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

[FRONTEND] add an attr for masked load without explicit other (#55)

Browse Source
This commit is contained in:
Shintaro Iwasaki
2022-08-18 09:51:37 -07:00
committed by GitHub
parent fc58250a06
commit d69ce77b19
16 changed files with 71 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
test/Conversion/tritongpu_to_llvm.mlir
View File

@@ -21,7 +21,7 @@ module attributes {"triton_gpu.num-warps" = 4 : i32} {
func @basic_load(%a_ptr_init : tensor<256x!tt.ptr<f32>, #blocked0>, %cst : tensor<256xi1, #blocked0>, %cst_0 : tensor<256xf32, #blocked0>) {
// CHECK: llvm.inline_asm
// CHECK: llvm.inline_asm
%1 = tt.load %a_ptr_init, %cst, %cst_0 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<256xf32, #blocked0>
%1 = tt.load %a_ptr_init, %cst, %cst_0 {cache = 1 : i32, evict = 1 : i32, isOtherUnspecified = false, isVolatile = false} : tensor<256xf32, #blocked0>
return
}
}
@@ -36,7 +36,7 @@ module attributes {"triton_gpu.num-warps" = 4 : i32} {
// CHECK-SAME: ld.global.v4.b32
// CHECK: llvm.inline_asm
// CHECK-SAME: ld.global.v4.b32
%1 = tt.load %a_ptr_init, %cst, %cst_0 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<256xf32, #blocked0>
%1 = tt.load %a_ptr_init, %cst, %cst_0 {cache = 1 : i32, evict = 1 : i32, isOtherUnspecified = false, isVolatile = false} : tensor<256xf32, #blocked0>
return
}
}
@@ -51,7 +51,7 @@ module attributes {"triton_gpu.num-warps" = 4 : i32} {
// CHECK-SAME: ld.global.v2.b32
// CHECK: llvm.inline_asm
// CHECK-SAME: ld.global.v2.b32
%1 = tt.load %a_ptr_init, %cst, %cst_0 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<256xf16, #blocked0>
%1 = tt.load %a_ptr_init, %cst, %cst_0 {cache = 1 : i32, evict = 1 : i32, isOtherUnspecified = false, isVolatile = false} : tensor<256xf16, #blocked0>
return
}
}
@@ -64,7 +64,7 @@ module attributes {"triton_gpu.num-warps" = 4 : i32} {
// CHECK-LABEL: masked_load_const_other
func @masked_load_const_other(%a_ptr_init : tensor<256x!tt.ptr<f32>, #blocked0>, %cst : tensor<256xi1, #blocked0>) {
%cst_0 = arith.constant dense<0.000000e+00> : tensor<256xf32, #blocked0>
%1 = tt.load %a_ptr_init, %cst, %cst_0 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<256xf32, #blocked0>
%1 = tt.load %a_ptr_init, %cst, %cst_0 {cache = 1 : i32, evict = 1 : i32, isOtherUnspecified = false, isVolatile = false} : tensor<256xf32, #blocked0>
return
}
}