diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/memory-legalizer-private-cluster.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/memory-legalizer-private-cluster.ll | 23347 |
1 files changed, 23347 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/memory-legalizer-private-cluster.ll b/llvm/test/CodeGen/AMDGPU/memory-legalizer-private-cluster.ll new file mode 100644 index 000000000000..2af195461d2e --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/memory-legalizer-private-cluster.ll @@ -0,0 +1,23347 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx600 < %s | FileCheck --check-prefixes=GFX6 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx700 < %s | FileCheck --check-prefixes=GFX7 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1010 < %s | FileCheck --check-prefixes=GFX10-WGP %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1010 -mattr=+cumode < %s | FileCheck --check-prefixes=GFX10-CU %s +; RUN: llc -mtriple=amdgcn-amd-amdpal -O0 -mcpu=gfx700 -amdgcn-skip-cache-invalidations < %s | FileCheck --check-prefixes=SKIP-CACHE-INV %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx90a < %s | FileCheck -check-prefixes=GFX90A-NOTTGSPLIT %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx90a -mattr=+tgsplit < %s | FileCheck -check-prefixes=GFX90A-TGSPLIT %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX942-NOTTGSPLIT %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx942 -mattr=+tgsplit < %s | FileCheck -check-prefixes=GFX942-TGSPLIT %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1100 < %s | FileCheck --check-prefixes=GFX11-WGP %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1100 -mattr=+cumode < %s | FileCheck --check-prefixes=GFX11-CU %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1200 < %s | FileCheck --check-prefixes=GFX12-WGP %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1200 -mattr=+cumode < %s | FileCheck --check-prefixes=GFX12-CU %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1250 < %s | FileCheck --check-prefixes=GFX1250 %s + +define amdgpu_kernel void @private_cluster_unordered_load( +; GFX6-LABEL: private_cluster_unordered_load: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_unordered_load: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_unordered_load: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_unordered_load: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_unordered_load: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_unordered_load: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_unordered_load: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_unordered_load: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_unordered_load: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_unordered_load: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_unordered_load: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_unordered_load: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_unordered_load: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_unordered_load: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: flat_load_b32 v0, v[0:1] +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %in, ptr addrspace(5) %out) { +entry: + %val = load atomic i32, ptr addrspace(5) %in syncscope("cluster") unordered, align 4 + store i32 %val, ptr addrspace(5) %out + ret void +} + +define amdgpu_kernel void @private_cluster_monotonic_load( +; GFX6-LABEL: private_cluster_monotonic_load: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_monotonic_load: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_monotonic_load: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_monotonic_load: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_monotonic_load: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_monotonic_load: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_monotonic_load: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_monotonic_load: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_monotonic_load: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_monotonic_load: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_monotonic_load: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_monotonic_load: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_monotonic_load: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_monotonic_load: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: flat_load_b32 v0, v[0:1] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %in, ptr addrspace(5) %out) { +entry: + %val = load atomic i32, ptr addrspace(5) %in syncscope("cluster") monotonic, align 4 + store i32 %val, ptr addrspace(5) %out + ret void +} + +define amdgpu_kernel void @private_cluster_acquire_load( +; GFX6-LABEL: private_cluster_acquire_load: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acquire_load: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acquire_load: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acquire_load: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acquire_load: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acquire_load: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acquire_load: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acquire_load: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acquire_load: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acquire_load: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acquire_load: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acquire_load: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acquire_load: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acquire_load: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: flat_load_b32 v0, v[0:1] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %in, ptr addrspace(5) %out) { +entry: + %val = load atomic i32, ptr addrspace(5) %in syncscope("cluster") acquire, align 4 + store i32 %val, ptr addrspace(5) %out + ret void +} + +define amdgpu_kernel void @private_cluster_seq_cst_load( +; GFX6-LABEL: private_cluster_seq_cst_load: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_seq_cst_load: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_seq_cst_load: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_seq_cst_load: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_seq_cst_load: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_seq_cst_load: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_seq_cst_load: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_seq_cst_load: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_seq_cst_load: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_seq_cst_load: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_seq_cst_load: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_seq_cst_load: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_seq_cst_load: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_seq_cst_load: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_load_b32 v0, v[0:1] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %in, ptr addrspace(5) %out) { +entry: + %val = load atomic i32, ptr addrspace(5) %in syncscope("cluster") seq_cst, align 4 + store i32 %val, ptr addrspace(5) %out + ret void +} + +define amdgpu_kernel void @private_cluster_unordered_store( +; GFX6-LABEL: private_cluster_unordered_store: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_unordered_store: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_unordered_store: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_unordered_store: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_unordered_store: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_unordered_store: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_unordered_store: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_unordered_store: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_unordered_store: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_unordered_store: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_unordered_store: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_unordered_store: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_unordered_store: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_unordered_store: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_store_b32 v[0:1], v2 +; GFX1250-NEXT: s_endpgm + i32 %in, ptr addrspace(5) %out) { +entry: + store atomic i32 %in, ptr addrspace(5) %out syncscope("cluster") unordered, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_monotonic_store( +; GFX6-LABEL: private_cluster_monotonic_store: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_monotonic_store: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_monotonic_store: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_monotonic_store: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_monotonic_store: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_monotonic_store: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_monotonic_store: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_monotonic_store: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_monotonic_store: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_monotonic_store: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_monotonic_store: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_monotonic_store: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_monotonic_store: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_monotonic_store: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_store_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + i32 %in, ptr addrspace(5) %out) { +entry: + store atomic i32 %in, ptr addrspace(5) %out syncscope("cluster") monotonic, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_release_store( +; GFX6-LABEL: private_cluster_release_store: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_release_store: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_release_store: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_release_store: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_release_store: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_release_store: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_release_store: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_release_store: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_release_store: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_release_store: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_release_store: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_release_store: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_release_store: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_release_store: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_store_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + i32 %in, ptr addrspace(5) %out) { +entry: + store atomic i32 %in, ptr addrspace(5) %out syncscope("cluster") release, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_seq_cst_store( +; GFX6-LABEL: private_cluster_seq_cst_store: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_seq_cst_store: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_seq_cst_store: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_seq_cst_store: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_seq_cst_store: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_seq_cst_store: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_seq_cst_store: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_seq_cst_store: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_seq_cst_store: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_seq_cst_store: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_seq_cst_store: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_seq_cst_store: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_seq_cst_store: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_seq_cst_store: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_store_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + i32 %in, ptr addrspace(5) %out) { +entry: + store atomic i32 %in, ptr addrspace(5) %out syncscope("cluster") seq_cst, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_monotonic_atomicrmw( +; GFX6-LABEL: private_cluster_monotonic_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_monotonic_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_monotonic_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_monotonic_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_monotonic_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_monotonic_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_monotonic_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_monotonic_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_monotonic_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_monotonic_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_monotonic_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_monotonic_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_monotonic_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_monotonic_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster") monotonic + ret void +} + +define amdgpu_kernel void @private_cluster_acquire_atomicrmw( +; GFX6-LABEL: private_cluster_acquire_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acquire_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acquire_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acquire_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acquire_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acquire_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acquire_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acquire_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acquire_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acquire_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acquire_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acquire_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acquire_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acquire_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster") acquire + ret void +} + +define amdgpu_kernel void @private_cluster_release_atomicrmw( +; GFX6-LABEL: private_cluster_release_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_release_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_release_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_release_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_release_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_release_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_release_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_release_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_release_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_release_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_release_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_release_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_release_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_release_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster") release + ret void +} + +define amdgpu_kernel void @private_cluster_acq_rel_atomicrmw( +; GFX6-LABEL: private_cluster_acq_rel_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acq_rel_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acq_rel_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acq_rel_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acq_rel_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acq_rel_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acq_rel_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acq_rel_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acq_rel_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acq_rel_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acq_rel_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acq_rel_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acq_rel_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acq_rel_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster") acq_rel + ret void +} + +define amdgpu_kernel void @private_cluster_seq_cst_atomicrmw( +; GFX6-LABEL: private_cluster_seq_cst_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_seq_cst_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_seq_cst_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_seq_cst_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_seq_cst_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_seq_cst_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_seq_cst_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_seq_cst_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_seq_cst_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_seq_cst_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_seq_cst_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_seq_cst_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_seq_cst_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_seq_cst_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster") seq_cst + ret void +} + +define amdgpu_kernel void @private_cluster_acquire_ret_atomicrmw( +; GFX6-LABEL: private_cluster_acquire_ret_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_mov_b32_e32 v2, s4 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(1) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acquire_ret_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_mov_b32_e32 v2, s4 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(1) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acquire_ret_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acquire_ret_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acquire_ret_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(1) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acquire_ret_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acquire_ret_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acquire_ret_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acquire_ret_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acquire_ret_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acquire_ret_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acquire_ret_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acquire_ret_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acquire_ret_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s0 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[2:3], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[2:3] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s0, s3 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v0, v[0:1], v2 th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster") acquire + store i32 %val, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_acq_rel_ret_atomicrmw( +; GFX6-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_mov_b32_e32 v2, s4 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(1) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_mov_b32_e32 v2, s4 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(1) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(1) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acq_rel_ret_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s0 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[2:3], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[2:3] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s0, s3 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v0, v[0:1], v2 th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster") acq_rel + store i32 %val, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_seq_cst_ret_atomicrmw( +; GFX6-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_mov_b32_e32 v2, s4 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(1) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_mov_b32_e32 v2, s4 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(1) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(1) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_seq_cst_ret_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s0 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[2:3], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[2:3] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s0, s3 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v0, v[0:1], v2 th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster") seq_cst + store i32 %val, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_monotonic_monotonic_cmpxchg( +; GFX6-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_monotonic_monotonic_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") monotonic monotonic + ret void +} + +define amdgpu_kernel void @private_cluster_acquire_monotonic_cmpxchg( +; GFX6-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acquire_monotonic_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") acquire monotonic + ret void +} + +define amdgpu_kernel void @private_cluster_release_monotonic_cmpxchg( +; GFX6-LABEL: private_cluster_release_monotonic_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_release_monotonic_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_release_monotonic_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_release_monotonic_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_release_monotonic_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_release_monotonic_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_release_monotonic_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_release_monotonic_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_release_monotonic_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_release_monotonic_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_release_monotonic_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_release_monotonic_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_release_monotonic_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_release_monotonic_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") release monotonic + ret void +} + +define amdgpu_kernel void @private_cluster_acq_rel_monotonic_cmpxchg( +; GFX6-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acq_rel_monotonic_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") acq_rel monotonic + ret void +} + +define amdgpu_kernel void @private_cluster_seq_cst_monotonic_cmpxchg( +; GFX6-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_seq_cst_monotonic_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") seq_cst monotonic + ret void +} + +define amdgpu_kernel void @private_cluster_monotonic_acquire_cmpxchg( +; GFX6-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_monotonic_acquire_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") monotonic acquire + ret void +} + +define amdgpu_kernel void @private_cluster_acquire_acquire_cmpxchg( +; GFX6-LABEL: private_cluster_acquire_acquire_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acquire_acquire_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acquire_acquire_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acquire_acquire_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acquire_acquire_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acquire_acquire_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acquire_acquire_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acquire_acquire_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acquire_acquire_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acquire_acquire_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acquire_acquire_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acquire_acquire_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acquire_acquire_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acquire_acquire_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") acquire acquire + ret void +} + +define amdgpu_kernel void @private_cluster_release_acquire_cmpxchg( +; GFX6-LABEL: private_cluster_release_acquire_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_release_acquire_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_release_acquire_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_release_acquire_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_release_acquire_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_release_acquire_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_release_acquire_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_release_acquire_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_release_acquire_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_release_acquire_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_release_acquire_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_release_acquire_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_release_acquire_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_release_acquire_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") release acquire + ret void +} + +define amdgpu_kernel void @private_cluster_acq_rel_acquire_cmpxchg( +; GFX6-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acq_rel_acquire_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") acq_rel acquire + ret void +} + +define amdgpu_kernel void @private_cluster_seq_cst_acquire_cmpxchg( +; GFX6-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_seq_cst_acquire_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") seq_cst acquire + ret void +} + +define amdgpu_kernel void @private_cluster_monotonic_seq_cst_cmpxchg( +; GFX6-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_monotonic_seq_cst_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") monotonic seq_cst + ret void +} + +define amdgpu_kernel void @private_cluster_acquire_seq_cst_cmpxchg( +; GFX6-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acquire_seq_cst_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") acquire seq_cst + ret void +} + +define amdgpu_kernel void @private_cluster_release_seq_cst_cmpxchg( +; GFX6-LABEL: private_cluster_release_seq_cst_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_release_seq_cst_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_release_seq_cst_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_release_seq_cst_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_release_seq_cst_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_release_seq_cst_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_release_seq_cst_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_release_seq_cst_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_release_seq_cst_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_release_seq_cst_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_release_seq_cst_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_release_seq_cst_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_release_seq_cst_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_release_seq_cst_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") release seq_cst + ret void +} + +define amdgpu_kernel void @private_cluster_acq_rel_seq_cst_cmpxchg( +; GFX6-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acq_rel_seq_cst_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") acq_rel seq_cst + ret void +} + +define amdgpu_kernel void @private_cluster_seq_cst_seq_cst_cmpxchg( +; GFX6-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_seq_cst_seq_cst_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") seq_cst seq_cst + ret void +} + +define amdgpu_kernel void @private_cluster_monotonic_monotonic_ret_cmpxchg( +; GFX6-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_monotonic_monotonic_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") monotonic monotonic + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_acquire_monotonic_ret_cmpxchg( +; GFX6-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acquire_monotonic_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") acquire monotonic + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_release_monotonic_ret_cmpxchg( +; GFX6-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_release_monotonic_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") release monotonic + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_acq_rel_monotonic_ret_cmpxchg( +; GFX6-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acq_rel_monotonic_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") acq_rel monotonic + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_seq_cst_monotonic_ret_cmpxchg( +; GFX6-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_seq_cst_monotonic_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") seq_cst monotonic + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_monotonic_acquire_ret_cmpxchg( +; GFX6-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_monotonic_acquire_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") monotonic acquire + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_acquire_acquire_ret_cmpxchg( +; GFX6-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acquire_acquire_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") acquire acquire + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_release_acquire_ret_cmpxchg( +; GFX6-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_release_acquire_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") release acquire + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_acq_rel_acquire_ret_cmpxchg( +; GFX6-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acq_rel_acquire_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") acq_rel acquire + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_seq_cst_acquire_ret_cmpxchg( +; GFX6-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_seq_cst_acquire_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") seq_cst acquire + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_monotonic_seq_cst_ret_cmpxchg( +; GFX6-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_monotonic_seq_cst_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") monotonic seq_cst + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_acquire_seq_cst_ret_cmpxchg( +; GFX6-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acquire_seq_cst_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") acquire seq_cst + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_release_seq_cst_ret_cmpxchg( +; GFX6-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_release_seq_cst_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") release seq_cst + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_acq_rel_seq_cst_ret_cmpxchg( +; GFX6-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_acq_rel_seq_cst_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") acq_rel seq_cst + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_seq_cst_seq_cst_ret_cmpxchg( +; GFX6-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_seq_cst_seq_cst_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster") seq_cst seq_cst + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_unordered_load( +; GFX6-LABEL: private_cluster_one_as_unordered_load: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_unordered_load: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_unordered_load: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_unordered_load: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_unordered_load: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_unordered_load: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_unordered_load: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_unordered_load: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_unordered_load: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_unordered_load: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_unordered_load: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_unordered_load: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_unordered_load: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_unordered_load: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: flat_load_b32 v0, v[0:1] +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %in, ptr addrspace(5) %out) { +entry: + %val = load atomic i32, ptr addrspace(5) %in syncscope("cluster-one-as") unordered, align 4 + store i32 %val, ptr addrspace(5) %out + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_monotonic_load( +; GFX6-LABEL: private_cluster_one_as_monotonic_load: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_monotonic_load: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_monotonic_load: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_monotonic_load: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_monotonic_load: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_load: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_monotonic_load: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_load: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_monotonic_load: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_monotonic_load: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_monotonic_load: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_monotonic_load: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_monotonic_load: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_monotonic_load: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: flat_load_b32 v0, v[0:1] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %in, ptr addrspace(5) %out) { +entry: + %val = load atomic i32, ptr addrspace(5) %in syncscope("cluster-one-as") monotonic, align 4 + store i32 %val, ptr addrspace(5) %out + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acquire_load( +; GFX6-LABEL: private_cluster_one_as_acquire_load: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acquire_load: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acquire_load: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acquire_load: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acquire_load: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_load: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acquire_load: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_load: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acquire_load: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acquire_load: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acquire_load: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acquire_load: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acquire_load: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acquire_load: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: flat_load_b32 v0, v[0:1] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %in, ptr addrspace(5) %out) { +entry: + %val = load atomic i32, ptr addrspace(5) %in syncscope("cluster-one-as") acquire, align 4 + store i32 %val, ptr addrspace(5) %out + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_seq_cst_load( +; GFX6-LABEL: private_cluster_one_as_seq_cst_load: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_seq_cst_load: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_seq_cst_load: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_seq_cst_load: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_seq_cst_load: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_load: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_load: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_load: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_load: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_seq_cst_load: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_seq_cst_load: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_seq_cst_load: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_seq_cst_load: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_seq_cst_load: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: flat_load_b32 v0, v[0:1] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %in, ptr addrspace(5) %out) { +entry: + %val = load atomic i32, ptr addrspace(5) %in syncscope("cluster-one-as") seq_cst, align 4 + store i32 %val, ptr addrspace(5) %out + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_unordered_store( +; GFX6-LABEL: private_cluster_one_as_unordered_store: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_unordered_store: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_unordered_store: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_unordered_store: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_unordered_store: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_unordered_store: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_unordered_store: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_unordered_store: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_unordered_store: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_unordered_store: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_unordered_store: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_unordered_store: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_unordered_store: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_unordered_store: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_store_b32 v[0:1], v2 +; GFX1250-NEXT: s_endpgm + i32 %in, ptr addrspace(5) %out) { +entry: + store atomic i32 %in, ptr addrspace(5) %out syncscope("cluster-one-as") unordered, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_monotonic_store( +; GFX6-LABEL: private_cluster_one_as_monotonic_store: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_monotonic_store: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_monotonic_store: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_monotonic_store: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_monotonic_store: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_store: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_monotonic_store: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_store: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_monotonic_store: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_monotonic_store: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_monotonic_store: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_monotonic_store: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_monotonic_store: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_monotonic_store: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_store_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + i32 %in, ptr addrspace(5) %out) { +entry: + store atomic i32 %in, ptr addrspace(5) %out syncscope("cluster-one-as") monotonic, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_release_store( +; GFX6-LABEL: private_cluster_one_as_release_store: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_release_store: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_release_store: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_release_store: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_release_store: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_release_store: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_release_store: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_release_store: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_release_store: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_release_store: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_release_store: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_release_store: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_release_store: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_release_store: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_store_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + i32 %in, ptr addrspace(5) %out) { +entry: + store atomic i32 %in, ptr addrspace(5) %out syncscope("cluster-one-as") release, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_seq_cst_store( +; GFX6-LABEL: private_cluster_one_as_seq_cst_store: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_seq_cst_store: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_seq_cst_store: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_seq_cst_store: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_seq_cst_store: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_store: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_store: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_store: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_store: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_seq_cst_store: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_seq_cst_store: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_seq_cst_store: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_seq_cst_store: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_seq_cst_store: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_store_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + i32 %in, ptr addrspace(5) %out) { +entry: + store atomic i32 %in, ptr addrspace(5) %out syncscope("cluster-one-as") seq_cst, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_monotonic_atomicrmw( +; GFX6-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_monotonic_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster-one-as") monotonic + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acquire_atomicrmw( +; GFX6-LABEL: private_cluster_one_as_acquire_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acquire_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acquire_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acquire_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acquire_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acquire_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acquire_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acquire_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acquire_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acquire_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acquire_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acquire_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster-one-as") acquire + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_release_atomicrmw( +; GFX6-LABEL: private_cluster_one_as_release_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_release_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_release_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_release_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_release_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_release_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_release_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_release_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_release_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_release_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_release_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_release_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_release_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_release_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster-one-as") release + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acq_rel_atomicrmw( +; GFX6-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acq_rel_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster-one-as") acq_rel + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_seq_cst_atomicrmw( +; GFX6-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v0, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: v_mov_b32_e32 v0, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_seq_cst_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s1, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s1 +; GFX1250-NEXT: s_mov_b32 s1, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s1, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s1, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s2, s3 +; GFX1250-NEXT: s_cselect_b32 s2, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s1, v2, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s1, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s1, v0, s2 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s0 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v[0:1], v2 scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster-one-as") seq_cst + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acquire_ret_atomicrmw( +; GFX6-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_mov_b32_e32 v2, s4 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(1) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_mov_b32_e32 v2, s4 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(1) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(1) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acquire_ret_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s0 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[2:3], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[2:3] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s0, s3 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v0, v[0:1], v2 th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster-one-as") acquire + store i32 %val, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acq_rel_ret_atomicrmw( +; GFX6-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_mov_b32_e32 v2, s4 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(1) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_mov_b32_e32 v2, s4 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(1) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(1) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acq_rel_ret_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s0 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[2:3], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[2:3] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s0, s3 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v0, v[0:1], v2 th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster-one-as") acq_rel + store i32 %val, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_seq_cst_ret_atomicrmw( +; GFX6-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_mov_b32_e32 v2, s4 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: s_waitcnt vmcnt(1) +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_mov_b32_e32 v2, s4 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: s_waitcnt vmcnt(1) +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s5 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s5 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(1) +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(1) +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: v_mov_b32_e32 v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: v_mov_b32_e32 v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX12-WGP-NEXT: v_mov_b32_e32 v1, s1 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX12-CU-NEXT: v_mov_b32_e32 v1, s1 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_seq_cst_ret_atomicrmw: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s0 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[2:3], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[2:3] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s3, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s0, s3 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_swap_b32 v0, v[0:1], v2 th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in) { +entry: + %val = atomicrmw volatile xchg ptr addrspace(5) %out, i32 %in syncscope("cluster-one-as") seq_cst + store i32 %val, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_monotonic_monotonic_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_monotonic_monotonic_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") monotonic monotonic + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acquire_monotonic_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acquire_monotonic_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") acquire monotonic + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_release_monotonic_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_release_monotonic_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") release monotonic + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acq_rel_monotonic_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acq_rel_monotonic_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") acq_rel monotonic + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_seq_cst_monotonic_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_seq_cst_monotonic_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") seq_cst monotonic + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_monotonic_acquire_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_monotonic_acquire_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") monotonic acquire + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acquire_acquire_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acquire_acquire_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") acquire acquire + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_release_acquire_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_release_acquire_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") release acquire + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acq_rel_acquire_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acq_rel_acquire_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") acq_rel acquire + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_seq_cst_acquire_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_seq_cst_acquire_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") seq_cst acquire + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_monotonic_seq_cst_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_monotonic_seq_cst_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") monotonic seq_cst + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acquire_seq_cst_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acquire_seq_cst_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") acquire seq_cst + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_release_seq_cst_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_release_seq_cst_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") release seq_cst + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acq_rel_seq_cst_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acq_rel_seq_cst_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") acq_rel seq_cst + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_seq_cst_seq_cst_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX6-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX6-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX6-NEXT: s_mov_b32 s7, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s4, s4, s7 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX7-NEXT: s_load_dword s5, s[8:9], 0x1 +; GFX7-NEXT: s_load_dword s6, s[8:9], 0x2 +; GFX7-NEXT: s_mov_b32 s7, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s4, s4, s7 +; GFX7-NEXT: v_mov_b32_e32 v0, s4 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v0, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[4:5], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s1, s[4:5], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[4:5], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s3, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s0, s0, s3 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s0 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s1 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[4:5], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s1, s[4:5], 0x4 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[4:5], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s3, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s0, s0, s3 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s0 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s1 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v0, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s3, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s0, s0, s3 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s3, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s0, s0, s3 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s0 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v0, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 offset:16 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_seq_cst_seq_cst_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s3, s2, s3 +; GFX1250-NEXT: s_mov_b32 s2, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s2 +; GFX1250-NEXT: s_mov_b32 s2, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s2, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[4:5], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[4:5] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[4:5], 0 +; GFX1250-NEXT: s_mov_b32 s2, s5 +; GFX1250-NEXT: s_mov_b32 s6, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s3, s6 +; GFX1250-NEXT: s_cselect_b32 s3, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s2, v2, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s2, s4 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s2, v0, s3 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s1 +; GFX1250-NEXT: v_mov_b32_e32 v4, s0 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v[0:1], v[2:3] scope:SCOPE_SE +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") seq_cst seq_cst + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_monotonic_monotonic_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_monotonic_monotonic_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") monotonic monotonic + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acquire_monotonic_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acquire_monotonic_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") acquire monotonic + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acq_rel_monotonic_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") acq_rel monotonic + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_seq_cst_monotonic_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") seq_cst monotonic + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_monotonic_acquire_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_monotonic_acquire_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") monotonic acquire + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acquire_acquire_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acquire_acquire_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") acquire acquire + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_release_acquire_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_release_acquire_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") release acquire + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acq_rel_acquire_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acq_rel_acquire_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") acq_rel acquire + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_seq_cst_acquire_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_seq_cst_acquire_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") seq_cst acquire + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_monotonic_seq_cst_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") monotonic seq_cst + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acquire_seq_cst_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acquire_seq_cst_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") acquire seq_cst + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_release_seq_cst_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_release_seq_cst_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") release seq_cst + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_acq_rel_seq_cst_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") acq_rel seq_cst + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} + +define amdgpu_kernel void @private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg( +; GFX6-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_add_u32 s0, s0, s15 +; GFX6-NEXT: s_addc_u32 s1, s1, 0 +; GFX6-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX6-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX6-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX6-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX6-NEXT: s_mov_b32 s5, 16 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_add_i32 s5, s4, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s5 +; GFX6-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX6-NEXT: v_mov_b32_e32 v1, s8 +; GFX6-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX6-NEXT: v_mov_b32_e32 v2, s5 +; GFX6-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX6-NEXT: s_endpgm +; +; GFX7-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; GFX7: ; %bb.0: ; %entry +; GFX7-NEXT: s_add_u32 s0, s0, s17 +; GFX7-NEXT: s_addc_u32 s1, s1, 0 +; GFX7-NEXT: s_mov_b64 s[6:7], s[8:9] +; GFX7-NEXT: s_load_dword s4, s[6:7], 0x0 +; GFX7-NEXT: s_load_dword s8, s[6:7], 0x1 +; GFX7-NEXT: s_load_dword s6, s[6:7], 0x2 +; GFX7-NEXT: s_mov_b32 s5, 16 +; GFX7-NEXT: s_waitcnt lgkmcnt(0) +; GFX7-NEXT: s_add_i32 s5, s4, s5 +; GFX7-NEXT: v_mov_b32_e32 v0, s5 +; GFX7-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen +; GFX7-NEXT: s_waitcnt vmcnt(0) +; GFX7-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX7-NEXT: v_mov_b32_e32 v1, s8 +; GFX7-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX7-NEXT: v_mov_b32_e32 v2, s5 +; GFX7-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen +; GFX7-NEXT: v_mov_b32_e32 v1, s4 +; GFX7-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX7-NEXT: s_endpgm +; +; GFX10-WGP-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; GFX10-WGP: ; %bb.0: ; %entry +; GFX10-WGP-NEXT: s_add_u32 s0, s0, s17 +; GFX10-WGP-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-WGP-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-WGP-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-WGP-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-WGP-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-WGP-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX10-WGP-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-WGP-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-WGP-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-WGP-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-WGP-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-WGP-NEXT: s_endpgm +; +; GFX10-CU-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; GFX10-CU: ; %bb.0: ; %entry +; GFX10-CU-NEXT: s_add_u32 s0, s0, s17 +; GFX10-CU-NEXT: s_addc_u32 s1, s1, 0 +; GFX10-CU-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX10-CU-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX10-CU-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX10-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-CU-NEXT: v_mov_b32_e32 v0, s4 +; GFX10-CU-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: s_waitcnt vmcnt(0) +; GFX10-CU-NEXT: v_cmp_eq_u32_e64 s6, v0, s6 +; GFX10-CU-NEXT: v_cndmask_b32_e64 v1, v0, s5, s6 +; GFX10-CU-NEXT: v_mov_b32_e32 v2, s4 +; GFX10-CU-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX10-CU-NEXT: v_mov_b32_e32 v1, s4 +; GFX10-CU-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX10-CU-NEXT: s_endpgm +; +; SKIP-CACHE-INV-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; SKIP-CACHE-INV: ; %bb.0: ; %entry +; SKIP-CACHE-INV-NEXT: s_getpc_b64 s[12:13] +; SKIP-CACHE-INV-NEXT: s_mov_b32 s12, s0 +; SKIP-CACHE-INV-NEXT: s_load_dwordx4 s[12:15], s[12:13], 0x0 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_u32 s12, s12, s11 +; SKIP-CACHE-INV-NEXT: s_addc_u32 s13, s13, 0 +; SKIP-CACHE-INV-NEXT: s_mov_b64 s[2:3], s[4:5] +; SKIP-CACHE-INV-NEXT: s_load_dword s0, s[2:3], 0x0 +; SKIP-CACHE-INV-NEXT: s_load_dword s4, s[2:3], 0x1 +; SKIP-CACHE-INV-NEXT: s_load_dword s2, s[2:3], 0x2 +; SKIP-CACHE-INV-NEXT: s_mov_b32 s1, 16 +; SKIP-CACHE-INV-NEXT: s_waitcnt lgkmcnt(0) +; SKIP-CACHE-INV-NEXT: s_add_i32 s1, s0, s1 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v0, s1 +; SKIP-CACHE-INV-NEXT: buffer_load_dword v0, v0, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_waitcnt vmcnt(0) +; SKIP-CACHE-INV-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s4 +; SKIP-CACHE-INV-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v2, s1 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: v_mov_b32_e32 v1, s0 +; SKIP-CACHE-INV-NEXT: buffer_store_dword v0, v1, s[12:15], 0 offen +; SKIP-CACHE-INV-NEXT: s_endpgm +; +; GFX90A-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; GFX90A-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX90A-NOTTGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-NOTTGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-NOTTGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-NOTTGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX90A-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; GFX90A-TGSPLIT: ; %bb.0: ; %entry +; GFX90A-TGSPLIT-NEXT: s_add_u32 s0, s0, s17 +; GFX90A-TGSPLIT-NEXT: s_addc_u32 s1, s1, 0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s4, s[8:9], 0x0 +; GFX90A-TGSPLIT-NEXT: s_load_dword s5, s[8:9], 0x4 +; GFX90A-TGSPLIT-NEXT: s_load_dword s6, s[8:9], 0x8 +; GFX90A-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-TGSPLIT-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX90A-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s6 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s5 +; GFX90A-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[6:7] +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v2, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v1, v2, s[0:3], 0 offen offset:16 +; GFX90A-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX90A-TGSPLIT-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen +; GFX90A-TGSPLIT-NEXT: s_endpgm +; +; GFX942-NOTTGSPLIT-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; GFX942-NOTTGSPLIT: ; %bb.0: ; %entry +; GFX942-NOTTGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-NOTTGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-NOTTGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-NOTTGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-NOTTGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-NOTTGSPLIT-NEXT: s_nop 0 +; GFX942-NOTTGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-NOTTGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-NOTTGSPLIT-NEXT: s_endpgm +; +; GFX942-TGSPLIT-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; GFX942-TGSPLIT: ; %bb.0: ; %entry +; GFX942-TGSPLIT-NEXT: s_mov_b64 s[2:3], s[4:5] +; GFX942-TGSPLIT-NEXT: s_load_dword s0, s[2:3], 0x0 +; GFX942-TGSPLIT-NEXT: s_load_dword s4, s[2:3], 0x4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: s_load_dword s2, s[2:3], 0x8 +; GFX942-TGSPLIT-NEXT: s_mov_b32 s1, 16 +; GFX942-TGSPLIT-NEXT: s_waitcnt lgkmcnt(0) +; GFX942-TGSPLIT-NEXT: s_add_i32 s1, s0, s1 +; GFX942-TGSPLIT-NEXT: scratch_load_dword v0, off, s1 +; GFX942-TGSPLIT-NEXT: s_waitcnt vmcnt(0) +; GFX942-TGSPLIT-NEXT: v_cmp_eq_u32_e64 s[2:3], v0, s2 +; GFX942-TGSPLIT-NEXT: v_mov_b32_e32 v1, s4 +; GFX942-TGSPLIT-NEXT: s_nop 0 +; GFX942-TGSPLIT-NEXT: v_cndmask_b32_e64 v1, v0, v1, s[2:3] +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v1, s1 +; GFX942-TGSPLIT-NEXT: scratch_store_dword off, v0, s0 +; GFX942-TGSPLIT-NEXT: s_endpgm +; +; GFX11-WGP-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; GFX11-WGP: ; %bb.0: ; %entry +; GFX11-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-WGP-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-WGP-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-WGP-NEXT: s_mov_b32 s1, 16 +; GFX11-WGP-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-WGP-NEXT: s_add_i32 s1, s0, s1 +; GFX11-WGP-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-WGP-NEXT: s_waitcnt vmcnt(0) +; GFX11-WGP-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-WGP-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-WGP-NEXT: s_endpgm +; +; GFX11-CU-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; GFX11-CU: ; %bb.0: ; %entry +; GFX11-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX11-CU-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX11-CU-NEXT: s_load_b32 s3, s[4:5], 0x8 +; GFX11-CU-NEXT: s_mov_b32 s1, 16 +; GFX11-CU-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-CU-NEXT: s_add_i32 s1, s0, s1 +; GFX11-CU-NEXT: scratch_load_b32 v0, off, s1 +; GFX11-CU-NEXT: s_waitcnt vmcnt(0) +; GFX11-CU-NEXT: v_cmp_eq_u32_e64 s3, v0, s3 +; GFX11-CU-NEXT: v_cndmask_b32_e64 v1, v0, s2, s3 +; GFX11-CU-NEXT: scratch_store_b32 off, v1, s1 +; GFX11-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX11-CU-NEXT: s_endpgm +; +; GFX12-WGP-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; GFX12-WGP: ; %bb.0: ; %entry +; GFX12-WGP-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-WGP-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-WGP-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-WGP-NEXT: s_wait_kmcnt 0x0 +; GFX12-WGP-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-WGP-NEXT: s_wait_loadcnt 0x0 +; GFX12-WGP-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-WGP-NEXT: s_wait_alu 0xf1ff +; GFX12-WGP-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-WGP-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-WGP-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-WGP-NEXT: s_endpgm +; +; GFX12-CU-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; GFX12-CU: ; %bb.0: ; %entry +; GFX12-CU-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX12-CU-NEXT: s_load_b32 s1, s[4:5], 0x4 +; GFX12-CU-NEXT: s_load_b32 s2, s[4:5], 0x8 +; GFX12-CU-NEXT: s_wait_kmcnt 0x0 +; GFX12-CU-NEXT: scratch_load_b32 v0, off, s0 offset:16 +; GFX12-CU-NEXT: s_wait_loadcnt 0x0 +; GFX12-CU-NEXT: v_cmp_eq_u32_e64 s2, v0, s2 +; GFX12-CU-NEXT: s_wait_alu 0xf1ff +; GFX12-CU-NEXT: v_cndmask_b32_e64 v1, v0, s1, s2 +; GFX12-CU-NEXT: scratch_store_b32 off, v1, s0 offset:16 +; GFX12-CU-NEXT: scratch_store_b32 off, v0, s0 +; GFX12-CU-NEXT: s_endpgm +; +; GFX1250-LABEL: private_cluster_one_as_seq_cst_seq_cst_ret_cmpxchg: +; GFX1250: ; %bb.0: ; %entry +; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0 +; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x4 +; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x8 +; GFX1250-NEXT: s_mov_b32 s3, 16 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: s_add_co_i32 s4, s0, s3 +; GFX1250-NEXT: s_mov_b32 s3, 0 +; GFX1250-NEXT: v_mbcnt_lo_u32_b32 v0, -1, s3 +; GFX1250-NEXT: s_mov_b32 s3, 20 +; GFX1250-NEXT: v_lshlrev_b32_e64 v2, s3, v0 +; GFX1250-NEXT: v_mov_b32_e32 v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: s_mov_b64 s[6:7], src_flat_scratch_base_lo +; GFX1250-NEXT: v_add_nc_u64_e64 v[0:1], v[0:1], s[6:7] +; GFX1250-NEXT: v_mov_b32_e32 v2, v1 +; GFX1250-NEXT: s_mov_b64 s[6:7], 0 +; GFX1250-NEXT: s_mov_b32 s3, s7 +; GFX1250-NEXT: s_mov_b32 s5, -1 +; GFX1250-NEXT: s_cmp_lg_u32 s4, s5 +; GFX1250-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1250-NEXT: v_cndmask_b32_e64 v2, s3, v2, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 killed $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: s_mov_b32 s3, s6 +; GFX1250-NEXT: v_cndmask_b32_e64 v0, s3, v0, s4 +; GFX1250-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v1, v2 +; GFX1250-NEXT: v_mov_b32_e32 v2, s2 +; GFX1250-NEXT: v_mov_b32_e32 v4, s1 +; GFX1250-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec +; GFX1250-NEXT: v_mov_b32_e32 v3, v4 +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: s_wait_storecnt 0x0 +; GFX1250-NEXT: s_wait_xcnt 0x0 +; GFX1250-NEXT: flat_atomic_cmpswap_b32 v0, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_SE +; GFX1250-NEXT: s_wait_loadcnt 0x0 +; GFX1250-NEXT: global_inv scope:SCOPE_SE +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: scratch_store_b32 off, v0, s0 +; GFX1250-NEXT: s_endpgm + ptr addrspace(5) %out, i32 %in, i32 %old) { +entry: + %gep = getelementptr i32, ptr addrspace(5) %out, i32 4 + %val = cmpxchg volatile ptr addrspace(5) %gep, i32 %old, i32 %in syncscope("cluster-one-as") seq_cst seq_cst + %val0 = extractvalue { i32, i1 } %val, 0 + store i32 %val0, ptr addrspace(5) %out, align 4 + ret void +} |