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

.

Browse Source
This commit is contained in:
Phil Tillet
2022-12-08 21:36:47 -08:00
parent 13644e7ac4
commit fa6dbbff60
2 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
lib/Dialect/TritonGPU/IR/Dialect.cpp
View File

@@ -593,6 +593,8 @@ Attribute DotOperandEncodingAttr::parse(AsmParser &parser, Type type) {
if(parent.isa<MmaEncodingAttr>() && if(parent.isa<MmaEncodingAttr>() &&
parent.cast<MmaEncodingAttr>().getVersion() == 1){ parent.cast<MmaEncodingAttr>().getVersion() == 1){
isMMAv1Row = attrs.get("isMMAv1Row"); isMMAv1Row = attrs.get("isMMAv1Row");
if(!isMMAv1Row)
llvm::report_fatal_error("isMMAv1Row attribute is missing");
} }
return parser.getChecked<DotOperandEncodingAttr>(parser.getContext(), opIdx, return parser.getChecked<DotOperandEncodingAttr>(parser.getContext(), opIdx,
parent, isMMAv1Row); parent, isMMAv1Row);

4
test/Conversion/tritongpu_to_llvm.mlir
View File

@@ -879,8 +879,8 @@ module attributes {"triton_gpu.num-warps" = 4 : i32} {
#blocked = #triton_gpu.blocked<{sizePerThread = [1, 4], threadsPerWarp = [2, 16], warpsPerCTA = [1, 4], order = [1, 0]}> #blocked = #triton_gpu.blocked<{sizePerThread = [1, 4], threadsPerWarp = [2, 16], warpsPerCTA = [1, 4], order = [1, 0]}>
#shared = #triton_gpu.shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [1, 0]}> #shared = #triton_gpu.shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [1, 0]}>
#mma = #triton_gpu.mma<{version = 1, warpsPerCTA = [2, 2]}> #mma = #triton_gpu.mma<{version = 1, warpsPerCTA = [2, 2]}>
#dot_operand_a = #triton_gpu.dot_op<{opIdx=0, parent=#mma}> #dot_operand_a = #triton_gpu.dot_op<{opIdx=0, parent=#mma, isMMAv1Row=true}>
#dot_operand_b = #triton_gpu.dot_op<{opIdx=1, parent=#mma}> #dot_operand_b = #triton_gpu.dot_op<{opIdx=1, parent=#mma, isMMAv1Row=true}>
module attributes {"triton_gpu.num-warps" = 4 : i32} { module attributes {"triton_gpu.num-warps" = 4 : i32} {
func @matmul884_kernel_dot_operand_layout(%ptr:!tt.ptr<f32> {tt.divisibility = 16 : i32}, func @matmul884_kernel_dot_operand_layout(%ptr:!tt.ptr<f32> {tt.divisibility = 16 : i32},
%a:tensor<128x32xf16, #shared>, %b:tensor<32x256xf16, #shared>) { %a:tensor<128x32xf16, #shared>, %b:tensor<32x256xf16, #shared>) {