[AMD backend] Fix unit test test_dot_without_load
#3338
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
zhanglx13
reviewed
Mar 11, 2024
| @@ -87,7 +87,7 @@ Type TritonGPUToLLVMTypeConverter::getElementTypeForStruct( | |||
| return elemTy; | |||
| if (auto mfmaParent = | |||
| dotOpLayout.getParent().dyn_cast<AMDMfmaEncodingAttr>()) { | |||
| return vec_ty(elemTy, dotOpLayout.getKWidth()); | |||
| return elemTy; | |||
There was a problem hiding this comment.
Since this is not a special case anymore, we can just remove it.
zhanglx13
reviewed
Mar 11, 2024
lib/Dialect/TritonGPU/IR/Dialect.cpp
Outdated
| @@ -43,7 +43,15 @@ namespace gpu { | |||
| unsigned getTotalElemsPerThread(Attribute layout, ArrayRef<int64_t> shape, | |||
| Type eltTy) { | |||
| if (auto tritonGPUAttr = layout.dyn_cast<TritonGPU_AttrTrait>()) { | |||
| return tritonGPUAttr.getTotalElemsPerThread(shape, eltTy); | |||
| unsigned elemNum = tritonGPUAttr.getTotalElemsPerThread(shape, eltTy); | |||
There was a problem hiding this comment.
I think it's better not to touch this "interface" function. The changes are for dotOp with mfma as parent anyway. Can we move the changes in DotOperandEncodingAttr::getTotalElemsPerThread?
There was a problem hiding this comment.
Thanks. Changed accordingly.
scxiao
force-pushed
the
fix_dot_without_load
branch
from
694cd67 to
eda0398
Compare
zhanglx13
approved these changes
Mar 11, 2024
|
Hi @ThomasRaoux, could you please take a look at this PR to see if you have any comments? Thanks. |
|
looks like this fails CI: |
OK, let me check that. Sorry about that. |
zhanglx13
added a commit
that referenced
this pull request
Mar 12, 2024
This reverts commit e0e5a36.
ThomasRaoux
pushed a commit
that referenced
this pull request
Mar 12, 2024
zhanglx13
pushed a commit
that referenced
this pull request
Mar 13, 2024
This PR is actually fix the regression in the reverted PR: #3338, which caused a regression for the test `test_masked_load_shared_memory`. The reason is for type used in packing dot_op for bfloat16. We should use the type `i16` for `bf16` when packing dot_op for mfma. This time I ran all the tests in `test_core.py` locally, and all work fine.
htyu
pushed a commit
to htyu/triton
that referenced
this pull request
Mar 20, 2024
This PR is to fix the unit test `test_dot_without_load` by by changing dot op from `vector<vector<type>>` to `vector<type>`, so the constantOp can be converted to mfma `dot_op` with existing lowering code.
htyu
pushed a commit
to htyu/triton
that referenced
this pull request
Mar 20, 2024
htyu
pushed a commit
to htyu/triton
that referenced
this pull request
Mar 20, 2024
This PR is actually fix the regression in the reverted PR: triton-lang#3338, which caused a regression for the test `test_masked_load_shared_memory`. The reason is for type used in packing dot_op for bfloat16. We should use the type `i16` for `bf16` when packing dot_op for mfma. This time I ran all the tests in `test_core.py` locally, and all work fine.
karupayun
pushed a commit
to openxla/triton
that referenced
this pull request
Apr 3, 2024
This PR is to fix the unit test `test_dot_without_load` by by changing dot op from `vector<vector<type>>` to `vector<type>`, so the constantOp can be converted to mfma `dot_op` with existing lowering code.
karupayun
pushed a commit
to openxla/triton
that referenced
this pull request
Apr 3, 2024
karupayun
pushed a commit
to openxla/triton
that referenced
this pull request
Apr 3, 2024
This PR is actually fix the regression in the reverted PR: triton-lang#3338, which caused a regression for the test `test_masked_load_shared_memory`. The reason is for type used in packing dot_op for bfloat16. We should use the type `i16` for `bf16` when packing dot_op for mfma. This time I ran all the tests in `test_core.py` locally, and all work fine.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR is to fix the unit test
test_dot_without_loadby by changing dot op fromvector<vector<type>>tovector<type>, so the constantOp can be converted to mfmadot_opwith existing lowering code.