[mlir][linalg] Retire Linalg's Vectorization Pattern

[guraypp]

Summary

This revision retires the LinalgCodegenStrategy vectorization pattern. Please see the context: https://discourse.llvm.org/t/psa-retire-linalg-filter-based-patterns/63785.
This revision improves the transform dialect's VectorizeOp in different ways below:

• Adds LinalgDialect as a dependent dialect. When transform.structured.vectorize vectorizes tensor.pad, it generates linalg.init_tensor. In this case, linalg dialect must be registered.
• Inserts CopyVectorizationPattern in order to vectorize memref.copy.
• Creates two attributes: disable_multi_reduction_to_contract_patterns and disable_transfer_permutation_map_lowering_patterns. They are limiting the power of vectorization and are currently intended for testing purposes.

It also removes some of the "CHECK: vector.transfer_write" in the vectorization.mlir test. They are redundant writes, at the end of the code there is a rewrite to the same place. Transform dialect no longer generates them.

Depends on D133684 that retires the LinalgCodegenStrategy vectorization pass.

[guraypp]

removed accidental code in the test

[guraypp]

update the pattern

[guraypp]

Add two options reduction_to_contract and permutation_map to VectorizeOp so we can use them for the tests in vectorization.mlir. Nevertheless, there are still tests failing. This revision marks the failing tests as "CHECKFAILS".

[nicolasvasilache]

mlir/test/lib/Dialect/Linalg/TestLinalgTransforms.cpp

438

This should go away completely.
Fine to do in a separate commit but please don't leave it behind.

[nicolasvasilache]

mlir/test/Dialect/Linalg/vectorization.mlir

373

memref.copy vectorization can be added by adding patterns.add<CopyVectorizationPattern>(ctx); to the VectorizeOp.

794

tensor.pad vectorization can be activated with the vectorize_padding = true flag on the transform.structured.vectorize

[guraypp]

This commit changes the followings:

Adds LinalgDialect as dependent dialect. Because tensor.pad generates linalg.init_tensor. 
Adds CopyVectorizationPattern pattern.
Deletes "CHECK: vector.transfer_write" in the test. They look redundant writes to me.

[guraypp]

fix indendation

[nicolasvasilache]

The description should be part of the commit message.

In D133699#3788935, @guraypp wrote:

This commit changes the followings:

Adds LinalgDialect as dependent dialect. Because tensor.pad generates linalg.init_tensor.

This needs to be better explained, adding dialect dependencies is not a trivial thing.
What is the dependence added to, why is it necessary now and wasn't before etc.

Adds CopyVectorizationPattern pattern.

This revision also adds the CopyVectorizationPattern to the list of patterns applied by VectorizeOp::applyToOne.

Deletes "CHECK: vector.transfer_write" in the test. They look redundant writes to me.

mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td

777

The op semantics start to become more complex and it is unclear with the current doc what this does.

Please make these 2 new attributes very loud and spell them as UnitAttr:$disable_multi_reduction_to_contract_patterns and UnitAttr:$disable_transfer_permutation_map_lowering_patterns.
Please document the options and mark them as "limiting the power of vectorization and are currently intended for testing purposes".

mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp

1163

Please put this in n anon namespace and document the fact that this only exists so we can call vectorize via a pattern inside of VectorizeOp::applyToOne.

mlir/test/Dialect/Linalg/vectorization.mlir

878

Please don't change the testing behavior in the same commit.
At best a followup commit could do this but I do not see value in removing those atm: this is what vectorization currently produces, if we need to improve this then let's change the algorithm in a followup commit.

[nicolasvasilache]

Please rephrase this commit message into something more meaningful..

We retire linalg's strategy vectorize patterns. Our goal is to use transform dialect instead of passes.

This revision retires the LinalgCodegenStrategy vectorization patterns and passes.
Context: https://discourse.llvm.org/t/psa-retire-linalg-filter-based-patterns/63785.

The retiring the pattern is done, however, the vectorization.mlir needs to be changed drastically. I changed some of it and put it here. If that's what we want, I'll replace them all and update the revision.

This is stale, please drop.

It depends on D133684

This is how one specifies dependent commits: https://llvm.org/docs/Phabricator.html#using-patch-summaries

[guraypp]

Addressed nicolasvasilache's comments

[guraypp]

did rebase

[nicolasvasilache]

Thanks! Let's land this and its parent.

[guraypp]

rebase

mlir/include/mlir/Dialect/Linalg/Passes.h

 
 /// Create a LinalgStrategyPeelPass.
 std::unique_ptr<OperationPass<func::FuncOp>> createLinalgStrategyPeelPass(
     StringRef opName = "",
     const linalg::LinalgPeelOptions &opt = linalg::LinalgPeelOptions(),
     const linalg::LinalgTransformationFilter &filter =
         linalg::LinalgTransformationFilter());
 
-/// Create a LinalgStrategyVectorizePass.
-std::unique_ptr<OperationPass<func::FuncOp>> createLinalgStrategyVectorizePass(
-    StringRef opName = "",
-    linalg::LinalgVectorizationOptions opt =
-        linalg::LinalgVectorizationOptions(),
-    const linalg::LinalgTransformationFilter &filter =
-        linalg::LinalgTransformationFilter(),
-    bool padVectorize = false);
-
 /// Create a LinalgStrategyLowerVectorsPass.
 std::unique_ptr<OperationPass<func::FuncOp>>
 createLinalgStrategyLowerVectorsPass(
     linalg::LinalgVectorLoweringOptions opt =
         linalg::LinalgVectorLoweringOptions(),
     const linalg::LinalgTransformationFilter &filter =
         linalg::LinalgTransformationFilter());
 

mlir/include/mlir/Dialect/Linalg/Passes.td

def LinalgStrategyPeelPass
   let options = [
     Option<"anchorFuncName", "anchor-func", "std::string", /*default=*/"",
       "Which func op is the anchor to latch on.">,
     Option<"anchorOpName", "anchor-op", "std::string", /*default=*/"",
       "Which linalg op within the func is the anchor to latch on.">,
   ];
 }
 
-def LinalgStrategyVectorizePass
-    : Pass<"linalg-strategy-vectorize-pass", "func::FuncOp"> {
-  let summary = "Configurable pass to apply pattern-based linalg vectorization.";
-  let constructor = "mlir::createLinalgStrategyVectorizePass()";
-  let dependentDialects = ["linalg::LinalgDialect"];
-  let options = [
-    Option<"anchorFuncName", "anchor-func", "std::string", /*default=*/"",
-      "Which func op is the anchor to latch on.">,
-    Option<"anchorOpName", "anchor-op", "std::string", /*default=*/"",
-      "Which linalg op within the func is the anchor to latch on.">,
-    Option<"vectorizePadding", "vectorize-padding", "bool", "false",
-      "Enable vectorization of padding ops.">,
-  ];
-}
-
 def LinalgStrategyLowerVectorsPass
     : Pass<"linalg-strategy-lower-vectors-pass", "func::FuncOp"> {
   let summary = "Configurable pass to lower vector operations.";
   let constructor = "mlir::createLinalgStrategyLowerVectorsPass()";
   let dependentDialects = ["linalg::LinalgDialect"];
   let options = [
     Option<"anchorFuncName", "anchor-func", "std::string", /*default=*/"",
       "Which func op is the anchor to latch on.">,

mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td

let description = [{
     
     This operation produces `definiteFailure` if vectorization fails for any
     reason.
     The operation always returns the handle to the target op that is expected 
     to be isolated from above.
   }];
 
   let arguments = (ins PDL_Operation:$target,
-                   DefaultValuedAttr<BoolAttr, "false">:$vectorize_padding);
+                   DefaultValuedAttr<BoolAttr, "false">:$vectorize_padding,                   
+                   DefaultValuedAttr<BoolAttr, "true">:$reduction_to_contract,                   
+                   DefaultValuedAttr<BoolAttr, "true">:$permutation_map);
   let results = (outs PDL_Operation:$transformed);
 
   let assemblyFormat = "$target attr-dict";
 
   let extraClassDeclaration = [{
     ::mlir::DiagnosedSilenceableFailure applyToOne(
         ::mlir::Operation *target, 
         ::llvm::SmallVectorImpl<::mlir::Operation *> &results, 
         ::mlir::transform::TransformState &state);
   }];
 }
 
 #endif // LINALG_TRANSFORM_OPS

mlir/include/mlir/Dialect/Linalg/Transforms/CodegenStrategy.h

void addToPassPipeline(OpPassManager &pm,
     pm.addPass(createLinalgStrategyPeelPass(opName, options, m));
   }
 
 private:
   std::string opName;
   linalg::LinalgPeelOptions options;
 };
 
-/// Represent one application of createLinalgStrategyVectorizePass.
-struct Vectorize : public Transformation {
-  explicit Vectorize(linalg::LinalgVectorizationOptions options,
-                     LinalgTransformationFilter::FilterFunction f = nullptr,
-                     bool padVectorize = false)
-      : Transformation(std::move(f)), options(options),
-        vectorizePadding(padVectorize) {}
-
-  Vectorize(StringRef name, linalg::LinalgVectorizationOptions options,
-            LinalgTransformationFilter::FilterFunction f = nullptr,
-            bool padVectorize = false)
-      : Transformation(std::move(f)), opName(name), options(options),
-        vectorizePadding(padVectorize) {}
-
-  void addToPassPipeline(OpPassManager &pm,
-                         LinalgTransformationFilter m) const override {
-    pm.addPass(createLinalgStrategyVectorizePass(opName, options, m,
-                                                 vectorizePadding));
-  }
-
-private:
-  std::string opName;
-  linalg::LinalgVectorizationOptions options;
-  bool vectorizePadding;
-};
-
 /// Represent one application of createLinalgStrategyLowerVectorsPass.
 struct VectorLowering : public Transformation {
   explicit VectorLowering(
       linalg::LinalgVectorLoweringOptions options,
       LinalgTransformationFilter::FilterFunction f = nullptr)
       : Transformation(std::move(f)), options(options) {}
 
   void addToPassPipeline(OpPassManager &pm,
pad(StringRef opName, const linalg::LinalgPaddingOptions &options,
     return *this;
   }
   /// Conditionally append a pattern to pad and hoist the operands of Op
   /// `opName` with padding `options`.
   CodegenStrategy &
   padIf(bool b, StringRef opName, linalg::LinalgPaddingOptions options,
         LinalgTransformationFilter::FilterFunction f = nullptr) {
     return b ? pad(opName, std::move(options), std::move(f)) : *this;
-  }  
+  }
   /// Append patterns to decompose convolutions.
   CodegenStrategy &
   decompose(const LinalgTransformationFilter::FilterFunction &f = nullptr) {
     transformationSequence.emplace_back(std::make_unique<Decompose>(f));
     return *this;
   }
   /// Conditionally append patterns to decompose convolutions.
   CodegenStrategy &
peel(StringRef opName, const LinalgPeelOptions &options,
     return *this;
   }
   /// Conditionally append a pattern to peel 'LinalgOpType'.
   CodegenStrategy &
   peelIf(bool b, StringRef opName, const LinalgPeelOptions &options,
          LinalgTransformationFilter::FilterFunction f = nullptr) {
     return b ? peel(opName, options, std::move(f)) : *this;
   }
-  /// Append a pattern to rewrite `LinalgOpType` as a vector operation.
-  CodegenStrategy &
-  vectorize(StringRef opName,
-            const LinalgTransformationFilter::FilterFunction &f = nullptr,
-            bool vectorizePadding = false) {
-    transformationSequence.emplace_back(std::make_unique<Vectorize>(
-        opName, linalg::LinalgVectorizationOptions(), f, vectorizePadding));
-    return *this;
-  }
-  /// Conditionally append a pattern to rewrite `LinalgOpType` as a vector
-  /// operation.
-  CodegenStrategy &
-  vectorizeIf(bool b, StringRef opName,
-              LinalgTransformationFilter::FilterFunction f = nullptr,
-              bool vectorizePadding = false) {
-    return b ? vectorize(opName, std::move(f), vectorizePadding) : *this;
-  }
   /// Append a pattern to lower all vector operations.
   CodegenStrategy &vectorLowering(LinalgVectorLoweringOptions options) {
     transformationSequence.emplace_back(
         std::make_unique<VectorLowering>(options));
     return *this;
   }
   /// Configure the post staged-patterns global enabling passes options.
   CodegenStrategy &

mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h

 ///
 /// Linalg vectorization patterns.
 ///
 /// Empty for now, used for SFINAE purposes only.
 struct LinalgVectorizationOptions {};
 
 /// `filter` controls LinalgTransformMarker matching and update when specified.
 /// See `vectorizeLinalgOp` for more details.
-struct LinalgVectorizationPattern : public OpInterfaceRewritePattern<LinalgOp> {
-  /// Construct a generic pattern applied to all LinalgOp that verify `filter`.
-  LinalgVectorizationPattern(
-      MLIRContext *context,
-      LinalgTransformationFilter f = LinalgTransformationFilter(),
-      LinalgVectorizationOptions options = LinalgVectorizationOptions(),
-      PatternBenefit benefit = 1);
-
-  /// Construct a pattern specifically applied to `opName`.
-  LinalgVectorizationPattern(
-      StringRef opName, MLIRContext *context,
-      LinalgVectorizationOptions options = LinalgVectorizationOptions(),
-      LinalgTransformationFilter f = LinalgTransformationFilter(),
-      PatternBenefit benefit = 1);
-
-  LogicalResult matchAndRewrite(LinalgOp linalgOp,
-                                PatternRewriter &rewriter) const override;
-
-private:
-  /// LinalgTransformMarker handles special attribute manipulations.
-  LinalgTransformationFilter filter;
-};
-
-/// `filter` controls LinalgTransformMarker matching and update when specified.
-/// See `vectorizeLinalgOp` for more details.
 struct CopyVectorizationPattern : public OpRewritePattern<memref::CopyOp> {
   using OpRewritePattern<memref::CopyOp>::OpRewritePattern;
 
   LogicalResult matchAndRewrite(memref::CopyOp copyOp,
                                 PatternRewriter &rewriter) const override;
 };
 
 /// Return vector::CombiningKind for the given op.

 template <>
 class VectorizationPatterns<> {
 public:
   static void insert(RewritePatternSet &patterns,
                      const LinalgVectorizationOptions &options,
                      const LinalgTransformationFilter &f) {}
 };
 
-template <typename OpTy, typename... OpTypes>
-class VectorizationPatterns<OpTy, OpTypes...> {
-public:
-  static void insert(RewritePatternSet &patterns,
-                     const LinalgVectorizationOptions &options,
-                     const LinalgTransformationFilter &f) {
-    patterns.add<LinalgVectorizationPattern>(OpTy::getOperationName(),
-                                             patterns.getContext(), options, f);
-    VectorizationPatterns<OpTypes...>::insert(patterns, options, f);
-  }
-};
-
 template <typename... OpTypes>
 class TilingPatterns;
 
 template <>
 class TilingPatterns<> {
 public:
   static void insert(RewritePatternSet &patterns,
                      const LinalgTilingOptions &options,

mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp

if (getMixedNumThreads().empty() == getMixedTileSizes().empty())
     return emitOpError("either num_threads or tile_sizes must be specified");
   return success();
 }
 
 //===----------------------------------------------------------------------===//
 // VectorizeOp
 //===----------------------------------------------------------------------===//
 
+struct VectorizationPattern : public RewritePattern {
+  explicit VectorizationPattern(MLIRContext *context)
+    : RewritePattern(MatchAnyOpTypeTag(), /*benefit=*/1, context) {}
+  LogicalResult matchAndRewrite(Operation *op, PatternRewriter &rewriter) const override {
+      LinalgOp linalgOp = dyn_cast<LinalgOp>(op);
+      if(!linalgOp) return failure();
+      return vectorize(rewriter, linalgOp);
+  }
+};
+
 DiagnosedSilenceableFailure
 transform::VectorizeOp::applyToOne(Operation *target,
                                    SmallVectorImpl<Operation *> &results,
                                    transform::TransformState &state) {
   if (!target->hasTrait<OpTrait::IsIsolatedFromAbove>()) {
     auto diag = this->emitOpError("requires isolated-from-above targets");
     diag.attachNote(target->getLoc()) << "non-isolated target";
     return DiagnosedSilenceableFailure::definiteFailure();
   }
 
   MLIRContext *ctx = getContext();
   RewritePatternSet patterns(ctx);
-  patterns.add<LinalgVectorizationPattern>(ctx);
+  patterns.add<VectorizationPattern>(ctx);
 
+  if(getPermutationMap())
     vector::populateVectorTransferPermutationMapLoweringPatterns(patterns);
+    
+  if(getReductionToContract())
     vector::populateVectorReductionToContractPatterns(patterns);
+
   patterns.add<linalg::LinalgCopyVTRForwardingPattern,
                linalg::LinalgCopyVTWForwardingPattern>(ctx,
                                                        /*benefit=*/2);
   vector::TransferReadOp::getCanonicalizationPatterns(patterns, ctx);
   vector::TransferWriteOp::getCanonicalizationPatterns(patterns, ctx);
+  
+  patterns.add<CopyVectorizationPattern>(ctx);
+  
   if (getVectorizePadding())
     linalg::populatePadOpVectorizationPatterns(patterns);
 
   if (failed(applyPatternsAndFoldGreedily(target, std::move(patterns))))
     return DiagnosedSilenceableFailure(reportUnknownTransformError(target));
 
   results.push_back(target);
   return DiagnosedSilenceableFailure(success());

 class LinalgTransformDialectExtension
     : public transform::TransformDialectExtension<
           LinalgTransformDialectExtension> {
 public:
   using Base::Base;
 
   void init() {
     declareDependentDialect<pdl::PDLDialect>();
-
+    declareDependentDialect<LinalgDialect>();
     declareGeneratedDialect<AffineDialect>();
     declareGeneratedDialect<arith::ArithmeticDialect>();
     declareGeneratedDialect<scf::SCFDialect>();
     declareGeneratedDialect<vector::VectorDialect>();
 
     registerTransformOps<
 #define GET_OP_LIST
 #include "mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp.inc"

mlir/lib/Dialect/Linalg/Transforms/LinalgStrategyPasses.cpp

 #include <utility>
 
 namespace mlir {
 #define GEN_PASS_DEF_LINALGSTRATEGYTILEANDFUSEPASS
 #define GEN_PASS_DEF_LINALGSTRATEGYTILEPASS
 #define GEN_PASS_DEF_LINALGSTRATEGYPADPASS
 #define GEN_PASS_DEF_LINALGSTRATEGYDECOMPOSEPASS
 #define GEN_PASS_DEF_LINALGSTRATEGYPEELPASS
-#define GEN_PASS_DEF_LINALGSTRATEGYVECTORIZEPASS
 #define GEN_PASS_DEF_LINALGSTRATEGYLOWERVECTORSPASS
 #define GEN_PASS_DEF_LINALGSTRATEGYREMOVEMARKERSPASS
 #include "mlir/Dialect/Linalg/Passes.h.inc"
 } // namespace mlir
 
 using namespace mlir;
 using namespace mlir::vector;
 using namespace linalg;
if (failed(
             applyPatternsAndFoldGreedily(funcOp, std::move(peelingPatterns))))
       return signalPassFailure();
   }
 
   LinalgPeelOptions options;
   LinalgTransformationFilter filter;
 };
 
-/// Configurable pass to apply pattern-based linalg vectorization.
-struct LinalgStrategyVectorizePass
-    : public impl::LinalgStrategyVectorizePassBase<
-          LinalgStrategyVectorizePass> {
-
-  LinalgStrategyVectorizePass() = default;
-
-  LinalgStrategyVectorizePass(StringRef opName, LinalgVectorizationOptions opt,
-                              LinalgTransformationFilter filt,
-                              bool padVectorize = false)
-      : options(opt), filter(std::move(filt)) {
-    this->anchorOpName.setValue(opName.str());
-    this->vectorizePadding.setValue(padVectorize);
-  }
-
-  void runOnOperation() override {
-    auto funcOp = getOperation();
-    if (!anchorFuncName.empty() && funcOp.getName() != anchorFuncName)
-      return;
-
-    RewritePatternSet vectorizationPatterns(funcOp.getContext());
-    if (!anchorOpName.empty()) {
-      vectorizationPatterns.add<LinalgVectorizationPattern>(
-          anchorOpName, funcOp.getContext(), options, filter);
-    } else {
-      vectorizationPatterns.add<LinalgVectorizationPattern>(funcOp.getContext(),
-                                                            filter, options);
-    }
-    vector::populateVectorTransferPermutationMapLoweringPatterns(
-        vectorizationPatterns);
-    vector::populateVectorReductionToContractPatterns(vectorizationPatterns);
-    vectorizationPatterns.add<linalg::LinalgCopyVTRForwardingPattern,
-                              linalg::LinalgCopyVTWForwardingPattern>(
-        funcOp.getContext(), /*benefit=*/2);
-    TransferReadOp::getCanonicalizationPatterns(vectorizationPatterns,
-                                                funcOp.getContext());
-    TransferWriteOp::getCanonicalizationPatterns(vectorizationPatterns,
-                                                 funcOp.getContext());
-    (void)applyPatternsAndFoldGreedily(funcOp,
-                                       std::move(vectorizationPatterns));
-
-    // Apply the pad tensor op vectorization separately to avoid running the
-    // GenericPadOpVectorizationPattern too early.
-    // TODO: Improve once we have better infrastructure to control pattern
-    // application.
-    if (vectorizePadding) {
-      RewritePatternSet patterns(funcOp.getContext());
-      linalg::populatePadOpVectorizationPatterns(patterns);
-      (void)applyPatternsAndFoldGreedily(funcOp, std::move(patterns));
-    }
-  }
-
-  LinalgVectorizationOptions options;
-  LinalgTransformationFilter filter;
-};
-
 /// Configurable pass to lower vector operations.
 struct LinalgStrategyLowerVectorsPass
     : public impl::LinalgStrategyLowerVectorsPassBase<
           LinalgStrategyLowerVectorsPass> {
 
   LinalgStrategyLowerVectorsPass(LinalgVectorLoweringOptions opt,
                                  LinalgTransformationFilter filt)
       : options(opt), filter(std::move(filt)) {}

 /// Create a LinalgStrategyPeelPass.
 std::unique_ptr<OperationPass<func::FuncOp>>
 mlir::createLinalgStrategyPeelPass(StringRef opName,
                                    const LinalgPeelOptions &opt,
                                    const LinalgTransformationFilter &filter) {
   return std::make_unique<LinalgStrategyPeelPass>(opName, opt, filter);
 }
 
-/// Create a LinalgStrategyVectorizePass.
-std::unique_ptr<OperationPass<func::FuncOp>>
-mlir::createLinalgStrategyVectorizePass(
-    StringRef opName, LinalgVectorizationOptions opt,
-    const LinalgTransformationFilter &filter, bool padVectorize) {
-  return std::make_unique<LinalgStrategyVectorizePass>(opName, opt, filter,
-                                                       padVectorize);
-}
-
 /// Create a LinalgStrategyLowerVectorsPass.
 std::unique_ptr<OperationPass<func::FuncOp>>
 mlir::createLinalgStrategyLowerVectorsPass(
     LinalgVectorLoweringOptions opt, const LinalgTransformationFilter &filter) {
   return std::make_unique<LinalgStrategyLowerVectorsPass>(opt, filter);
 }
 
 /// Create a LinalgStrategyRemoveMarkersPass.
 std::unique_ptr<OperationPass<func::FuncOp>>
 mlir::createLinalgStrategyRemoveMarkersPass() {
   return std::make_unique<LinalgStrategyRemoveMarkersPass>();
 }

mlir/lib/Dialect/Linalg/Transforms/Transforms.cpp

if (!options.loopsToPeelComputationFunction)
     return failure();
 
   SmallVector<scf::ForOp, 4> loopsToPeel;
   options.loopsToPeelComputationFunction(rewriter, linalgOp, loopsToPeel);
   peelLoops(rewriter, loopsToPeel);
   return success();
 }
 
-mlir::linalg::LinalgVectorizationPattern::LinalgVectorizationPattern(
-    MLIRContext *context, LinalgTransformationFilter f,
-    LinalgVectorizationOptions options, PatternBenefit benefit)
-    : OpInterfaceRewritePattern<LinalgOp>(context, benefit),
-      filter(std::move(f)) {}
-
-mlir::linalg::LinalgVectorizationPattern::LinalgVectorizationPattern(
-    StringRef opName, MLIRContext *context, LinalgVectorizationOptions options,
-    LinalgTransformationFilter f, PatternBenefit benefit)
-    : OpInterfaceRewritePattern<LinalgOp>(context, benefit),
-      filter(f.addOpNameFilter(opName)) {}
-
-LogicalResult mlir::linalg::LinalgVectorizationPattern::matchAndRewrite(
-    LinalgOp linalgOp, PatternRewriter &rewriter) const {
-  if (failed(filter.checkAndNotify(rewriter, linalgOp)))
-    return failure();
-  return vectorize(rewriter, linalgOp);
-}
-
 LogicalResult mlir::linalg::CopyVectorizationPattern::matchAndRewrite(
     memref::CopyOp copyOp, PatternRewriter &rewriter) const {
   return vectorizeCopy(rewriter, copyOp);
 }
 
 LogicalResult mlir::linalg::applyStagedPatterns(
     Operation *op, ArrayRef<FrozenRewritePatternSet> stage1Patterns,
     const FrozenRewritePatternSet &stage2Patterns,

mlir/test/Dialect/Linalg/vectorization.mlir

-// RUN: mlir-opt %s -test-linalg-transform-patterns=test-linalg-to-vector-patterns -split-input-file | FileCheck %s
+// RUN: mlir-opt %s -test-transform-dialect-interpreter -split-input-file | FileCheck %s
 
 // -----
 
 // CHECK-LABEL: contraction_dot
 func.func @contraction_dot(%A: memref<1584xf32>, %B: memref<1584xf32>, %C: memref<f32>) {
 
 // CHECK: arith.mulf %{{.*}}, %{{.*}} : vector<1584xf32>
 // CHECK: vector.multi_reduction <add>, %{{.*}}, {{.*}} [0] : vector<1584xf32> to f32
   linalg.dot ins(%A, %B: memref<1584xf32>, memref<1584xf32>)
             outs(%C: memref<f32>)
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.dot"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {reduction_to_contract = false, permutation_map = false }
+  }
+}
+
 // -----
 
 // CHECK-LABEL: contraction_matvec
 func.func @contraction_matvec(%A: memref<1584x1584xf32>, %B: memref<1584xf32>, %C: memref<1584xf32>) {
 
 // CHECK: arith.mulf %{{.*}}, %{{.*}} : vector<1584x1584xf32>
 // CHECK: vector.multi_reduction <add>, %{{.*}}, {{.*}} [1] : vector<1584x1584xf32> to vector<1584xf32>
   linalg.matvec ins(%A, %B: memref<1584x1584xf32>, memref<1584xf32>)
             outs(%C: memref<1584xf32>)
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.matvec"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {reduction_to_contract = false, permutation_map = false }
+  }
+}
+
 // -----
 
 // CHECK-LABEL: contraction_matmul
 func.func @contraction_matmul(%A: memref<1584x1584xf32>, %B: memref<1584x1584xf32>, %C: memref<1584x1584xf32>) {
 // CHECK: arith.mulf %{{.*}}, %{{.*}} : vector<1584x1584x1584xf32>
 // CHECK: vector.multi_reduction <add>, %{{.*}}, {{.*}} [2] : vector<1584x1584x1584xf32> to vector<1584x1584xf32>
   linalg.matmul ins(%A, %B: memref<1584x1584xf32>, memref<1584x1584xf32>)
             outs(%C: memref<1584x1584xf32>)
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.matmul"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {reduction_to_contract = false, permutation_map = false }
+  }
+}
+
 // -----
 
 // CHECK-LABEL: contraction_batch_matmul
 func.func @contraction_batch_matmul(%A: memref<1584x1584x1584xf32>, %B: memref<1584x1584x1584xf32>, %C: memref<1584x1584x1584xf32>) {
 // CHECK: arith.mulf %{{.*}}, %{{.*}} : vector<1584x1584x1584x1584xf32>
 // CHECK: vector.multi_reduction <add>, %{{.*}}, {{.*}} [3] : vector<1584x1584x1584x1584xf32> to vector<1584x1584x1584xf32>
   linalg.batch_matmul
     ins(%A, %B: memref<1584x1584x1584xf32>, memref<1584x1584x1584xf32>)
    outs(%C: memref<1584x1584x1584xf32>)
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.batch_matmul"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {reduction_to_contract = false, permutation_map = false }
+  }
+}
+
 // -----
 
 #matmul_trait = {
   args_in = 2,
   args_out = 1,
   indexing_maps = [
     affine_map<(m, n, k) -> (m, k)>,
     affine_map<(m, n, k) -> (k, n)>,
outs(%C : memref<8x32xf32>) {
     ^bb(%a: f32, %b: f32, %c: f32) :
       %d = arith.mulf %a, %b: f32
       %e = arith.addf %c, %d: f32
       linalg.yield %e : f32
   }
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {reduction_to_contract = false, permutation_map = false }
+  }
+}
+
 // -----
 
 #matmul_transpose_out_trait = {
   args_in = 2,
   args_out = 1,
   indexing_maps = [
     affine_map<(m, n, k) -> (m, k)>,
     affine_map<(m, n, k) -> (k, n)>,
outs(%C : memref<32x8xf32>) {
     ^bb(%a: f32, %b: f32, %c: f32) :
       %d = arith.mulf %a, %b: f32
       %e = arith.addf %c, %d: f32
       linalg.yield %e : f32
   }
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {reduction_to_contract = false, permutation_map = false }
+  }
+}
+
 // -----
 
 #map0 = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 #map1 = affine_map<(d0, d1, d2) -> (d1, d0, d2)>
 // CHECK: #[[MAP:.+]] = affine_map<(d0, d1, d2) -> (d1, d0, d2)>
 // CHECK: func @generic_interchanged_transpose
 func.func @generic_interchanged_transpose(%arg0: tensor<12x128x32xf32>) -> tensor<128x12x32xf32> {
   // CHECK: %[[IN:.+]] = vector.transfer_read
   // CHECK: vector.transfer_write %[[IN]], {{.+}} permutation_map = #[[MAP]]
   %0 = linalg.init_tensor [128, 12, 32] : tensor<128x12x32xf32>
   %1 = linalg.generic {indexing_maps = [#map0, #map1],
                        iterator_types = ["parallel", "parallel", "parallel"]}
     ins(%arg0 : tensor<12x128x32xf32>)
     outs(%0 : tensor<128x12x32xf32>) {
   ^bb0(%arg1: f32, %arg2: f32):
     linalg.yield %arg1 : f32
   } -> tensor<128x12x32xf32>
   return %1 : tensor<128x12x32xf32>
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {reduction_to_contract = false, permutation_map = false }
+  }
+}
+
 // -----
 
 #matmul_trait = {
   args_in = 2,
   args_out = 1,
   indexing_maps = [
     affine_map<(m, n, k) -> (m, k)>,
     affine_map<(m, n, k) -> (k, n)>,
outs(%C : memref<8x32xi32>) {
     ^bb(%a: i32, %b: i32, %c: i32) :
       %d = arith.muli %a, %b: i32
       %e = arith.addi %c, %d: i32
       linalg.yield %e : i32
   }
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {reduction_to_contract = false, permutation_map = false }
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @vectorization_test_2
 func.func @vectorization_test_2(%A: memref<8x16xf32>, %B: memref<16x32xf32>,
                          %C: memref<8x32xf32>) {
   //       CHECK: arith.mulf %{{.*}}, %{{.*}} : vector<8x32x16xf32>
   //       CHECK: vector.multi_reduction <add>, %{{.*}}, {{.*}} [2] : vector<8x32x16xf32> to vector<8x32xf32>
   linalg.matmul
     ins(%A, %B: memref<8x16xf32>, memref<16x32xf32>)
    outs(%C: memref<8x32xf32>)
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.matmul"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {reduction_to_contract = false, permutation_map = false }
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @test_vectorize_scalar_input
 func.func @test_vectorize_scalar_input(%A : memref<8x16xf32>, %arg0 : f32) {
   //       CHECK: %[[V:.*]] = vector.broadcast {{.*}} : f32 to vector<8x16xf32>
   //       CHECK: vector.transfer_write %[[V]], {{.*}} : vector<8x16xf32>, memref<8x16xf32>
   linalg.generic {
     indexing_maps = [affine_map<(m, n) -> ()>, affine_map<(m, n) -> (m, n)>],
     iterator_types = ["parallel", "parallel"]}
    ins(%arg0 : f32)
   outs(%A: memref<8x16xf32>) {
     ^bb(%0: f32, %1: f32) :
       linalg.yield %0 : f32
   }
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1 
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @test_do_not_vectorize_unsupported_element_types
 func.func @test_do_not_vectorize_unsupported_element_types(%A : memref<8x16xcomplex<f32>>, %arg0 : complex<f32>) {
   // CHECK-NOT: vector.broadcast
   // CHECK-NOT: vector.transfer_write
   linalg.generic {
     indexing_maps = [affine_map<(m, n) -> ()>, affine_map<(m, n) -> (m, n)>],
     iterator_types = ["parallel", "parallel"]}
    ins(%arg0 : complex<f32>)
   outs(%A: memref<8x16xcomplex<f32>>) {
     ^bb(%0: complex<f32>, %1: complex<f32>) :
       linalg.yield %0 : complex<f32>
   }
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1 
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @test_vectorize_fill
 func.func @test_vectorize_fill(%A : memref<8x16xf32>, %arg0 : f32) {
   //       CHECK: %[[V:.*]] = vector.broadcast {{.*}} : f32 to vector<8x16xf32>
   //       CHECK: vector.transfer_write %[[V]], {{.*}} : vector<8x16xf32>, memref<8x16xf32>
   linalg.fill ins(%arg0 : f32) outs(%A : memref<8x16xf32>)
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.fill"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1 
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @test_vectorize_fill
 func.func @test_vectorize_fill_scalar(%A : memref<f32>, %arg0 : f32) {
   // CHECK-SAME: (%[[M:.*]]: memref<f32>, %[[val:.*]]: f32)
   //      CHECK:   %[[VEC:.*]] = vector.broadcast %[[val]] : f32 to vector<f32>
   //      CHECK:   vector.transfer_write %[[VEC]], %[[M]][] : vector<f32>, memref<f32>
   linalg.fill ins(%arg0 : f32) outs(%A : memref<f32>)
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.fill"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1 
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @test_vectorize_copy
 func.func @test_vectorize_copy(%A : memref<8x16xf32>, %B : memref<8x16xf32>) {
   //       CHECK: %[[V:.*]] = vector.transfer_read {{.*}} : memref<8x16xf32>, vector<8x16xf32>
   //       CHECK: vector.transfer_write %[[V]], {{.*}} : vector<8x16xf32>, memref<8x16xf32>
   memref.copy %A, %B :  memref<8x16xf32> to memref<8x16xf32>
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["memref.copy"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1 
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @test_vectorize_copy_scalar
 func.func @test_vectorize_copy_scalar(%A : memref<f32>, %B : memref<f32>) {
   //  CHECK-SAME: (%[[A:.*]]: memref<f32>, %[[B:.*]]: memref<f32>)
   //       CHECK:   %[[V:.*]] = vector.transfer_read %[[A]][]{{.*}} : memref<f32>, vector<f32>
   //       CHECK:   %[[val:.*]] = vector.extractelement %[[V]][] : vector<f32>
   //       CHECK:   %[[VV:.*]] = vector.broadcast %[[val]] : f32 to vector<f32>
   //       CHECK:   vector.transfer_write %[[VV]], %[[B]][] : vector<f32>, memref<f32>
   memref.copy %A, %B :  memref<f32> to memref<f32>
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["memref.copy"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1 
+  }
+}
 // -----
 
 // CHECK-LABEL: func @test_vectorize_trailing_index
   //  CHECK-SAME: (%[[ARG0:.*]]: memref<1x2x4x8xindex>)
 func.func @test_vectorize_trailing_index(%arg0: memref<1x2x4x8xindex>) {
   //   CHECK-DAG:   %[[CST0:.*]] = arith.constant dense<[0, 1, 2, 3, 4, 5, 6, 7]> : vector<8xindex>
   //   CHECK-DAG:   %[[C0:.*]] = arith.constant 0 : index
   linalg.generic {
     indexing_maps = [
       affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>],
     iterator_types = ["parallel", "parallel", "parallel", "parallel"]}
   outs(%arg0: memref<1x2x4x8xindex>) {
   ^bb0(%arg1: index):
   //       CHECK:   %[[BCST:.*]] = vector.broadcast %[[CST0]] : vector<8xindex> to vector<1x2x4x8xindex>
   //       CHECK:   vector.transfer_write %[[BCST]], %[[ARG0]][%[[C0]], %[[C0]], %[[C0]], %[[C0]]] {{.*}} : vector<1x2x4x8xindex>, memref<1x2x4x8xindex>
     %0 = linalg.index 3 : index
     linalg.yield %0 : index
   }
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1 
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @test_vectorize_inner_index
   //  CHECK-SAME: (%[[ARG0:.*]]: memref<1x2x4x8xindex>)
 func.func @test_vectorize_inner_index(%arg0: memref<1x2x4x8xindex>) {
   //   CHECK-DAG:   %[[CST0:.*]] = arith.constant dense<[0, 1]> : vector<2xindex>
   //   CHECK-DAG:   %[[C0:.*]] = arith.constant 0 : index
   linalg.generic {
     indexing_maps = [
       affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>],
     iterator_types = ["parallel", "parallel", "parallel", "parallel"]}
   outs(%arg0: memref<1x2x4x8xindex>) {
   ^bb0(%arg1: index):
   //       CHECK:   %[[BCST:.*]] = vector.broadcast %[[CST0]] : vector<2xindex> to vector<1x8x4x2xindex>
   //       CHECK:   %[[TRAN:.*]] = vector.transpose %[[BCST]], [0, 3, 2, 1] : vector<1x8x4x2xindex> to vector<1x2x4x8xindex>
   //       CHECK:   vector.transfer_write %[[TRAN]], %[[ARG0]][%[[C0]], %[[C0]], %[[C0]], %[[C0]]] {{.*}} : vector<1x2x4x8xindex>, memref<1x2x4x8xindex>
     %0 = linalg.index 1 : index
     linalg.yield %0 : index
   }
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1 
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @generic_vectorize
   //  CHECK-SAME: (%[[ARG0:.*]]: memref<4x256xf32>, %[[ARG1:.*]]: memref<4x256xf32>,
   //  CHECK-SAME:  %[[ARG2:.*]]: memref<256xf32>, %[[ARG3:.*]]: f32)
 func.func @generic_vectorize(%arg0: memref<4x256xf32>,
                         %arg1: memref<4x256xf32>,
                         %arg2: memref<256xf32>, %i: f32) {
func.func @generic_vectorize(%arg0: memref<4x256xf32>,
   //       CHECK:   vector.transfer_write %[[SUB]], %[[ARG0]][%[[C0]], %[[C0]]] {{.*}} : vector<4x256xf32>, memref<4x256xf32>
   //       CHECK:   vector.transfer_write %[[TAN]], %[[ARG0]][%[[C0]], %[[C0]]] {{.*}} : vector<4x256xf32>, memref<4x256xf32>
     linalg.yield %6, %8, %c1_f32, %9, %10, %11, %12, %13, %14, %15 : f32, f32,
       f32, f32, f32, f32, f32, f32, f32, f32
   }
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {permutation_map = false }
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @generic_vectorize_tensor
 //  CHECK-SAME: (%[[ARG0:.*]]: tensor<4x256xf32>, %[[ARG1:.*]]: tensor<4x256xf32>,
 //  CHECK-SAME:  %[[ARG2:.*]]: tensor<256xf32>, %[[ARG3:.*]]: f32)
 func.func @generic_vectorize_tensor(%arg0: tensor<4x256xf32>,
   %arg1: tensor<4x256xf32>, %arg2: tensor<256xf32>,
   %i: f32) -> (tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>,
//       CHECK:   %[[R9:.*]] = vector.transfer_write %[[TAN]], %[[ARG0]][%[[C0]], %[[C0]]] {{.*}} : vector<4x256xf32>, tensor<4x256xf32>
     tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>)
   //       CHECK:   return %[[R0]], %[[R1]], %[[R2]], %[[R3]], %[[R4]], %[[R5]], %[[R6]], %[[R7]], %[[R8]], %[[R9]] : tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>
   return %r#0, %r#1, %r#2, %r#3, %r#4, %r#5, %r#6, %r#7, %r#8, %r#9:
     tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>,
     tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>,
     tensor<4x256xf32>, tensor<4x256xf32>, tensor<4x256xf32>
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  { permutation_map = false }
+  }
+}
+
 // -----
 
 // CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1) -> (d0, 0, 0, d1)>
 // CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0) -> (d0, 0, 0, 0)>
 // CHECK-DAG: #[[$MAP2:.*]] = affine_map<(d0) -> (0, 0, d0, 0)>
 // CHECK-DAG: #[[$MAP3:.*]] = affine_map<(d0, d1) -> (d1, 0, d0, 0)>
 //     CHECK: func @generic_vectorize_broadcast_transpose
 // CHECK-DAG:   %[[C0:.*]] = arith.constant 0 : index
^bb0(%arg0: f32, %arg1: f32, %arg2: f32, %arg3: f32, %arg4: f32):
     %s = arith.subf %arg0, %arg1 : f32
     %a = arith.addf %arg2, %s : f32
     %b = arith.addf %arg3, %a : f32
     linalg.yield %b : f32
   }
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {permutation_map = false }
+  }
+}
+
 // -----
 
 // Test different input maps.
 #matmul_trait = {
   indexing_maps = [
     affine_map<(d0, d1, d2, d3) -> (d1, d0)>,
     affine_map<(d0, d1, d2, d3) -> (d3, d1)>,
     affine_map<(d0, d1, d2, d3) -> (d3, d1, d0, d2)>,
outs(%D : memref<7x14x8x16xf32>) {
     ^bb(%a: f32, %b: f32, %c: f32, %d: f32) :
       %e = arith.addf %a, %b: f32
       %f = arith.addf %e, %c: f32
       linalg.yield %f : f32
   }
   return
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {permutation_map = false }
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @matmul_tensors
 //  CHECK-SAME: (%[[ARG0:.*]]: tensor<8x4xf32>, %[[ARG1:.*]]: tensor<4x12xf32>,
 //  CHECK-SAME:  %[[ARG2:.*]]: tensor<8x12xf32>) -> tensor<8x12xf32>
 func.func @matmul_tensors(
   %arg0: tensor<8x4xf32>, %arg1: tensor<4x12xf32>, %arg2: tensor<8x12xf32>)
     -> tensor<8x12xf32> {
func.func @matmul_tensors(
   //       CHECK:   %[[W:.*]] = vector.transfer_write %[[R]], %[[ARG2]][%[[C0]], %[[C0]]] {in_bounds = [true, true]} : vector<8x12xf32>, tensor<8x12xf32>
   %0 = linalg.matmul  ins(%arg0, %arg1: tensor<8x4xf32>, tensor<4x12xf32>)
                      outs(%arg2: tensor<8x12xf32>)
     -> tensor<8x12xf32>
   //       CHECK:   return %[[W]] : tensor<8x12xf32>
   return %0 : tensor<8x12xf32>
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.matmul"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {reduction_to_contract = false, permutation_map = false }
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @pad_static(
 //  CHECK-SAME:                  %[[ARG0:.*]]: tensor<2x?x2xf32>, %[[PAD:.*]]: f32
 //   CHECK-NOT:   tensor.pad
 //   CHECK-DAG:   %[[C0:.*]] = arith.constant 0 : index
 //   CHECK-DAG:   %[[C2:.*]] = arith.constant 2 : index
 //   CHECK-DAG:   %[[INIT:.*]] = linalg.init_tensor [2, 3, 4] : tensor<2x3x4xf32>
 //   CHECK-DAG:   %[[VEC:.*]] = vector.broadcast %[[PAD]] : f32 to vector<2x3x4xf32>
 //       CHECK:   %[[FILL:.*]] = vector.transfer_write %[[VEC]], %[[INIT]]{{.*}} : vector<2x3x4xf32>, tensor<2x3x4xf32>
 //       CHECK:   %[[READ:.*]] = vector.transfer_read %[[ARG0]][%[[C0]], %[[C0]], %[[C0]]], %[[PAD]] {in_bounds = [true, false, true]} : tensor<2x?x2xf32>, vector<2x3x2xf32>
 //       CHECK:   %[[RESULT:.*]] = vector.transfer_write %[[READ]], %[[FILL]][%[[C0]], %[[C0]], %[[C2]]] {in_bounds = [true, true, true]} : vector<2x3x2xf32>, tensor<2x3x4xf32>
 //       CHECK:   return %[[RESULT]]
 func.func @pad_static(%arg0: tensor<2x?x2xf32>, %pad_value: f32) -> tensor<2x3x4xf32> {
   %0 = tensor.pad %arg0 low[0, 0, 2] high[0, 1, 0] {
     ^bb0(%arg1: index, %arg2: index, %arg3: index):
       tensor.yield %pad_value : f32
     } : tensor<2x?x2xf32> to tensor<2x3x4xf32>
   return %0 : tensor<2x3x4xf32>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["tensor.pad"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1 { vectorize_padding = true }
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @pad_static_source(
 //  CHECK-SAME:                  %[[ARG0:.*]]: tensor<2x5x2xf32>, %[[PAD:.*]]: f32
 //   CHECK-NOT:   tensor.pad
 //   CHECK-DAG:   %[[C0:.*]] = arith.constant 0 : index
 //   CHECK-DAG:   %[[C2:.*]] = arith.constant 2 : index
 //       CHECK:   %[[INIT:.*]] = linalg.init_tensor [2, 6, 4] : tensor<2x6x4xf32>
 //       CHECK:   %[[VEC:.*]] =  vector.broadcast %[[PAD]] : f32 to vector<2x6x4xf32>
 //       CHECK:   %[[FILL:.*]] = vector.transfer_write %[[VEC]], %[[INIT]][%[[C0]], %[[C0]], %[[C0]]] {in_bounds = [true, true, true]} : vector<2x6x4xf32>, tensor<2x6x4xf32>
 //       CHECK:   %[[READ:.*]] = vector.transfer_read %[[ARG0]][%[[C0]], %[[C0]], %[[C0]]], %{{.*}} {in_bounds = [true, true, true]} : tensor<2x5x2xf32>, vector<2x5x2xf32>
 //       CHECK:   %[[WRITE:.*]] = vector.transfer_write %[[READ]], %[[FILL]][%[[C0]], %[[C0]], %[[C2]]] {in_bounds = [true, true, true]} : vector<2x5x2xf32>, tensor<2x6x4xf32>
 //       CHECK:   return %[[WRITE]]
 func.func @pad_static_source(%arg0: tensor<2x5x2xf32>, %pad_value: f32) -> tensor<2x6x4xf32> {
   %0 = tensor.pad %arg0 low[0, 0, 2] high[0, 1, 0] {
     ^bb0(%arg1: index, %arg2: index, %arg3: index):
       tensor.yield %pad_value : f32
     } : tensor<2x5x2xf32> to tensor<2x6x4xf32>
   return %0 : tensor<2x6x4xf32>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["tensor.pad"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  { vectorize_padding = true }
+  }
+}
+
+
 // -----
 
 // CHECK-LABEL: func @pad_static_dynamic(
 //  CHECK-SAME:                          %[[SRC:.*]]: tensor<1x2x2x?xf32>, %[[LOW:.*]]: index, %[[HIGH:.*]]: index
 //   CHECK-NOT:   tensor.pad
 //   CHECK-DAG:   %[[C2:.*]] = arith.constant 2 : index
 //   CHECK-DAG:   %[[C3:.*]] = arith.constant 3 : index
 //   CHECK-DAG:   %[[C5:.*]] = arith.constant 5 : index
func.func @pad_static_dynamic(%arg0: tensor<1x2x2x?xf32>, %low: index, %high: index,
                   %pad_value: f32) -> tensor<6x?x?x?xf32> {
   %0 = tensor.pad %arg0 low[2, %low, 3, 3] high[3, 3, %high, 2] {
     ^bb0(%arg1: index, %arg2: index, %arg3: index, %arg4: index):
       tensor.yield %pad_value : f32
     } : tensor<1x2x2x?xf32> to tensor<6x?x?x?xf32>
   return %0 : tensor<6x?x?x?xf32>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["tensor.pad"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  { vectorize_padding = true }
+  }
+}
+
+
 // -----
 
 // CHECK-LABEL: func @pad_and_transfer_read
 //  CHECK-SAME:     %[[ARG0:.*]]: tensor<5x6xf32>
 //   CHECK-NOT:   tensor.pad
 //   CHECK-DAG:   %[[C0:.*]] = arith.constant 0 : index
 //   CHECK-DAG:   %[[C5:.*]] = arith.constant 5.0
-//       CHECK:   %[[RESULT:.*]] = vector.transfer_read %[[ARG0]][%[[C0]], %[[C0]]], %[[C5]] : tensor<5x6xf32>, vector<7x9xf32>
-//       CHECK:   return %[[RESULT]]
+//  CHECK:   %[[RESULT:.*]] = vector.transfer_read %[[ARG0]][%[[C0]], %[[C0]]], %[[C5]] : tensor<5x6xf32>, vector<7x9xf32>
+//  CHECK:   return %[[RESULT]]
 func.func @pad_and_transfer_read(%arg0: tensor<5x6xf32>) -> vector<7x9xf32> {
   %c0 = arith.constant 0 : index
   %c5 = arith.constant 5.0 : f32
   %c6 = arith.constant 6.0 : f32
   %0 = tensor.pad %arg0 low[0, 0] high[5, 7] {
     ^bb0(%arg1: index, %arg2: index):
       tensor.yield %c5 : f32
   } : tensor<5x6xf32> to tensor<10x13xf32>
   %1 = vector.transfer_read %0[%c0, %c0], %c6
       : tensor<10x13xf32>, vector<7x9xf32>
   return %1 : vector<7x9xf32>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["tensor.pad"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1 { vectorize_padding = true }
+  }
+}
+
 // -----
 
 func.func private @make_vector() -> vector<7x9xf32>
 
 // CHECK-LABEL: func @pad_and_transfer_write_static
 //  CHECK-SAME:     %[[ARG0:.*]]: tensor<5x6xf32>
 //   CHECK-NOT:   tensor.pad
 //       CHECK:   %[[C0:.*]] = arith.constant 0 : index
func.func @pad_and_transfer_write_static(
   } : tensor<5x6xf32> to tensor<10x13xf32>
   %1 = call @make_vector() : () -> vector<7x9xf32>
   %2 = vector.transfer_write %1, %0[%c0, %c0]
       : vector<7x9xf32>, tensor<10x13xf32>
   %3 = tensor.extract_slice %2[0, 0] [5, 6] [1, 1] : tensor<10x13xf32> to tensor<5x6xf32>
   return %3 : tensor<5x6xf32>
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["tensor.pad"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4  { vectorize_padding = true }
+  }
+}
+
+
 // -----
 
 func.func private @make_vector() -> vector<7x9xf32>
 
 // CHECK-LABEL: func @pad_and_transfer_write_dynamic_static
 //  CHECK-SAME:     %[[ARG0:.*]]: tensor<?x?xf32>, %[[SIZE:.*]]: index, %[[PADDING:.*]]: index
 //   CHECK-NOT:   tensor.pad
 //       CHECK:   %[[C0:.*]] = arith.constant 0 : index
func.func @pad_and_transfer_write_dynamic_static(
   } : tensor<?x6xf32> to tensor<?x13xf32>
   %1 = call @make_vector() : () -> vector<7x9xf32>
   %2 = vector.transfer_write %1, %0[%c0, %c0]
       : vector<7x9xf32>, tensor<?x13xf32>
   %3 = tensor.extract_slice %2[0, 0] [%size, 6] [1, 1] : tensor<?x13xf32> to tensor<?x6xf32>
   return %3 : tensor<?x6xf32>
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["tensor.pad"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4 { vectorize_padding = true }
+  }
+}
+
+
 // -----
 
 func.func private @make_vector() -> tensor<12x13xf32>
 
 // CHECK-LABEL: func @pad_and_insert_slice_source
 //  CHECK-SAME:     %[[ARG0:.*]]: tensor<5x6xf32>
 //   CHECK-NOT:   tensor.pad
 //   CHECK-DAG:   %[[C0:.*]] = arith.constant 0 : index
%0 = tensor.pad %arg0 low[0, 0] high[2, 3] {
     ^bb0(%arg2: index, %arg3: index):
       tensor.yield %c5 : f32
   } : tensor<5x6xf32> to tensor<7x9xf32>
   %1 = call @make_vector() : () -> tensor<12x13xf32>
   %r = tensor.insert_slice %0 into %1[0, 0][7, 9][1, 1] : tensor<7x9xf32> into tensor<12x13xf32>
   return %r : tensor<12x13xf32>
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["tensor.pad"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4  { vectorize_padding = true }
+  }
+}
+
+
 // -----
 
 func.func private @make_vector() -> tensor<12x13xf32>
 
 // CHECK-LABEL: func @pad_and_insert_slice_dest
 // Check the insert slice is not rewritten if the padded result is used by the destination operand.
 //       CHECK:   %[[T1:.*]] = call @make_vector() : () -> tensor<12x13xf32>
 //       CHECK:   = tensor.insert_slice %[[T1]] into
 func.func @pad_and_insert_slice_dest(
     %arg0: tensor<1x5x6xf32>) -> tensor<1x12x13xf32> {
   %c5 = arith.constant 5.0 : f32
   %0 = tensor.pad %arg0 low[0, 0, 0] high[0, 7, 7] {
     ^bb0(%arg2: index, %arg3: index, %arg4: index):
       tensor.yield %c5 : f32
   } : tensor<1x5x6xf32> to tensor<1x12x13xf32>
   %1 = call @make_vector() : () -> tensor<12x13xf32>
   %r = tensor.insert_slice %1 into %0[0, 0, 0][1, 12, 13][1, 1, 1] : tensor<12x13xf32> into tensor<1x12x13xf32>
   return %r : tensor<1x12x13xf32>
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["tensor.pad"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4 
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @pad_tensor_non_const_pad_value
 //  CHECK-SAME:     %[[ARG0:.*]]: tensor<5x6xf32>
 //   CHECK-NOT:   tensor.pad
 //   CHECK-DAG:   %[[C0:.*]] = arith.constant 0 : index
 //   CHECK-DAG:   %[[C3:.*]] = arith.constant 3 : index
 //   CHECK-DAG:   %[[C4:.*]] = arith.constant 4 : index
^bb0(%arg1: index, %arg2: index):
       %f1 = arith.sitofp %i1 : i32 to f32
       %f2 = arith.sitofp %i2 : i32 to f32
       %m = arith.mulf %f1, %f2 : f32
       tensor.yield %m : f32
   } : tensor<5x6xf32> to tensor<12x13xf32>
   return %0 : tensor<12x13xf32>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["tensor.pad"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4  { vectorize_padding = true }
+  }
+}
+
 // -----
 
 // CHECK-LABEL: func @sum_exp
 func.func @sum_exp(%input: tensor<4x16x8xf32>, %output: tensor<4x16xf32>)
   -> tensor<4x16xf32>
 {
   // CHECK: vector.transfer_read {{.*}} : tensor<4x16x8xf32>, vector<4x16x8xf32>
   // CHECK: vector.transfer_read {{.*}} {in_bounds = [true, true]} : tensor<4x16xf32>, vector<4x16xf32>
%0 = linalg.generic {
     ^bb0(%arg0: f32, %arg1: f32):
       %1 = math.exp %arg0 : f32
       %2 = arith.addf %1, %arg1 : f32
       linalg.yield %2 : f32
     } -> tensor<4x16xf32>
   return %0 : tensor<4x16xf32>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4 
+  }
+}
+
 // -----
 
 // CHECK-DAG: #[[$M1:.*]] =  affine_map<(d0, d1) -> (d1, d0, 0, 0)>
 // CHECK-DAG: #[[$M2:.*]] =  affine_map<(d0, d1) -> (0, 0, d1, d0)>
 // CHECK-DAG: #[[$M3:.*]] =  affine_map<(d0, d1) -> (d1, d0)>
 
 // CHECK-LABEL: func @sum_exp_2
 func.func @sum_exp_2(%input: tensor<3x2xf32>, %input_2: tensor<5x4xf32>, %output: tensor<5x2xf32>)
^bb0(%arg0: f32, %arg1: f32, %arg2: f32):
       %2 = math.exp %arg1 : f32
       %3 = arith.addf %1, %2 : f32
       %4 = arith.addf %3, %arg2 : f32
       linalg.yield %4 : f32
     } -> tensor<5x2xf32>
   return %0 : tensor<5x2xf32>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4  {reduction_to_contract = false, permutation_map = false }
+  }
+}
+
 // -----
 
 // CHECK-LABEL:   func @red_max_2d(
 func.func @red_max_2d(%arg0: tensor<4x4xf32>) -> tensor<4xf32> {
   // CHECK: %[[CMINF:.+]] = arith.constant dense<-3.402820e+38> : vector<4xf32>
   // CHECK: linalg.init_tensor [4] : tensor<4xf32>
-  // CHECK: vector.transfer_write {{.*}} : vector<4xf32>, tensor<4xf32>
   // CHECK: vector.multi_reduction <maxf>, {{.*}}, %[[CMINF]] [1] : vector<4x4xf32> to vector<4xf32>
   // CHECK: vector.transfer_write {{.*}} : vector<4xf32>, tensor<4xf32>
   %ident = arith.constant -3.40282e+38 : f32
   %init = linalg.init_tensor [4] : tensor<4xf32>
   %fill = linalg.fill ins(%ident : f32) outs(%init : tensor<4xf32>) -> tensor<4xf32>
   %red = linalg.generic {indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>,
                                           affine_map<(d0, d1) -> (d0)>],
                          iterator_types = ["parallel", "reduction"]}
                          ins(%arg0 : tensor<4x4xf32>) outs(%fill : tensor<4xf32>) {
   ^bb0(%in0: f32, %out0: f32):
     %max = arith.maxf %in0, %out0 : f32
     linalg.yield %max : f32
   } -> tensor<4xf32>
   return %red : tensor<4xf32>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4 { vectorize_padding = true }
+  }
+}
+
 // -----
 
 // CHECK-LABEL:   func @red_min_2d(
 func.func @red_min_2d(%arg0: tensor<4x4xf32>) -> tensor<4xf32> {
   // CHECK: %[[CMAXF:.+]] = arith.constant dense<3.402820e+38> : vector<4xf32>
   // CHECK: linalg.init_tensor [4] : tensor<4xf32>
-  // CHECK: vector.transfer_write {{.*}} : vector<4xf32>, tensor<4xf32>
   // CHECK: vector.transfer_read {{.*}} : tensor<4x4xf32>, vector<4x4xf32>
   // CHECK: vector.multi_reduction <minf>, {{.*}}, %[[CMAXF]] [1] : vector<4x4xf32> to vector<4xf32>
   // CHECK: vector.transfer_write {{.*}} : vector<4xf32>, tensor<4xf32>
   %maxf32 = arith.constant 3.40282e+38 : f32
   %init = linalg.init_tensor [4] : tensor<4xf32>
   %fill = linalg.fill ins(%maxf32 : f32) outs(%init : tensor<4xf32>) -> tensor<4xf32>
   %red = linalg.generic {indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>,
                                           affine_map<(d0, d1) -> (d0)>],
                          iterator_types = ["parallel", "reduction"]}
                          ins(%arg0 : tensor<4x4xf32>) outs(%fill : tensor<4xf32>) {
   ^bb0(%in0: f32, %out0: f32):
     %min = arith.minf %out0, %in0 : f32
     linalg.yield %min : f32
   } -> tensor<4xf32>
   return %red : tensor<4xf32>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4 
+  }
+}
+
 // -----
 
 // CHECK-LABEL:   func @red_mul_2d(
 func.func @red_mul_2d(%arg0: tensor<4x4xf32>) -> tensor<4xf32> {
   // CHECK: linalg.init_tensor [4] : tensor<4xf32>
-  // CHECK: vector.transfer_write {{.*}} : vector<4xf32>, tensor<4xf32>
   // CHECK: vector.transfer_read {{.*}} : tensor<4x4xf32>, vector<4x4xf32>
   // CHECK: vector.multi_reduction <mul>, {{.*}}, {{.*}} [1] : vector<4x4xf32> to vector<4xf32>
   // CHECK: vector.transfer_write {{.*}} : vector<4xf32>, tensor<4xf32>
   %ident = arith.constant 1.0 : f32
   %init = linalg.init_tensor [4] : tensor<4xf32>
   %fill = linalg.fill ins(%ident : f32) outs(%init : tensor<4xf32>) -> tensor<4xf32>
   %red = linalg.generic {indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>,
                                           affine_map<(d0, d1) -> (d0)>],
                          iterator_types = ["parallel", "reduction"]}
                          ins(%arg0 : tensor<4x4xf32>) outs(%fill : tensor<4xf32>) {
   ^bb0(%in0: f32, %out0: f32):
     %mul = arith.mulf %in0, %out0 : f32
     linalg.yield %mul : f32
   } -> tensor<4xf32>
   return %red : tensor<4xf32>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4 
+  }
+}
+
 // -----
 
 // CHECK-LABEL:   func @red_or_2d(
 func.func @red_or_2d(%arg0: tensor<4x4xi1>) -> tensor<4xi1> {
   // CHECK: linalg.init_tensor [4] : tensor<4xi1>
-  // CHECK: vector.transfer_write {{.*}} : vector<4xi1>, tensor<4xi1>
   // CHECK: vector.transfer_read {{.*}} : tensor<4x4xi1>, vector<4x4xi1>
   // CHECK: vector.multi_reduction <or>, {{.*}}, {{.*}} [1] : vector<4x4xi1> to vector<4xi1>
   // CHECK: vector.transfer_write {{.*}} : vector<4xi1>, tensor<4xi1>
   %ident = arith.constant false
   %init = linalg.init_tensor [4] : tensor<4xi1>
   %fill = linalg.fill ins(%ident : i1) outs(%init : tensor<4xi1>) -> tensor<4xi1>
   %red = linalg.generic {indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>,
                                           affine_map<(d0, d1) -> (d0)>],
                          iterator_types = ["parallel", "reduction"]}
                          ins(%arg0 : tensor<4x4xi1>) outs(%fill : tensor<4xi1>) {
   ^bb0(%in0: i1, %out0: i1):
     %or = arith.ori %in0, %out0 : i1
     linalg.yield %or : i1
   } -> tensor<4xi1>
   return %red : tensor<4xi1>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4 
+  }
+}
+
 // -----
 
 // CHECK-LABEL:   func @red_and_2d(
 func.func @red_and_2d(%arg0: tensor<4x4xi1>) -> tensor<4xi1> {
   // CHECK: linalg.init_tensor [4] : tensor<4xi1>
-  // CHECK: vector.transfer_write {{.*}} : vector<4xi1>, tensor<4xi1>
   // CHECK: vector.transfer_read {{.*}} : tensor<4x4xi1>, vector<4x4xi1>
   // CHECK: vector.multi_reduction <and>, {{.*}}, {{.*}} [1] : vector<4x4xi1> to vector<4xi1>
   // CHECK: vector.transfer_write {{.*}} : vector<4xi1>, tensor<4xi1>
   %ident = arith.constant true
   %init = linalg.init_tensor [4] : tensor<4xi1>
   %fill = linalg.fill ins(%ident : i1) outs(%init : tensor<4xi1>) -> tensor<4xi1>
   %red = linalg.generic {indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>,
                                           affine_map<(d0, d1) -> (d0)>],
                          iterator_types = ["parallel", "reduction"]}
                          ins(%arg0 : tensor<4x4xi1>) outs(%fill : tensor<4xi1>) {
   ^bb0(%in0: i1, %out0: i1):
     %and = arith.andi %in0, %out0 : i1
     linalg.yield %and : i1
   } -> tensor<4xi1>
   return %red : tensor<4xi1>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4 
+  }
+}
+
 // -----
 
 // CHECK-LABEL:   func @red_xor_2d(
 func.func @red_xor_2d(%arg0: tensor<4x4xi1>) -> tensor<4xi1> {
   // CHECK: linalg.init_tensor [4] : tensor<4xi1>
-  // CHECK: vector.transfer_write {{.*}} : vector<4xi1>, tensor<4xi1>
   // CHECK: vector.transfer_read {{.*}} : tensor<4x4xi1>, vector<4x4xi1>
   // CHECK: vector.multi_reduction <xor>, {{.*}}, {{.*}} [1] : vector<4x4xi1> to vector<4xi1>
   // CHECK: vector.transfer_write {{.*}} : vector<4xi1>, tensor<4xi1>
   %ident = arith.constant false
   %init = linalg.init_tensor [4] : tensor<4xi1>
   %fill = linalg.fill ins(%ident : i1) outs(%init : tensor<4xi1>) -> tensor<4xi1>
   %red = linalg.generic {indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>,
                                           affine_map<(d0, d1) -> (d0)>],
                          iterator_types = ["parallel", "reduction"]}
                          ins(%arg0 : tensor<4x4xi1>) outs(%fill : tensor<4xi1>) {
   ^bb0(%in0: i1, %out0: i1):
     %xor = arith.xori %in0, %out0 : i1
     linalg.yield %xor : i1
   } -> tensor<4xi1>
   return %red : tensor<4xi1>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4 
+  }
+}
+
 // -----
 
 // CHECK-DAG: #[[$M5:.*]] = affine_map<(d0, d1) -> (d0, 0)>
 
 // CHECK-LABEL:   func @explicit_broadcast(
 func.func @explicit_broadcast(%arg0: tensor<4x4xf32>, %arg1: tensor<4x1xf32>) -> tensor<4x4xf32> {
   // CHECK: vector.transfer_read {{.*}} {in_bounds = [true, true]} : tensor<4x4xf32>, vector<4x4xf32>
   // CHECK: vector.transfer_read {{.*}} {in_bounds = [true, true], permutation_map = #[[$M5]]} : tensor<4x1xf32>, vector<4x4xf32>
func.func @explicit_broadcast(%arg0: tensor<4x4xf32>, %arg1: tensor<4x1xf32>) -> tensor<4x4xf32> {
    outs(%fill : tensor<4x4xf32>) {
     ^bb0(%arg7: f32, %arg8: f32, %arg9: f32):
       %40 = arith.subf %arg7, %arg8 : f32
       linalg.yield %40 : f32
     } -> tensor<4x4xf32>
   return %red : tensor<4x4xf32>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %3 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4 
+  }
+}
+
 // -----
 
 // CHECK-DAG: #[[$M6:.*]] = affine_map<(d0, d1) -> (d0, 0)>
 
 // CHECK-LABEL:   func @fused_broadcast_red_2d
 func.func @fused_broadcast_red_2d(%arg0: tensor<4x4xf32>, %arg1: tensor<4x1xf32>) -> tensor<4xf32> {
   // CHECK: vector.transfer_read {{.*}} {in_bounds = [true, true]} : tensor<4x4xf32>, vector<4x4xf32>
   // CHECK: vector.transfer_read {{.*}} {in_bounds = [true, true], permutation_map = #[[$M6]]} : tensor<4x1xf32>, vector<4x4xf32>
^bb0(%arg7: f32, %arg8: f32, %arg9: f32):
       %40 = arith.subf %arg7, %arg8 : f32
       %41 = math.exp %40 : f32
       %42 = arith.addf %41, %arg9 : f32
       linalg.yield %42 : f32
     } -> tensor<4xf32>
   return %red : tensor<4xf32>
 }
 
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.fill"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1 
+    
+    %3 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %4 = get_closest_isolated_parent %3
+    %5 = transform.structured.vectorize %4 
+  }
+}
+
 // -----
 
 //  CHECK-LABEL: func @reduce_1d(
 //   CHECK-SAME:   %[[A:.*]]: tensor<32xf32>
 func.func @reduce_1d(%arg0: tensor<32xf32>) -> tensor<f32> {
   //  CHECK-DAG: %[[vF0:.*]] = arith.constant dense<0.000000e+00> : vector<f32>
   //  CHECK-DAG: %[[F0:.*]] = arith.constant 0.000000e+00 : f32
   //  CHECK-DAG: %[[C0:.*]] = arith.constant 0 : index
   %f0 = arith.constant 0.000000e+00 : f32
 
   //      CHECK: %[[init:.*]] = linalg.init_tensor [] : tensor<f32>
   %0 = linalg.init_tensor [] : tensor<f32>
 
-  //      CHECK: %[[f:.*]] = vector.transfer_write %[[vF0]], %[[init]][]
-  // CHECK-SAME:   : vector<f32>, tensor<f32>
   %1 = linalg.fill ins(%f0 : f32) outs(%0 : tensor<f32>) -> tensor<f32>
   //      CHECK: %[[r:.*]] = vector.transfer_read %[[A]][%[[C0]]]
   // CHECK-SAME:   : tensor<32xf32>, vector<32xf32>
   //      CHECK: %[[f0:.*]] = vector.extractelement %[[vF0]][] : vector<f32>
   //      CHECK: %[[red:.*]] = vector.multi_reduction <add>, %[[r]], %[[f0]] [0]
   // CHECK-SAME:   : vector<32xf32> to f32
   //      CHECK: %[[red_v1:.*]] = vector.broadcast %[[red]] : f32 to vector<f32>
-  //      CHECK: %[[res:.*]] = vector.transfer_write %[[red_v1]], %[[f]][]
-  // CHECK-SAME:   : vector<f32>, tensor<f32>
+  //      CHECK: %[[res:.*]] = vector.transfer_write %[[red_v1]], %[[init]][]
+  //      CHECK:   vector<f32>, tensor<f32>
   %2 = linalg.generic {
          indexing_maps = [affine_map<(d0) -> (d0)>,
                           affine_map<(d0) -> ()>],
          iterator_types = ["reduction"]}
          ins(%arg0 : tensor<32xf32>)
          outs(%1 : tensor<f32>) {
     ^bb0(%a: f32, %b: f32):
       %3 = arith.addf %a, %b : f32
       linalg.yield %3 : f32
     } -> tensor<f32>
 
   return %2 : tensor<f32>
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1 
+  }
+}
+
 
 // -----
 
 // This test checks that vectorization does not occur when an input indexing map
 // is not a projected permutation. In the future, this can be converted to a 
 // positive test when support is added.
 
 // CHECK-LABEL:   func @not_projected_permutation
 func.func @not_projected_permutation(%arg0: tensor<8x8xf32>) -> tensor<6x6x3x3xf32> {
   %c0 = arith.constant 0.0 : f32
   %init = linalg.init_tensor [6, 6, 3, 3] : tensor<6x6x3x3xf32>
   %fill = linalg.fill ins(%c0 : f32) outs(%init : tensor<6x6x3x3xf32>) -> tensor<6x6x3x3xf32>
   // CHECK: linalg.generic
   %result = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2, d3) -> (d0 + d2, d1 + d3)>,
                                              affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>],
    iterator_types = ["parallel", "parallel", "parallel", "parallel"]}
    ins(%arg0 : tensor<8x8xf32>)
    outs(%fill : tensor<6x6x3x3xf32>) {
     ^bb0(%arg7: f32, %arg9: f32):
       linalg.yield %arg7 : f32
     } -> tensor<6x6x3x3xf32>
   return %result : tensor<6x6x3x3xf32>
 }
 
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1 
+  }
+}
+
 // -----
 
 // Check vectorization can handle cases where outputs are a mix of reduced and non-reduced values.
 func.func @mixed_parallel_reduced_results(%arg0 : tensor<2x4x8xf32>,
     %arg1 : tensor<2x4xf32>, %arg2 : tensor<2x4x8xf32>, %arg3 : tensor<2x4xf32>) ->
     (tensor<2x4x8xf32>, tensor<2x4xf32>) {
   %0:2 = linalg.generic {
       indexing_maps = [affine_map<(d0, d1, d2) -> (d0, d1, d2)>, affine_map<(d0, d1, d2) -> (d0, d1)>,

 //  CHECK-SAME:     %[[ARG3:[a-zA-Z0-9]+]]: tensor<2x4xf32>
 //   CHECK-DAG:   %[[V0:.+]] = vector.transfer_read %[[ARG0]]
 //   CHECK-DAG:   %[[V1:.+]] = vector.transfer_read %[[ARG1]]
 //   CHECK-DAG:   %[[V2:.+]] = vector.transfer_read %[[ARG3]]
 //   CHECK-DAG:   %[[MUL:.+]] = arith.mulf %[[V0]], %[[V1]]
 //   CHECK-DAG:   %[[ADD:.+]] = vector.multi_reduction <add>, %[[MUL]], %[[V2]]
 //   CHECK-DAG:   vector.transfer_write %[[MUL]], %[[ARG2]]
 //   CHECK-DAG:   vector.transfer_write %[[ADD]], %[[ARG3]]
+
+transform.with_pdl_patterns {
+^bb0(%arg0: !pdl.operation):
+  transform.sequence %arg0 failures(propagate) {
+  ^bb1(%arg1: !pdl.operation):
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1
+    %1 = get_closest_isolated_parent %0
+    %2 = transform.structured.vectorize %1  {reduction_to_contract = false, permutation_map = false }
+  }
+}

mlir/test/lib/Dialect/Linalg/TestLinalgTransforms.cpp

static void applyPatterns(func::FuncOp funcOp) {
   //===--------------------------------------------------------------------===//
   // Linalg distribution patterns.
   //===--------------------------------------------------------------------===//
   LinalgLoopDistributionOptions distributionOptions;
 
   //===--------------------------------------------------------------------===//
   // Linalg to vector contraction patterns.
   //===--------------------------------------------------------------------===//
-  patterns.add<LinalgVectorizationPattern>(
-      ctx, LinalgTransformationFilter(StringAttr::get(ctx, "VECTORIZE"))
-               .addOpFilter<MatmulOp, FillOp, GenericOp>());
   patterns.add<CopyVectorizationPattern>(ctx);
 
   (void)applyPatternsAndFoldGreedily(funcOp, std::move(patterns));
 
   // Drop the marker.
   funcOp.walk([](LinalgOp op) {
     op->removeAttr(LinalgTransforms::kLinalgTransformMarker);
   });

 
 static void applyVectorTransferForwardingPatterns(func::FuncOp funcOp) {
   RewritePatternSet forwardPattern(funcOp.getContext());
   forwardPattern.add<LinalgCopyVTRForwardingPattern>(funcOp.getContext());
   forwardPattern.add<LinalgCopyVTWForwardingPattern>(funcOp.getContext());
   (void)applyPatternsAndFoldGreedily(funcOp, std::move(forwardPattern));
 }
 
 static void applyLinalgToVectorPatterns(func::FuncOp funcOp) {
   RewritePatternSet patterns(funcOp.getContext());
-  auto *ctx = funcOp.getContext();
-  patterns.add<LinalgVectorizationPattern>(
-      ctx, LinalgTransformationFilter()
-               .addOpFilter<ContractionOpInterface, FillOp, GenericOp>());
+  auto *ctx = funcOp.getContext();  
   patterns.add<CopyVectorizationPattern>(ctx);
   populatePadOpVectorizationPatterns(patterns);
   populateConvolutionVectorizationPatterns(patterns);
   (void)applyPatternsAndFoldGreedily(funcOp, std::move(patterns));
 }
 
 static void applyPadTensorToGenericPatterns(func::FuncOp funcOp) {
   RewritePatternSet patterns(funcOp.getContext());