Merge branch 'main' into users/mingmingl-llvm/samplefdo-profile-formatusers/mingmingl-llvm/samplefdo-profile-format

1 files changed, 479 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cooperative.atomic-singlethread.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cooperative.atomic-singlethread.ll
new file mode 100644
index 000000000000..f97bced6364d
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cooperative.atomic-singlethread.ll
@@ -0,0 +1,479 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-GISEL %s
+
+define i32 @test_flat_amdgcn_cooperative_atomic_load_32x4B_relaxed(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_load_32x4B_relaxed:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b32 v0, v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call i32 @llvm.amdgcn.cooperative.atomic.load.32x4B.p0(ptr %addr, i32 0, metadata !0)
+  ret i32 %0
+}
+
+define <2 x i32> @test_flat_amdgcn_cooperative_atomic_load_16x8B_relaxed(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_load_16x8B_relaxed:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b64 v[0:1], v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call <2 x i32> @llvm.amdgcn.cooperative.atomic.load.16x8B.p0(ptr %addr, i32 0, metadata !0)
+  ret <2 x i32> %0
+}
+
+define <4 x i32> @test_flat_amdgcn_cooperative_atomic_load_8x16B_relaxed(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_load_8x16B_relaxed:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b128 v[0:3], v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call <4 x i32> @llvm.amdgcn.cooperative.atomic.load.8x16B.p0(ptr %addr, i32 0, metadata !0)
+  ret <4 x i32> %0
+}
+
+define void @test_flat_amdgcn_cooperative_atomic_store_32x4B_relaxed(ptr noundef %addr, i32 noundef %val)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_store_32x4B_relaxed:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b32 v[0:1], v2 scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.32x4B.p0(ptr %addr, i32 %val, i32 0, metadata !0)
+  ret void
+}
+
+define void @test_flat_amdgcn_cooperative_atomic_store_16x8B_relaxed(ptr noundef %addr, <2 x i32> noundef %val)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_store_16x8B_relaxed:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b64 v[0:1], v[2:3] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.16x8B.p0(ptr %addr, <2 x i32> %val, i32 0, metadata !0)
+  ret void
+}
+
+define void @test_flat_amdgcn_cooperative_atomic_store_8x16B_relaxed(ptr noundef %addr, <4 x i32> noundef %val)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_store_8x16B_relaxed:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b128 v[0:1], v[2:5] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.8x16B.p0(ptr %addr, <4 x i32> %val, i32 0, metadata !0)
+  ret void
+}
+
+define i32 @test_flat_amdgcn_cooperative_atomic_load_32x4B_acquire(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_load_32x4B_acquire:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b32 v0, v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call i32 @llvm.amdgcn.cooperative.atomic.load.32x4B.p0(ptr %addr, i32 2, metadata !0)
+  ret i32 %0
+}
+
+define <2 x i32> @test_flat_amdgcn_cooperative_atomic_load_16x8B_acquire(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_load_16x8B_acquire:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b64 v[0:1], v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call <2 x i32> @llvm.amdgcn.cooperative.atomic.load.16x8B.p0(ptr %addr, i32 2, metadata !0)
+  ret <2 x i32> %0
+}
+
+define <4 x i32> @test_flat_amdgcn_cooperative_atomic_load_8x16B_acquire(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_load_8x16B_acquire:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b128 v[0:3], v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call <4 x i32> @llvm.amdgcn.cooperative.atomic.load.8x16B.p0(ptr %addr, i32 2, metadata !0)
+  ret <4 x i32> %0
+}
+
+define void @test_flat_amdgcn_cooperative_atomic_store_32x4B_release(ptr noundef %addr, i32 noundef %val)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_store_32x4B_release:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b32 v[0:1], v2 scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.32x4B.p0(ptr %addr, i32 %val, i32 3, metadata !0)
+  ret void
+}
+
+define void @test_flat_amdgcn_cooperative_atomic_store_16x8B_release(ptr noundef %addr, <2 x i32> noundef %val)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_store_16x8B_release:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b64 v[0:1], v[2:3] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.16x8B.p0(ptr %addr, <2 x i32> %val, i32 3, metadata !0)
+  ret void
+}
+
+define void @test_flat_amdgcn_cooperative_atomic_store_8x16B_release(ptr noundef %addr, <4 x i32> noundef %val)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_store_8x16B_release:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b128 v[0:1], v[2:5] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.8x16B.p0(ptr %addr, <4 x i32> %val, i32 3, metadata !0)
+  ret void
+}
+
+define i32 @test_flat_amdgcn_cooperative_atomic_load_32x4B_seq_cst(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_load_32x4B_seq_cst:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b32 v0, v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call i32 @llvm.amdgcn.cooperative.atomic.load.32x4B.p0(ptr %addr, i32 5, metadata !0)
+  ret i32 %0
+}
+
+define <2 x i32> @test_flat_amdgcn_cooperative_atomic_load_16x8B_seq_cst(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_load_16x8B_seq_cst:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b64 v[0:1], v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call <2 x i32> @llvm.amdgcn.cooperative.atomic.load.16x8B.p0(ptr %addr, i32 5, metadata !0)
+  ret <2 x i32> %0
+}
+
+define <4 x i32> @test_flat_amdgcn_cooperative_atomic_load_8x16B_seq_cst(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_load_8x16B_seq_cst:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b128 v[0:3], v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call <4 x i32> @llvm.amdgcn.cooperative.atomic.load.8x16B.p0(ptr %addr, i32 5, metadata !0)
+  ret <4 x i32> %0
+}
+
+define void @test_flat_amdgcn_cooperative_atomic_store_32x4B_seq_cst(ptr noundef %addr, i32 noundef %val)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_store_32x4B_seq_cst:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b32 v[0:1], v2 scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.32x4B.p0(ptr %addr, i32 %val, i32 5, metadata !0)
+  ret void
+}
+
+define void @test_flat_amdgcn_cooperative_atomic_store_16x8B_seq_cst(ptr noundef %addr, <2 x i32> noundef %val)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_store_16x8B_seq_cst:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b64 v[0:1], v[2:3] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.16x8B.p0(ptr %addr, <2 x i32> %val, i32 5, metadata !0)
+  ret void
+}
+
+define void @test_flat_amdgcn_cooperative_atomic_store_8x16B_seq_cst(ptr noundef %addr, <4 x i32> noundef %val)  {
+; GFX1250-LABEL: test_flat_amdgcn_cooperative_atomic_store_8x16B_seq_cst:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b128 v[0:1], v[2:5] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.8x16B.p0(ptr %addr, <4 x i32> %val, i32 5, metadata !0)
+  ret void
+}
+
+define i32 @test_one_as_flat_amdgcn_cooperative_atomic_load_32x4B_relaxed(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_load_32x4B_relaxed:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b32 v0, v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call i32 @llvm.amdgcn.cooperative.atomic.load.32x4B.p0(ptr %addr, i32 0, metadata !1)
+  ret i32 %0
+}
+
+define <2 x i32> @test_one_as_flat_amdgcn_cooperative_atomic_load_16x8B_relaxed(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_load_16x8B_relaxed:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b64 v[0:1], v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call <2 x i32> @llvm.amdgcn.cooperative.atomic.load.16x8B.p0(ptr %addr, i32 0, metadata !1)
+  ret <2 x i32> %0
+}
+
+define <4 x i32> @test_one_as_flat_amdgcn_cooperative_atomic_load_8x16B_relaxed(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_load_8x16B_relaxed:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b128 v[0:3], v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call <4 x i32> @llvm.amdgcn.cooperative.atomic.load.8x16B.p0(ptr %addr, i32 0, metadata !1)
+  ret <4 x i32> %0
+}
+
+define void @test_one_as_flat_amdgcn_cooperative_atomic_store_32x4B_relaxed(ptr noundef %addr, i32 noundef %val)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_store_32x4B_relaxed:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b32 v[0:1], v2 scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.32x4B.p0(ptr %addr, i32 %val, i32 0, metadata !1)
+  ret void
+}
+
+define void @test_one_as_flat_amdgcn_cooperative_atomic_store_16x8B_relaxed(ptr noundef %addr, <2 x i32> noundef %val)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_store_16x8B_relaxed:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b64 v[0:1], v[2:3] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.16x8B.p0(ptr %addr, <2 x i32> %val, i32 0, metadata !1)
+  ret void
+}
+
+define void @test_one_as_flat_amdgcn_cooperative_atomic_store_8x16B_relaxed(ptr noundef %addr, <4 x i32> noundef %val)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_store_8x16B_relaxed:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b128 v[0:1], v[2:5] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.8x16B.p0(ptr %addr, <4 x i32> %val, i32 0, metadata !1)
+  ret void
+}
+
+define i32 @test_one_as_flat_amdgcn_cooperative_atomic_load_32x4B_acquire(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_load_32x4B_acquire:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b32 v0, v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call i32 @llvm.amdgcn.cooperative.atomic.load.32x4B.p0(ptr %addr, i32 2, metadata !1)
+  ret i32 %0
+}
+
+define <2 x i32> @test_one_as_flat_amdgcn_cooperative_atomic_load_16x8B_acquire(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_load_16x8B_acquire:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b64 v[0:1], v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call <2 x i32> @llvm.amdgcn.cooperative.atomic.load.16x8B.p0(ptr %addr, i32 2, metadata !1)
+  ret <2 x i32> %0
+}
+
+define <4 x i32> @test_one_as_flat_amdgcn_cooperative_atomic_load_8x16B_acquire(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_load_8x16B_acquire:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b128 v[0:3], v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call <4 x i32> @llvm.amdgcn.cooperative.atomic.load.8x16B.p0(ptr %addr, i32 2, metadata !1)
+  ret <4 x i32> %0
+}
+
+define void @test_one_as_flat_amdgcn_cooperative_atomic_store_32x4B_release(ptr noundef %addr, i32 noundef %val)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_store_32x4B_release:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b32 v[0:1], v2 scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.32x4B.p0(ptr %addr, i32 %val, i32 3, metadata !1)
+  ret void
+}
+
+define void @test_one_as_flat_amdgcn_cooperative_atomic_store_16x8B_release(ptr noundef %addr, <2 x i32> noundef %val)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_store_16x8B_release:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b64 v[0:1], v[2:3] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.16x8B.p0(ptr %addr, <2 x i32> %val, i32 3, metadata !1)
+  ret void
+}
+
+define void @test_one_as_flat_amdgcn_cooperative_atomic_store_8x16B_release(ptr noundef %addr, <4 x i32> noundef %val)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_store_8x16B_release:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b128 v[0:1], v[2:5] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.8x16B.p0(ptr %addr, <4 x i32> %val, i32 3, metadata !1)
+  ret void
+}
+
+define i32 @test_one_as_flat_amdgcn_cooperative_atomic_load_32x4B_seq_cst(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_load_32x4B_seq_cst:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b32 v0, v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call i32 @llvm.amdgcn.cooperative.atomic.load.32x4B.p0(ptr %addr, i32 5, metadata !1)
+  ret i32 %0
+}
+
+define <2 x i32> @test_one_as_flat_amdgcn_cooperative_atomic_load_16x8B_seq_cst(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_load_16x8B_seq_cst:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b64 v[0:1], v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call <2 x i32> @llvm.amdgcn.cooperative.atomic.load.16x8B.p0(ptr %addr, i32 5, metadata !1)
+  ret <2 x i32> %0
+}
+
+define <4 x i32> @test_one_as_flat_amdgcn_cooperative_atomic_load_8x16B_seq_cst(ptr noundef readonly %addr)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_load_8x16B_seq_cst:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_load_b128 v[0:3], v[0:1] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  %0 = tail call <4 x i32> @llvm.amdgcn.cooperative.atomic.load.8x16B.p0(ptr %addr, i32 5, metadata !1)
+  ret <4 x i32> %0
+}
+
+define void @test_one_as_flat_amdgcn_cooperative_atomic_store_32x4B_seq_cst(ptr noundef %addr, i32 noundef %val)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_store_32x4B_seq_cst:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b32 v[0:1], v2 scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.32x4B.p0(ptr %addr, i32 %val, i32 5, metadata !1)
+  ret void
+}
+
+define void @test_one_as_flat_amdgcn_cooperative_atomic_store_16x8B_seq_cst(ptr noundef %addr, <2 x i32> noundef %val)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_store_16x8B_seq_cst:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b64 v[0:1], v[2:3] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.16x8B.p0(ptr %addr, <2 x i32> %val, i32 5, metadata !1)
+  ret void
+}
+
+define void @test_one_as_flat_amdgcn_cooperative_atomic_store_8x16B_seq_cst(ptr noundef %addr, <4 x i32> noundef %val)  {
+; GFX1250-LABEL: test_one_as_flat_amdgcn_cooperative_atomic_store_8x16B_seq_cst:
+; GFX1250:       ; %bb.0: ; %entry
+; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-NEXT:    flat_store_b128 v[0:1], v[2:5] scope:SCOPE_DEV
+; GFX1250-NEXT:    s_wait_dscnt 0x0
+; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+entry:
+  tail call void @llvm.amdgcn.cooperative.atomic.store.8x16B.p0(ptr %addr, <4 x i32> %val, i32 5, metadata !1)
+  ret void
+}
+
+!0 = !{ !"singlethread" }
+!1 = !{ !"singlethread-one-as" }
+
+;; GFX1250: {{.*}}
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; GFX1250-GISEL: {{.*}}
+; GFX1250-SDAG: {{.*}}