gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[BACKEND] Add C++ tests for PTXFormat and some tiny refinement (#109)

Browse Source 
This PR does

1. Add some C++ tests for `PTXFormat`
2. Enhance the functionality of `PTXFormat`, make a `PTXInstr` instance
can be called multiple times similar as a C function.
This commit is contained in:
Yan Chunwei
2022-09-10 00:15:07 +08:00
committed by GitHub
parent a9464f4993
commit 2a852044d9
10 changed files with 231 additions and 92 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
test/Conversion/triton_to_llvm.mlir
View File

@@ -29,8 +29,8 @@ func @test_store_splat(%ptr: !tt.ptr<f32>) {
%vs = tt.splat %a : (f32) -> tensor<128xf32>
%mask = tt.splat %true : (i1) -> tensor<128xi1>
// CHECK: %{{.*}} = llvm.inline_asm has_side_effects asm_dialect = att operand_attrs = [] "@$1 st.global.b32 [ $2 + 0 ], { $0 };",
// CHECK-SAME: "r,b,l" %{{.*}}, %{{.*}}, %{{.*}} : (i32, i1, !llvm.ptr<f32, 1>) -> !llvm.void
// CHECK: %{{.*}} = llvm.inline_asm has_side_effects asm_dialect = att operand_attrs = [] "@${{.*}} st.global.b32 [ ${{.*}} + 0 ], { ${{.*}} };",
// CHECK-SAME: "r,l,b" %{{.*}}, %{{.*}}, %{{.*}} : (i32, !llvm.ptr<f32, 1>, i1) -> !llvm.void
tt.store %ptrs, %vs, %mask : tensor<128xf32>
return

4
test/Conversion/tritongpu_to_llvm.mlir
View File

@@ -217,9 +217,9 @@ module attributes {"triton_gpu.num-warps" = 4 : i32} {
// CHECK-LABEL: basic_store
func @basic_store(%ptrs: tensor<256x!tt.ptr<f32>, #blocked0>, %vals: tensor<256xf32, #blocked0>, %mask: tensor<256xi1, #blocked0>) {
// CHECK: llvm.inline_asm has_side_effects asm_dialect = att
// CHECK-SAME: st.global.b32 [ ${{.*}} + 0 ], { ${{.*}} };", "r,b,l" %{{.*}}, %{{.*}}, %{{.*}} : (i32, i1, !llvm.ptr<f32, 1>) -> !llvm.void
// CHECK-SAME: @${{.*}} st.global.b32 [ ${{.*}} + 0 ], { ${{.*}} };", "r,l,b" %{{.*}}, %{{.*}}, %{{.*}} : (i32, !llvm.ptr<f32, 1>, i1) -> !llvm.void
// CHECK: llvm.inline_asm has_side_effects asm_dialect = att
// CHECK-SAME: st.global.b32 [ ${{.*}} + 0 ], { ${{.*}} };", "r,b,l" %{{.*}}, %{{.*}}, %{{.*}} : (i32, i1, !llvm.ptr<f32, 1>) -> !llvm.void
// CHECK-SAME: @${{.*}} st.global.b32 [ ${{.*}} + 0 ], { ${{.*}} };", "r,l,b" %{{.*}}, %{{.*}}, %{{.*}} : (i32, !llvm.ptr<f32, 1>, i1) -> !llvm.void
tt.store %ptrs, %vals, %mask : tensor<256xf32, #blocked0>
return
}