[mlir][Vector][NFC] Add vector-transform-options flag to ConvertVectorToLLVMPass
#123491
Conversation
…torToLLVMPass This flag enables the configuration of some transformation such as the lowering of contractions and transposes. The default configuration preserves the existing behavior.
dcaballe
requested review from
banach-space and
nicolasvasilache
as code owners
|
@llvm/pr-subscribers-mlir Author: Diego Caballero (dcaballe) ChangesThis flag enables the configuration of some transformation such as the lowering of contractions and transposes. The default configuration preserves the existing behavior. Full diff: https://github.com/llvm/llvm-project/pull/123491.diff 3 Files Affected:
diff --git a/mlir/include/mlir/Conversion/Passes.td b/mlir/include/mlir/Conversion/Passes.td
index 0f42ffb3a80266..7600d723c30936 100644
--- a/mlir/include/mlir/Conversion/Passes.td
+++ b/mlir/include/mlir/Conversion/Passes.td
@@ -1440,7 +1440,11 @@ def ConvertVectorToLLVMPass : Pass<"convert-vector-to-llvm"> {
Option<"x86Vector", "enable-x86vector",
"bool", /*default=*/"false",
"Enables the use of X86Vector dialect while lowering the vector "
- "dialect.">
+ "dialect.">,
+ Option<"vectorTransformsOptions", "vector-transform-options",
+ "vector::VectorTransformsOptions",
+ /*default=*/"vector::VectorTransformsOptions()",
+ "Options to lower some operations like contractions and transposes.">,
];
}
diff --git a/mlir/include/mlir/Conversion/VectorToLLVM/ConvertVectorToLLVMPass.h b/mlir/include/mlir/Conversion/VectorToLLVM/ConvertVectorToLLVMPass.h
index 4661d31b6364d5..410b881db79590 100644
--- a/mlir/include/mlir/Conversion/VectorToLLVM/ConvertVectorToLLVMPass.h
+++ b/mlir/include/mlir/Conversion/VectorToLLVM/ConvertVectorToLLVMPass.h
@@ -9,6 +9,7 @@
#define MLIR_CONVERSION_VECTORTOLLVM_CONVERTVECTORTOLLVMPASS_H_
#include "mlir/Conversion/VectorToLLVM/ConvertVectorToLLVM.h"
+#include "mlir/Dialect/Vector/Transforms/VectorTransforms.h"
namespace mlir {
class Pass;
diff --git a/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVMPass.cpp b/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVMPass.cpp
index 2c4c5ada9815d6..e3a81bd20212d8 100644
--- a/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVMPass.cpp
+++ b/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVMPass.cpp
@@ -69,12 +69,11 @@ void ConvertVectorToLLVMPass::runOnOperation() {
populateVectorToVectorCanonicalizationPatterns(patterns);
populateVectorBitCastLoweringPatterns(patterns);
populateVectorBroadcastLoweringPatterns(patterns);
- populateVectorContractLoweringPatterns(patterns, VectorTransformsOptions());
+ populateVectorContractLoweringPatterns(patterns, vectorTransformsOptions);
populateVectorMaskOpLoweringPatterns(patterns);
populateVectorShapeCastLoweringPatterns(patterns);
populateVectorInterleaveLoweringPatterns(patterns);
- populateVectorTransposeLoweringPatterns(patterns,
- VectorTransformsOptions());
+ populateVectorTransposeLoweringPatterns(patterns, vectorTransformsOptions);
// Vector transfer ops with rank > 1 should be lowered with VectorToSCF.
populateVectorTransferLoweringPatterns(patterns, /*maxTransferRank=*/1);
populateVectorMaskMaterializationPatterns(patterns,
|
|
Hello! @dcaballe Could you help with this issue? Maybe there are some WA or example how to write printer for this. |
This was raised on Discourse: Subsequent fix: |
Sorry about that. I just reviewed the fix but it needs more reviewers to take a look. If this is blocking you, please, let me know and I'll revert this PR in the meantime. |
|
Ok, got it, thanks. Sorry, google didn't show me the discussion on Discourse for some reason. |
This flag enables the configuration of some transformation such as the lowering of contractions and transposes. The default configuration preserves the existing behavior.