[mlir][nvvm] Fix the PTX lowering of wgmma.mma_async #76150
Merged
Conversation
This file contains hidden or 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
apaszke
force-pushed
the
wgmma-transpose-b
branch
from
5eef4c0 to
0aff6ae
Compare
|
@llvm/pr-subscribers-mlir-llvm @llvm/pr-subscribers-mlir Author: Adam Paszke (apaszke) ChangesThe default layout of A and B matrices is row- and column-major respectively, meaning that the transpose flags have opposite meanings between those two operands. Full diff: https://github.com/llvm/llvm-project/pull/76150.diff 1 Files Affected:
diff --git a/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp b/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
index 4f5d71e10f68c1..a4de89d928e1be 100644
--- a/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
+++ b/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
@@ -1003,7 +1003,7 @@ void NVVM::WgmmaMmaAsyncOp::getAsmValues(
{makeConstantI32(rewriter, static_cast<int>(getLayoutA())),
mlir::NVVM::PTXRegisterMod::Read});
asmValues.push_back(
- {makeConstantI32(rewriter, static_cast<int>(getLayoutB())),
+ {makeConstantI32(rewriter, 1 - static_cast<int>(getLayoutB())),
mlir::NVVM::PTXRegisterMod::Read});
}
}
|
|
Can this be exercise by a unit-test? |
|
Yeah seems like |
The default layout of A and B matrices is row- and column-major respectively, meaning that the transpose flags have opposite meanings between those two operands.
apaszke
force-pushed
the
wgmma-transpose-b
branch
from
0aff6ae to
3c22209
Compare
|
Ok the test should be updated now. It does a col-col matmul, so the right transpose args are |
|
Good catch. |
grypp
approved these changes
Dec 21, 2023
|
Seems like the Windows failure is unrelated to this PR? |
|
Yes, let me merge this |
grypp
added a commit
to grypp/llvm-project
that referenced
this pull request
Jan 24, 2024
The llvm#76150 fixed meaning of `transposeB` in NVVM dialect which was initially implemented with opposite meaning. This PR fixes the lowering of `nvgpu.warpgroup.mma` to NVVM dialect. This will fix two integration tests: gemm_f32_f16_f16_128x128x128.mlir gemm_pred_f32_f16_f16_128x128x128.mlir
grypp
added a commit
that referenced
this pull request
Jan 25, 2024
The #76150 fixed meaning of `transposeB` in NVVM dialect which was initially implemented with opposite meaning. This PR fixes the lowering of `nvgpu.warpgroup.mma` to NVVM dialect. This will fix two integration tests: gemm_f32_f16_f16_128x128x128.mlir gemm_pred_f32_f16_f16_128x128x128.mlir
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.
The default layout of A and B matrices is row- and column-major respectively, meaning that the transpose flags have opposite meanings between those two operands.