diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll | 94 |
1 files changed, 44 insertions, 50 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll index 97df2a0dbd44..258bc2959f39 100644 --- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll +++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll @@ -5548,7 +5548,6 @@ define amdgpu_kernel void @sub_i64_constant(ptr addrspace(1) %out, ptr addrspace ; GFX7LESS: ; %bb.0: ; %entry ; GFX7LESS-NEXT: s_mov_b64 s[6:7], exec ; GFX7LESS-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX7LESS-NEXT: s_mov_b32 s4, 0 ; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, s6, 0 ; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v4, s7, v0 ; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4 @@ -5557,33 +5556,32 @@ define amdgpu_kernel void @sub_i64_constant(ptr addrspace(1) %out, ptr addrspace ; GFX7LESS-NEXT: s_cbranch_execz .LBB9_4 ; GFX7LESS-NEXT: ; %bb.1: ; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0) -; GFX7LESS-NEXT: s_load_dwordx2 s[14:15], s[2:3], 0x0 -; GFX7LESS-NEXT: s_bcnt1_i32_b64 s5, s[6:7] +; GFX7LESS-NEXT: s_load_dwordx2 s[4:5], s[2:3], 0x0 +; GFX7LESS-NEXT: s_bcnt1_i32_b64 s6, s[6:7] ; GFX7LESS-NEXT: s_mov_b64 s[10:11], 0 -; GFX7LESS-NEXT: v_mov_b32_e32 v5, s4 ; GFX7LESS-NEXT: s_mov_b32 s7, 0xf000 -; GFX7LESS-NEXT: s_mul_i32 s12, s5, 5 +; GFX7LESS-NEXT: s_mul_i32 s12, s6, 5 ; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0) -; GFX7LESS-NEXT: v_mov_b32_e32 v0, s14 -; GFX7LESS-NEXT: v_mov_b32_e32 v1, s15 +; GFX7LESS-NEXT: v_mov_b32_e32 v0, s4 +; GFX7LESS-NEXT: v_mov_b32_e32 v1, s5 ; GFX7LESS-NEXT: s_mov_b32 s6, -1 ; GFX7LESS-NEXT: s_mov_b32 s4, s2 ; GFX7LESS-NEXT: s_mov_b32 s5, s3 ; GFX7LESS-NEXT: .LBB9_2: ; %atomicrmw.start ; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7LESS-NEXT: v_mov_b32_e32 v9, v1 -; GFX7LESS-NEXT: v_mov_b32_e32 v8, v0 -; GFX7LESS-NEXT: v_subrev_i32_e32 v6, vcc, s12, v8 -; GFX7LESS-NEXT: v_subb_u32_e32 v7, vcc, v9, v5, vcc +; GFX7LESS-NEXT: v_mov_b32_e32 v8, v1 +; GFX7LESS-NEXT: v_mov_b32_e32 v7, v0 +; GFX7LESS-NEXT: v_subrev_i32_e32 v5, vcc, s12, v7 +; GFX7LESS-NEXT: v_subbrev_u32_e32 v6, vcc, 0, v8, vcc ; GFX7LESS-NEXT: s_waitcnt expcnt(0) -; GFX7LESS-NEXT: v_mov_b32_e32 v0, v6 -; GFX7LESS-NEXT: v_mov_b32_e32 v1, v7 -; GFX7LESS-NEXT: v_mov_b32_e32 v2, v8 -; GFX7LESS-NEXT: v_mov_b32_e32 v3, v9 +; GFX7LESS-NEXT: v_mov_b32_e32 v0, v5 +; GFX7LESS-NEXT: v_mov_b32_e32 v1, v6 +; GFX7LESS-NEXT: v_mov_b32_e32 v2, v7 +; GFX7LESS-NEXT: v_mov_b32_e32 v3, v8 ; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[0:3], off, s[4:7], 0 glc ; GFX7LESS-NEXT: s_waitcnt vmcnt(0) ; GFX7LESS-NEXT: buffer_wbinvl1 -; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[8:9] +; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[7:8] ; GFX7LESS-NEXT: s_or_b64 s[10:11], vcc, s[10:11] ; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[10:11] ; GFX7LESS-NEXT: s_cbranch_execnz .LBB9_2 @@ -5611,39 +5609,37 @@ define amdgpu_kernel void @sub_i64_constant(ptr addrspace(1) %out, ptr addrspace ; GFX8-NEXT: s_mov_b64 s[6:7], exec ; GFX8-NEXT: v_mbcnt_lo_u32_b32 v0, s6, 0 ; GFX8-NEXT: v_mbcnt_hi_u32_b32 v4, s7, v0 -; GFX8-NEXT: s_mov_b32 s4, 0 ; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4 ; GFX8-NEXT: ; implicit-def: $vgpr0_vgpr1 ; GFX8-NEXT: s_and_saveexec_b64 s[8:9], vcc ; GFX8-NEXT: s_cbranch_execz .LBB9_4 ; GFX8-NEXT: ; %bb.1: ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: s_load_dwordx2 s[14:15], s[2:3], 0x0 -; GFX8-NEXT: s_bcnt1_i32_b64 s5, s[6:7] +; GFX8-NEXT: s_load_dwordx2 s[4:5], s[2:3], 0x0 +; GFX8-NEXT: s_bcnt1_i32_b64 s6, s[6:7] ; GFX8-NEXT: s_mov_b64 s[10:11], 0 -; GFX8-NEXT: v_mov_b32_e32 v5, s4 -; GFX8-NEXT: s_mul_i32 s12, s5, 5 -; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_mov_b32_e32 v0, s14 -; GFX8-NEXT: v_mov_b32_e32 v1, s15 ; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mul_i32 s12, s6, 5 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: v_mov_b32_e32 v0, s4 +; GFX8-NEXT: v_mov_b32_e32 v1, s5 ; GFX8-NEXT: s_mov_b32 s6, -1 ; GFX8-NEXT: s_mov_b32 s4, s2 ; GFX8-NEXT: s_mov_b32 s5, s3 ; GFX8-NEXT: .LBB9_2: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX8-NEXT: v_mov_b32_e32 v9, v1 -; GFX8-NEXT: v_mov_b32_e32 v8, v0 -; GFX8-NEXT: v_subrev_u32_e32 v6, vcc, s12, v8 -; GFX8-NEXT: v_subb_u32_e32 v7, vcc, v9, v5, vcc -; GFX8-NEXT: v_mov_b32_e32 v0, v6 -; GFX8-NEXT: v_mov_b32_e32 v1, v7 -; GFX8-NEXT: v_mov_b32_e32 v2, v8 -; GFX8-NEXT: v_mov_b32_e32 v3, v9 +; GFX8-NEXT: v_mov_b32_e32 v8, v1 +; GFX8-NEXT: v_mov_b32_e32 v7, v0 +; GFX8-NEXT: v_subrev_u32_e32 v5, vcc, s12, v7 +; GFX8-NEXT: v_subbrev_u32_e32 v6, vcc, 0, v8, vcc +; GFX8-NEXT: v_mov_b32_e32 v0, v5 +; GFX8-NEXT: v_mov_b32_e32 v1, v6 +; GFX8-NEXT: v_mov_b32_e32 v2, v7 +; GFX8-NEXT: v_mov_b32_e32 v3, v8 ; GFX8-NEXT: buffer_atomic_cmpswap_x2 v[0:3], off, s[4:7], 0 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1_vol -; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[8:9] +; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[7:8] ; GFX8-NEXT: s_or_b64 s[10:11], vcc, s[10:11] ; GFX8-NEXT: s_andn2_b64 exec, exec, s[10:11] ; GFX8-NEXT: s_cbranch_execnz .LBB9_2 @@ -5670,39 +5666,37 @@ define amdgpu_kernel void @sub_i64_constant(ptr addrspace(1) %out, ptr addrspace ; GFX9-NEXT: s_mov_b64 s[6:7], exec ; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, s6, 0 ; GFX9-NEXT: v_mbcnt_hi_u32_b32 v4, s7, v0 -; GFX9-NEXT: s_mov_b32 s4, 0 ; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4 ; GFX9-NEXT: ; implicit-def: $vgpr0_vgpr1 ; GFX9-NEXT: s_and_saveexec_b64 s[8:9], vcc ; GFX9-NEXT: s_cbranch_execz .LBB9_4 ; GFX9-NEXT: ; %bb.1: ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_load_dwordx2 s[14:15], s[2:3], 0x0 -; GFX9-NEXT: s_bcnt1_i32_b64 s5, s[6:7] +; GFX9-NEXT: s_load_dwordx2 s[4:5], s[2:3], 0x0 +; GFX9-NEXT: s_bcnt1_i32_b64 s6, s[6:7] ; GFX9-NEXT: s_mov_b64 s[10:11], 0 -; GFX9-NEXT: v_mov_b32_e32 v5, s4 -; GFX9-NEXT: s_mul_i32 s12, s5, 5 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_mov_b32_e32 v0, s14 -; GFX9-NEXT: v_mov_b32_e32 v1, s15 ; GFX9-NEXT: s_mov_b32 s7, 0xf000 +; GFX9-NEXT: s_mul_i32 s12, s6, 5 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_mov_b32_e32 v0, s4 +; GFX9-NEXT: v_mov_b32_e32 v1, s5 ; GFX9-NEXT: s_mov_b32 s6, -1 ; GFX9-NEXT: s_mov_b32 s4, s2 ; GFX9-NEXT: s_mov_b32 s5, s3 ; GFX9-NEXT: .LBB9_2: ; %atomicrmw.start ; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX9-NEXT: v_mov_b32_e32 v9, v1 -; GFX9-NEXT: v_mov_b32_e32 v8, v0 -; GFX9-NEXT: v_subrev_co_u32_e32 v6, vcc, s12, v8 -; GFX9-NEXT: v_subb_co_u32_e32 v7, vcc, v9, v5, vcc -; GFX9-NEXT: v_mov_b32_e32 v0, v6 -; GFX9-NEXT: v_mov_b32_e32 v1, v7 -; GFX9-NEXT: v_mov_b32_e32 v2, v8 -; GFX9-NEXT: v_mov_b32_e32 v3, v9 +; GFX9-NEXT: v_mov_b32_e32 v8, v1 +; GFX9-NEXT: v_mov_b32_e32 v7, v0 +; GFX9-NEXT: v_subrev_co_u32_e32 v5, vcc, s12, v7 +; GFX9-NEXT: v_subbrev_co_u32_e32 v6, vcc, 0, v8, vcc +; GFX9-NEXT: v_mov_b32_e32 v0, v5 +; GFX9-NEXT: v_mov_b32_e32 v1, v6 +; GFX9-NEXT: v_mov_b32_e32 v2, v7 +; GFX9-NEXT: v_mov_b32_e32 v3, v8 ; GFX9-NEXT: buffer_atomic_cmpswap_x2 v[0:3], off, s[4:7], 0 glc ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: buffer_wbinvl1_vol -; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[8:9] +; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[7:8] ; GFX9-NEXT: s_or_b64 s[10:11], vcc, s[10:11] ; GFX9-NEXT: s_andn2_b64 exec, exec, s[10:11] ; GFX9-NEXT: s_cbranch_execnz .LBB9_2 |