[mlir][vector] Create `VectorToLLVMDialectInterface` #121440

Hardcode84 · 2025-01-01T16:42:18Z

Create VectorToLLVMDialectInterface which allows automatic conversion discovery by generic --convert-to-llvm pass. This only covers final dialect conversion step and not any previous preparation steps. Also, currently there is no way to pass any additional parameters through this conversion interface, but most users using default parameters anyway.

llvmbot · 2025-01-01T16:42:55Z

@llvm/pr-subscribers-mlir-vector

@llvm/pr-subscribers-mlir

Author: Ivan Butygin (Hardcode84)

Changes

Create VectorToLLVMDialectInterface which allows automatic conversion discovery by generic --convert-to-llvm pass. This only covers final dialect conversion step and not any previous preparation steps. Also, currently there is no way to pass any additional parameters through this conversion interface, but most users using default parameters anyway.

Full diff: https://github.com/llvm/llvm-project/pull/121440.diff

5 Files Affected:

(modified) mlir/include/mlir/Conversion/VectorToLLVM/ConvertVectorToLLVM.h (+3)
(modified) mlir/include/mlir/InitAllExtensions.h (+2)
(modified) mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp (+25)
(modified) mlir/lib/Dialect/Vector/IR/VectorOps.cpp (+2)
(added) mlir/test/Conversion/VectorToLLVM/vector-to-llvm-interface.mlir (+14)

diff --git a/mlir/include/mlir/Conversion/VectorToLLVM/ConvertVectorToLLVM.h b/mlir/include/mlir/Conversion/VectorToLLVM/ConvertVectorToLLVM.h
index 5fda62e3584c79..1e29bfeb9c3921 100644
--- a/mlir/include/mlir/Conversion/VectorToLLVM/ConvertVectorToLLVM.h
+++ b/mlir/include/mlir/Conversion/VectorToLLVM/ConvertVectorToLLVM.h
@@ -24,6 +24,9 @@ void populateVectorToLLVMConversionPatterns(
     const LLVMTypeConverter &converter, RewritePatternSet &patterns,
     bool reassociateFPReductions = false, bool force32BitVectorIndices = false);
 
+namespace vector {
+void registerConvertVectorToLLVMInterface(DialectRegistry &registry);
+}
 } // namespace mlir
 
 #endif // MLIR_CONVERSION_VECTORTOLLVM_CONVERTVECTORTOLLVM_H_
diff --git a/mlir/include/mlir/InitAllExtensions.h b/mlir/include/mlir/InitAllExtensions.h
index 14a6a2787b3a5d..887db344ed88b6 100644
--- a/mlir/include/mlir/InitAllExtensions.h
+++ b/mlir/include/mlir/InitAllExtensions.h
@@ -26,6 +26,7 @@
 #include "mlir/Conversion/NVVMToLLVM/NVVMToLLVM.h"
 #include "mlir/Conversion/OpenMPToLLVM/ConvertOpenMPToLLVM.h"
 #include "mlir/Conversion/UBToLLVM/UBToLLVM.h"
+#include "mlir/Conversion/VectorToLLVM/ConvertVectorToLLVM.h"
 #include "mlir/Dialect/AMX/Transforms.h"
 #include "mlir/Dialect/Affine/TransformOps/AffineTransformOps.h"
 #include "mlir/Dialect/Bufferization/TransformOps/BufferizationTransformOps.h"
@@ -76,6 +77,7 @@ inline void registerAllExtensions(DialectRegistry &registry) {
   registerConvertAMXToLLVMInterface(registry);
   gpu::registerConvertGpuToLLVMInterface(registry);
   NVVM::registerConvertGpuToNVVMInterface(registry);
+  vector::registerConvertVectorToLLVMInterface(registry);
 
   // Register all transform dialect extensions.
   affine::registerTransformDialectExtension(registry);
diff --git a/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp b/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp
index 9657f583c375bb..a30f6479e8a705 100644
--- a/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp
+++ b/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp
@@ -9,6 +9,7 @@
 #include "mlir/Conversion/VectorToLLVM/ConvertVectorToLLVM.h"
 
 #include "mlir/Conversion/ArithCommon/AttrToLLVMConverter.h"
+#include "mlir/Conversion/ConvertToLLVM/ToLLVMInterface.h"
 #include "mlir/Conversion/LLVMCommon/PrintCallHelper.h"
 #include "mlir/Conversion/LLVMCommon/TypeConverter.h"
 #include "mlir/Conversion/LLVMCommon/VectorPattern.h"
@@ -1933,3 +1934,27 @@ void mlir::populateVectorToLLVMMatrixConversionPatterns(
   patterns.add<VectorMatmulOpConversion>(converter);
   patterns.add<VectorFlatTransposeOpConversion>(converter);
 }
+
+namespace {
+struct VectorToLLVMDialectInterface : public ConvertToLLVMPatternInterface {
+  using ConvertToLLVMPatternInterface::ConvertToLLVMPatternInterface;
+  void loadDependentDialects(MLIRContext *context) const final {
+    context->loadDialect<LLVM::LLVMDialect>();
+  }
+
+  /// Hook for derived dialect interface to provide conversion patterns
+  /// and mark dialect legal for the conversion target.
+  void populateConvertToLLVMConversionPatterns(
+      ConversionTarget &target, LLVMTypeConverter &typeConverter,
+      RewritePatternSet &patterns) const final {
+    populateVectorToLLVMConversionPatterns(typeConverter, patterns);
+  }
+};
+} // namespace
+
+void mlir::vector::registerConvertVectorToLLVMInterface(
+    DialectRegistry &registry) {
+  registry.addExtension(+[](MLIRContext *ctx, vector::VectorDialect *dialect) {
+    dialect->addInterfaces<VectorToLLVMDialectInterface>();
+  });
+}
diff --git a/mlir/lib/Dialect/Vector/IR/VectorOps.cpp b/mlir/lib/Dialect/Vector/IR/VectorOps.cpp
index ae1cf95732336a..b57d89f0a96c01 100644
--- a/mlir/lib/Dialect/Vector/IR/VectorOps.cpp
+++ b/mlir/lib/Dialect/Vector/IR/VectorOps.cpp
@@ -13,6 +13,7 @@
 
 #include "mlir/Dialect/Vector/IR/VectorOps.h"
 
+#include "mlir/Conversion/ConvertToLLVM/ToLLVMInterface.h"
 #include "mlir/Dialect/Affine/IR/ValueBoundsOpInterfaceImpl.h"
 #include "mlir/Dialect/Arith/IR/Arith.h"
 #include "mlir/Dialect/Arith/Utils/Utils.h"
@@ -429,6 +430,7 @@ void VectorDialect::initialize() {
                             TransferWriteOp>();
   declarePromisedInterface<SubsetExtractionOpInterface, TransferReadOp>();
   declarePromisedInterface<SubsetInsertionOpInterface, TransferWriteOp>();
+  declarePromisedInterface<ConvertToLLVMPatternInterface, VectorDialect>();
 }
 
 /// Materialize a single constant operation from a given attribute value with
diff --git a/mlir/test/Conversion/VectorToLLVM/vector-to-llvm-interface.mlir b/mlir/test/Conversion/VectorToLLVM/vector-to-llvm-interface.mlir
new file mode 100644
index 00000000000000..5252bb25ecab54
--- /dev/null
+++ b/mlir/test/Conversion/VectorToLLVM/vector-to-llvm-interface.mlir
@@ -0,0 +1,14 @@
+// Most of the vector lowering is tested in vector-to-llvm.mlir, this file only for the interface smoke test
+// RUN: mlir-opt --convert-to-llvm="filter-dialects=vector" --split-input-file %s | FileCheck %s
+
+func.func @bitcast_f32_to_i32_vector_0d(%arg0: vector<f32>) -> vector<i32> {
+  %0 = vector.bitcast %arg0 : vector<f32> to vector<i32>
+  return %0 : vector<i32>
+}
+
+// CHECK-LABEL: @bitcast_f32_to_i32_vector_0d
+// CHECK-SAME:  %[[ARG_0:.*]]: vector<f32>
+// CHECK:       %[[VEC_F32_1D:.*]] = builtin.unrealized_conversion_cast %[[ARG_0]] : vector<f32> to vector<1xf32>
+// CHECK:       %[[VEC_I32_1D:.*]] = llvm.bitcast %[[VEC_F32_1D]] : vector<1xf32> to vector<1xi32>
+// CHECK:       %[[VEC_I32_0D:.*]] = builtin.unrealized_conversion_cast %[[VEC_I32_1D]] : vector<1xi32> to vector<i32>
+// CHECK:       return %[[VEC_I32_0D]] : vector<i32>

Hardcode84 · 2025-01-12T16:18:21Z

ping

Hardcode84 · 2025-01-20T16:27:39Z

ping

banach-space · 2025-01-22T16:49:01Z

Thanks! Could this be used as a separate RUN line in https://github.com/llvm/llvm-project/blob/main/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir? Otherwise, what's the benefit?

Hardcode84 · 2025-01-22T19:10:19Z

Thanks! Could this be used as a separate RUN line in https://github.com/llvm/llvm-project/blob/main/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir? Otherwise, what's the benefit?

We cannot (easily) do this. These tests designed to work with convert-vector-to-llvm pass which includes some preparations patterns running through applyPatternsGreedily.
Also, this pass includes armNeon/armSVE/amx/x86Vector specific transforms (which should probably be handled through their own dedicated ToLLVMDialectInterfaces).

Motivation for this one is to unhardcode conversions from GPUtoROCDL/NVVM passes https://github.com/llvm/llvm-project/blob/main/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp#L296

banach-space · 2025-01-23T18:09:05Z

Thanks for getting back to me.

We cannot (easily) do this. These tests designed to work with convert-vector-to-llvm pass which includes some preparations patterns running through applyPatternsGreedily.

This doesn't sound right. The tests in "Conversion/VectorToLLVM" should exercise the dialect conversion, i.e.

llvm-project/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

Lines 1906 to 1929 in d839c06

    
           void mlir::populateVectorToLLVMConversionPatterns( 
        
               const LLVMTypeConverter &converter, RewritePatternSet &patterns, 
        
               bool reassociateFPReductions, bool force32BitVectorIndices) { 
        
             // This function populates only ConversionPatterns, not RewritePatterns. 
        
             MLIRContext *ctx = converter.getDialect()->getContext(); 
        
             patterns.add<VectorReductionOpConversion>(converter, reassociateFPReductions); 
        
             patterns.add<VectorCreateMaskOpConversion>(ctx, force32BitVectorIndices); 
        
             patterns.add<VectorBitCastOpConversion, VectorShuffleOpConversion, 
        
                          VectorExtractElementOpConversion, VectorExtractOpConversion, 
        
                          VectorFMAOp1DConversion, VectorInsertElementOpConversion, 
        
                          VectorInsertOpConversion, VectorPrintOpConversion, 
        
                          VectorTypeCastOpConversion, VectorScaleOpConversion, 
        
                          VectorLoadStoreConversion<vector::LoadOp>, 
        
                          VectorLoadStoreConversion<vector::MaskedLoadOp>, 
        
                          VectorLoadStoreConversion<vector::StoreOp>, 
        
                          VectorLoadStoreConversion<vector::MaskedStoreOp>, 
        
                          VectorGatherOpConversion, VectorScatterOpConversion, 
        
                          VectorExpandLoadOpConversion, VectorCompressStoreOpConversion, 
        
                          VectorSplatOpLowering, VectorSplatNdOpLowering, 
        
                          VectorScalableInsertOpLowering, VectorScalableExtractOpLowering, 
        
                          MaskedReductionOpConversion, VectorInterleaveOpLowering, 
        
                          VectorDeinterleaveOpLowering, VectorFromElementsLowering, 
        
                          VectorScalableStepOpLowering>(converter); 
        
           }

That's exactly what the newly proposed VectorToLLVMDialectInterface is meant to run, right?

So, could you provide specific examples where things break (i.e. my assumption doesn't hold)? It's possible that "vector-to-llvm.mlir" contains tests that shouldn't be there. If that's the case, we should use this opportunity to fix that.

Also, this pass includes armNeon/armSVE/amx/x86Vector specific transforms (which should probably be handled through their own dedicated ToLLVMDialectInterfaces).

There's nothing architecture specific in "vector-to-llvm.mlir", so none of these should be required. If that's not the case, we should that file.

Motivation for this one is to unhardcode conversions from GPUtoROCDL/NVVM passes https://github.com/llvm/llvm-project/blob/main/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp#L296

Right now it's not clear to me how that's connected, but if it's going to help, then we should make it happen. In fact, it would be good to make sure that this change unblocks something useful. So - why not include the changes for "LowerGpuOpsToROCDLOps.cpp" in this PR?

Hardcode84 · 2025-01-24T13:12:07Z

So, could you provide specific examples where things break (i.e. my assumption doesn't hold)? It's possible that "vector-to-llvm.mlir" contains tests that shouldn't be there. If that's the case, we should use this opportunity to fix that.

There are a lot of them, full log: https://gist.github.com/Hardcode84/e1de4dc58ea7b47275e1aaff89e5567b

Hardcode84 · 2025-01-24T13:19:39Z

What we can probably do is to extract these initial transforms into dedicated pass and run it before generic convert-to-llvm for the vector-to-llvm.mlir.

banach-space · 2025-01-24T16:49:09Z

What we can probably do is to extract these initial transforms into dedicated pass and run it before generic convert-to-llvm for the vector-to-llvm.mlir.

I think that we can move this discussion elsewhere:

[mlir][vector] What passes should we test in vector-to-llvm.mlir? #124308

As for this PR specifically, are you able to provide an example how it would benefit "LowerGpuOpsToROCDLOps.cpp"? To me that would be sufficient to unblock this.

Hardcode84 · 2025-01-26T02:00:28Z

As for this PR specifically, are you able to provide an example how it would benefit "LowerGpuOpsToROCDLOps.cpp"? To me that would be sufficient to unblock this.

#124439

Create `VectorToLLVMDialectInterface` which allows automatic conversion discovery by generic `--convert-to-llvm` pass. This only covers final dialect conversion step and not any previous preparation steps. Also, currently there is no way to pass any additional parameters through this conversion interface, but most users using default parameters anyway.

banach-space

LGTM, thanks!

Here's a follow-up to fully leverage the newly introduced flag:

[mlir][vector] Categorize Vector-to-LLVM conversion tests (nfc) #125918

(also a follow-up for #124308)

#121440

llvm/llvm-project#121440

Create `VectorToLLVMDialectInterface` which allows automatic conversion discovery by generic `--convert-to-llvm` pass. This only covers final dialect conversion step and not any previous preparation steps. Also, currently there is no way to pass any additional parameters through this conversion interface, but most users using default parameters anyway.

llvm#121440

Hardcode84 requested review from banach-space, dcaballe, nicolasvasilache and Groverkss as code owners January 1, 2025 16:42

llvmbot added mlir:vectorops mlir mlir:vector labels Jan 1, 2025

Hardcode84 force-pushed the vector-translate branch from 153dbeb to 1ced298 Compare January 12, 2025 16:18

banach-space mentioned this pull request Jan 24, 2025

[mlir][vector] What passes should we test in vector-to-llvm.mlir? #124308

Closed

Hardcode84 force-pushed the vector-translate branch from 1ced298 to 7743c6a Compare January 31, 2025 01:07

Hardcode84 mentioned this pull request Feb 3, 2025

[mlir][gpu] GPUToROCDL/NVVM: use generic llvm conversion interface instead of hardcoded conversions. #124439

Merged

banach-space mentioned this pull request Feb 5, 2025

[mlir][vector] Categorize Vector-to-LLVM conversion tests (nfc) #125918

Merged

banach-space approved these changes Feb 5, 2025

View reviewed changes

Hardcode84 merged commit 6e52a12 into llvm:main Feb 5, 2025
8 checks passed

Hardcode84 deleted the vector-translate branch February 5, 2025 20:21

rupprecht mentioned this pull request Feb 5, 2025

[bazel][mlir][vector] Port 6e52a12811a17e5871f403f04fd4082d43160bf3 #125936

Merged

rupprecht added a commit that referenced this pull request Feb 5, 2025

[bazel][mlir][vector] Port 6e52a12 (#125936)

db9caf6

#121440

github-actions bot pushed a commit to arm/arm-toolchain that referenced this pull request Feb 5, 2025

Automerge: [bazel][mlir][vector] Port 6e52a12 (#125936)

59c279e

llvm/llvm-project#121440

Icohedron pushed a commit to Icohedron/llvm-project that referenced this pull request Feb 11, 2025

[bazel][mlir][vector] Port 6e52a12 (llvm#125936)

bdf1d48

llvm#121440

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[mlir][vector] Create `VectorToLLVMDialectInterface` #121440

[mlir][vector] Create `VectorToLLVMDialectInterface` #121440

Uh oh!

Hardcode84 commented Jan 1, 2025

Uh oh!

llvmbot commented Jan 1, 2025 •

edited

Loading

Uh oh!

Hardcode84 commented Jan 12, 2025

Uh oh!

Hardcode84 commented Jan 20, 2025

Uh oh!

banach-space commented Jan 22, 2025

Uh oh!

Hardcode84 commented Jan 22, 2025 •

edited

Loading

Uh oh!

banach-space commented Jan 23, 2025

Uh oh!

Hardcode84 commented Jan 24, 2025 •

edited

Loading

Uh oh!

Hardcode84 commented Jan 24, 2025

Uh oh!

banach-space commented Jan 24, 2025

Uh oh!

Hardcode84 commented Jan 26, 2025

Uh oh!

banach-space left a comment

Uh oh!

Uh oh!

Uh oh!

[mlir][vector] Create VectorToLLVMDialectInterface #121440

[mlir][vector] Create VectorToLLVMDialectInterface #121440

Uh oh!

Conversation

Hardcode84 commented Jan 1, 2025

Uh oh!

llvmbot commented Jan 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Hardcode84 commented Jan 12, 2025

Uh oh!

Hardcode84 commented Jan 20, 2025

Uh oh!

banach-space commented Jan 22, 2025

Uh oh!

Hardcode84 commented Jan 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

banach-space commented Jan 23, 2025

Uh oh!

Hardcode84 commented Jan 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Hardcode84 commented Jan 24, 2025

Uh oh!

banach-space commented Jan 24, 2025

Uh oh!

Hardcode84 commented Jan 26, 2025

Uh oh!

banach-space left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

[mlir][vector] Create `VectorToLLVMDialectInterface` #121440

[mlir][vector] Create `VectorToLLVMDialectInterface` #121440

llvmbot commented Jan 1, 2025 •

edited

Loading

Hardcode84 commented Jan 22, 2025 •

edited

Loading

Hardcode84 commented Jan 24, 2025 •

edited

Loading