diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/llvm.amdgcn.ds.append.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/llvm.amdgcn.ds.append.ll | 9 |
1 files changed, 9 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.ds.append.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.ds.append.ll index 132e3f717c5..9f7aa5ea3a1 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.ds.append.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.ds.append.ll @@ -7,6 +7,7 @@ ; GCN: s_load_dword [[PTR:s[0-9]+]] ; GCN: s_mov_b32 m0, [[PTR]] ; GCN: ds_append [[RESULT:v[0-9]+]]{{$}} +; GCN-NOT: buffer_wbinvl1 ; GCN: {{.*}}store{{.*}} [[RESULT]] define amdgpu_kernel void @ds_append_lds(i32 addrspace(3)* %lds, i32 addrspace(1)* %out) #0 { %val = call i32 @llvm.amdgcn.ds.append.p3i32(i32 addrspace(3)* %lds, i1 false) @@ -18,6 +19,7 @@ define amdgpu_kernel void @ds_append_lds(i32 addrspace(3)* %lds, i32 addrspace(1 ; GCN: s_load_dword [[PTR:s[0-9]+]] ; GCN: s_mov_b32 m0, [[PTR]] ; GCN: ds_append [[RESULT:v[0-9]+]] offset:65532{{$}} +; GCN-NOT: buffer_wbinvl1 ; GCN: {{.*}}store{{.*}} [[RESULT]] define amdgpu_kernel void @ds_append_lds_max_offset(i32 addrspace(3)* %lds, i32 addrspace(1)* %out) #0 { %gep = getelementptr inbounds i32, i32 addrspace(3)* %lds, i32 16383 @@ -36,6 +38,7 @@ define amdgpu_kernel void @ds_append_lds_max_offset(i32 addrspace(3)* %lds, i32 ; CIPLUS: s_mov_b32 m0, [[PTR]] ; CIPLUS: ds_append [[RESULT:v[0-9]+]] offset:16{{$}} +; GCN-NOT: buffer_wbinvl1 ; GCN: {{.*}}store{{.*}} [[RESULT]] define amdgpu_kernel void @ds_append_no_fold_offset_si(i32 addrspace(3)* addrspace(4)* %lds.ptr, i32 addrspace(1)* %out) #0 { %lds = load i32 addrspace(3)*, i32 addrspace(3)* addrspace(4)* %lds.ptr, align 4 @@ -53,6 +56,7 @@ define amdgpu_kernel void @ds_append_no_fold_offset_si(i32 addrspace(3)* addrspa ; GCN: s_mov_b32 m0, [[PTR]] ; GCN: ds_append [[RESULT:v[0-9]+]]{{$}} +; GCN-NOT: buffer_wbinvl1 ; GCN: {{.*}}store{{.*}} [[RESULT]] define amdgpu_kernel void @ds_append_lds_over_max_offset(i32 addrspace(3)* %lds, i32 addrspace(1)* %out) #0 { %gep = getelementptr inbounds i32, i32 addrspace(3)* %lds, i32 16384 @@ -65,6 +69,7 @@ define amdgpu_kernel void @ds_append_lds_over_max_offset(i32 addrspace(3)* %lds, ; GCN: v_readfirstlane_b32 [[READLANE:s[0-9]+]], v0 ; GCN: s_mov_b32 m0, [[READLANE]] ; GCN: ds_append [[RESULT:v[0-9]+]]{{$}} +; GCN-NOT: buffer_wbinvl1 ; GCN: {{.*}}store{{.*}} [[RESULT]] define void @ds_append_lds_vgpr_addr(i32 addrspace(3)* %lds, i32 addrspace(1)* %out) #0 { %val = call i32 @llvm.amdgcn.ds.append.p3i32(i32 addrspace(3)* %lds, i1 false) @@ -76,6 +81,7 @@ define void @ds_append_lds_vgpr_addr(i32 addrspace(3)* %lds, i32 addrspace(1)* % ; GCN: s_load_dword [[PTR:s[0-9]+]] ; GCN: s_mov_b32 m0, [[PTR]] ; GCN: ds_append [[RESULT:v[0-9]+]] gds{{$}} +; GCN-NOT: buffer_wbinvl1 ; GCN: {{.*}}store{{.*}} [[RESULT]] define amdgpu_kernel void @ds_append_gds(i32 addrspace(2)* %gds, i32 addrspace(1)* %out) #0 { %val = call i32 @llvm.amdgcn.ds.append.p2i32(i32 addrspace(2)* %gds, i1 false) @@ -87,6 +93,7 @@ define amdgpu_kernel void @ds_append_gds(i32 addrspace(2)* %gds, i32 addrspace(1 ; GCN: s_load_dword [[PTR:s[0-9]+]] ; GCN: s_mov_b32 m0, [[PTR]] ; GCN: ds_append [[RESULT:v[0-9]+]] offset:65532 gds{{$}} +; GCN-NOT: buffer_wbinvl1 ; GCN: {{.*}}store{{.*}} [[RESULT]] define amdgpu_kernel void @ds_append_gds_max_offset(i32 addrspace(2)* %gds, i32 addrspace(1)* %out) #0 { %gep = getelementptr inbounds i32, i32 addrspace(2)* %gds, i32 16383 @@ -96,6 +103,7 @@ define amdgpu_kernel void @ds_append_gds_max_offset(i32 addrspace(2)* %gds, i32 } ; GCN-LABEL: {{^}}ds_append_gds_over_max_offset: +; GCN-NOT: buffer_wbinvl1 define amdgpu_kernel void @ds_append_gds_over_max_offset(i32 addrspace(2)* %gds, i32 addrspace(1)* %out) #0 { %gep = getelementptr inbounds i32, i32 addrspace(2)* %gds, i32 16384 %val = call i32 @llvm.amdgcn.ds.append.p2i32(i32 addrspace(2)* %gep, i1 false) @@ -107,6 +115,7 @@ define amdgpu_kernel void @ds_append_gds_over_max_offset(i32 addrspace(2)* %gds, ; GCN: s_load_dword [[PTR:s[0-9]+]] ; GCN: s_mov_b32 m0, [[PTR]] ; GCN: ds_append [[RESULT:v[0-9]+]]{{$}} +; GCN-NOT: buffer_wbinvl1 ; NOTGFX9: s_mov_b32 m0, -1 ; GFX9-NOT: m0 ; GCN: _store_dword |
