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

[Triton-MLIR] tt.dot operands now must have DotOperand layout; also added prefetch pass prototype (#712)

Browse Source 
Co-authored-by: Jokeren <kerenzhou@openai.com>
Co-authored-by: Phil Tillet <phil@openai.com>
Co-authored-by: Superjomn <yanchunwei@outlook.com>
This commit is contained in:
Da Yan
2022-11-10 13:57:27 +08:00
committed by GitHub
parent 8832e32683
commit 4946167241
29 changed files with 1227 additions and 507 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
lib/Dialect/TritonGPU/Transforms/TritonGPUConversion.cpp
View File

@@ -95,8 +95,8 @@ TritonGPUConversionTarget::TritonGPUConversionTarget(
dotOp.a().getType().cast<RankedTensorType>().getEncoding();
Attribute bEncoding =
dotOp.b().getType().cast<RankedTensorType>().getEncoding();
if (aEncoding && aEncoding.isa<triton::gpu::SharedEncodingAttr>() &&
bEncoding && bEncoding.isa<triton::gpu::SharedEncodingAttr>())
if (aEncoding && aEncoding.isa<triton::gpu::DotOperandEncodingAttr>() &&
bEncoding && bEncoding.isa<triton::gpu::DotOperandEncodingAttr>())
return true;
return false;
});