[Mlir-commits] [mlir] andrzej/tensor pack e2e vec (PR #117533)

Andrzej Warzyński llvmlistbot at llvm.org
Mon Nov 25 01:46:25 PST 2024


https://github.com/banach-space created https://github.com/llvm/llvm-project/pull/117533

- **[mlir][linalg] Extract `GeneralizePadOpPattern` into a standalone transformation**
- **[mlir][linalg][nfc] Update "pack-dynamic-inner-tile.mlir"**


>From 96eb4d3991d066dce0600549b7b463108a7edb47 Mon Sep 17 00:00:00 2001
From: Andrzej Warzynski <andrzej.warzynski at arm.com>
Date: Fri, 22 Nov 2024 14:00:32 +0000
Subject: [PATCH 1/2] [mlir][linalg] Extract `GeneralizePadOpPattern` into a
 standalone transformation

Currently, `GeneralizePadOpPattern` is grouped under
`populatePadOpVectorizationPatterns`. However, as noted in #111349, this
transformation "decomposes" rather than "vectorizes" `tensor.pad`. As
such, it functions as:
  * a vectorization _pre-processing_ transformation, not
  * a vectorization transformation itself.

To clarify its purpose, this PR turns `GeneralizePadOpPattern` into a
standalone transformation by:
  * introducing a dedicated `populateDecomposePadPatterns` method,
  * adding a `apply_patterns.linalg.decompose_pad` Transform Dialect Op, and
  * removing it from `populatePadOpVectorizationPatterns`.

In addition, to better reflect its role, it is renamed as
"decomposition" rather then "generalization". That's to better reflect
its role. This is in line with the recent renaming of similar ops, i.e.
tensor.pack/tensor.unpack Ops in #116439.
---
 .../Linalg/TransformOps/LinalgTransformOps.td        | 11 +++++++++++
 .../mlir/Dialect/Linalg/Transforms/Transforms.h      |  8 ++++++--
 .../lib/Conversion/TensorToLinalg/TensorToLinalg.cpp |  4 +++-
 .../Linalg/TransformOps/LinalgTransformOps.cpp       | 11 ++++++++++-
 mlir/lib/Dialect/Linalg/Transforms/Transforms.cpp    | 10 +++++++---
 mlir/lib/Dialect/Linalg/Transforms/Vectorization.cpp |  6 ------
 ...ize-pad-tensor.mlir => decompose-pad-tensor.mlir} |  2 +-
 .../Dialect/Linalg/vectorization-pad-patterns.mlir   |  6 ++++++
 .../test/lib/Dialect/Linalg/TestLinalgTransforms.cpp | 12 ++++++------
 9 files changed, 50 insertions(+), 20 deletions(-)
 rename mlir/test/Dialect/Linalg/{generalize-pad-tensor.mlir => decompose-pad-tensor.mlir} (98%)

diff --git a/mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td b/mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td
index e3084530bd11b5..dc10f3a1c58ae3 100644
--- a/mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td
+++ b/mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td
@@ -52,6 +52,17 @@ def ApplyDecomposeTensorPackUnpackPatternsOp
   let assemblyFormat = "attr-dict";
 }
 
+def ApplyDecomposeTensorPadPatternsOp
+    : Op<Transform_Dialect, "apply_patterns.linalg.decompose_pad",
+         [DeclareOpInterfaceMethods<PatternDescriptorOpInterface>]> {
+  let description = [{
+    Collect patterns to decompose tensor.pad into e.g. tensor::EmptyOp,
+    linalg::FillOp and tensor::InsertSliceOp.
+  }];
+
+  let assemblyFormat = "attr-dict";
+}
+
 def ApplyFoldUnitExtentDimsViaReshapesPatternsOp : Op<Transform_Dialect,
     "apply_patterns.linalg.fold_unit_extent_dims_via_reshapes",
     [DeclareOpInterfaceMethods<PatternDescriptorOpInterface>]> {
diff --git a/mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h b/mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h
index 51967f83fee377..3c160d55a38e75 100644
--- a/mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h
+++ b/mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h
@@ -1503,8 +1503,8 @@ using OptimizeCopyFn =
 
 /// Rewrite a tensor::PadOp into a sequence of EmptyOp, FillOp and
 /// InsertSliceOp. For now, only constant padding values are supported.
-struct GeneralizePadOpPattern : public OpRewritePattern<tensor::PadOp> {
-  GeneralizePadOpPattern(MLIRContext *context, PatternBenefit benefit = 1)
+struct DecomposePadOpPattern : public OpRewritePattern<tensor::PadOp> {
+  DecomposePadOpPattern(MLIRContext *context, PatternBenefit benefit = 1)
       : OpRewritePattern<tensor::PadOp>(context, benefit) {}
   LogicalResult matchAndRewrite(tensor::PadOp padOp,
                                 PatternRewriter &rewriter) const override;
@@ -1688,6 +1688,10 @@ void populateDecomposeConvolutionPatterns(RewritePatternSet &patterns,
 /// outer dims to be unit.
 void populateDecomposePackUnpackPatterns(RewritePatternSet &patterns);
 
+/// Populates patterns to decompose tensor.pad into e.g.
+/// tensor.empty, linalg.fill, tensor.insert_slice.
+void populateDecomposePadPatterns(RewritePatternSet &patterns);
+
 /// Populates patterns to transform linalg.conv_2d_xxx operations into
 /// linalg.generic (for img2col packing) and linalg.matmul.
 /// \see rewriteInIm2Col for more details.
diff --git a/mlir/lib/Conversion/TensorToLinalg/TensorToLinalg.cpp b/mlir/lib/Conversion/TensorToLinalg/TensorToLinalg.cpp
index 5bb79d4bc84e2b..b0ca0ca13d0624 100644
--- a/mlir/lib/Conversion/TensorToLinalg/TensorToLinalg.cpp
+++ b/mlir/lib/Conversion/TensorToLinalg/TensorToLinalg.cpp
@@ -25,5 +25,7 @@ using namespace mlir;
 //===----------------------------------------------------------------------===//
 
 void mlir::populateTensorToLinalgPatterns(RewritePatternSet &patterns) {
-  patterns.add<mlir::linalg::GeneralizePadOpPattern>(patterns.getContext());
+  // TODO: Add the remaining patterns, e.g. to decompose Pack/Unpack Ops.
+  // Alternatively, delete this file.
+  patterns.add<mlir::linalg::DecomposePadOpPattern>(patterns.getContext());
 }
diff --git a/mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp b/mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp
index ada80deacfdbfe..e08be7d2ebd6ae 100644
--- a/mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp
+++ b/mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp
@@ -234,6 +234,11 @@ void transform::ApplyDecomposeTensorPackUnpackPatternsOp::populatePatterns(
   linalg::populateDecomposePackUnpackPatterns(patterns);
 }
 
+void transform::ApplyDecomposeTensorPadPatternsOp::populatePatterns(
+    RewritePatternSet &patterns) {
+  linalg::populateDecomposePadPatterns(patterns);
+}
+
 void transform::ApplyFoldUnitExtentDimsViaReshapesPatternsOp::populatePatterns(
     RewritePatternSet &patterns) {
   linalg::ControlDropUnitDims options;
@@ -3491,8 +3496,12 @@ transform::VectorizeChildrenAndApplyPatternsOp::applyToOne(
   // Add misc. vectorization patterns (e.g. for tensor.insert_slice)
   linalg::populateInsertSliceVectorizationPatterns(patterns);
 
-  if (getVectorizePadding())
+  if (getVectorizePadding()) {
     linalg::populatePadOpVectorizationPatterns(patterns);
+    // This creates an alternative path for lowering tensor.pad - by
+    // decomposing it into e.g. linalg.fill.
+    linalg::populateDecomposePadPatterns(patterns);
+  }
   vector::populateVectorStepLoweringPatterns(patterns);
 
   TrackingListener listener(state, *this);
diff --git a/mlir/lib/Dialect/Linalg/Transforms/Transforms.cpp b/mlir/lib/Dialect/Linalg/Transforms/Transforms.cpp
index d92543d7264625..c3e176299317ef 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/Transforms.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/Transforms.cpp
@@ -921,7 +921,7 @@ LogicalResult mlir::linalg::CopyVectorizationPattern::matchAndRewrite(
 
 /// Filling `dest` using FillOp constant padding value if possible.
 /// Otherwise, generate a tensor::GenerateOp.
-Value GeneralizePadOpPattern::createFillOrGenerateOp(
+Value DecomposePadOpPattern::createFillOrGenerateOp(
     RewriterBase &rewriter, tensor::PadOp padOp, Value dest,
     const SmallVector<Value> &dynSizes) const {
   auto padValue = padOp.getConstantPaddingValue();
@@ -938,8 +938,8 @@ Value GeneralizePadOpPattern::createFillOrGenerateOp(
 }
 
 LogicalResult
-GeneralizePadOpPattern::matchAndRewrite(tensor::PadOp padOp,
-                                        PatternRewriter &rewriter) const {
+DecomposePadOpPattern::matchAndRewrite(tensor::PadOp padOp,
+                                       PatternRewriter &rewriter) const {
   // Given an OpFoldResult, return an index-typed value.
   auto getIdxValue = [&](OpFoldResult ofr) {
     if (auto val = llvm::dyn_cast_if_present<Value>(ofr))
@@ -1623,3 +1623,7 @@ void linalg::populateDecomposePackUnpackPatterns(RewritePatternSet &patterns) {
   // TODO: Add and test patterns for tensor.unpack
   patterns.add<DecomposeOuterUnitDimsPackOpPattern>(patterns.getContext());
 }
+
+void linalg::populateDecomposePadPatterns(RewritePatternSet &patterns) {
+  patterns.add<DecomposePadOpPattern>(patterns.getContext());
+}
diff --git a/mlir/lib/Dialect/Linalg/Transforms/Vectorization.cpp b/mlir/lib/Dialect/Linalg/Transforms/Vectorization.cpp
index 23b46a2ee55f8d..06bb6c0fb1cac9 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/Vectorization.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/Vectorization.cpp
@@ -2770,12 +2770,6 @@ void mlir::linalg::populateInsertSliceVectorizationPatterns(
 
 void mlir::linalg::populatePadOpVectorizationPatterns(
     RewritePatternSet &patterns, PatternBenefit baseBenefit) {
-  // TODO: The following pattern implements "decomposition" and
-  // optional "vectorization". Seperate "decomposition" into a sepereate
-  // pre-processing pattern group.
-  patterns.add<GeneralizePadOpPattern>(patterns.getContext(), baseBenefit);
-
-  // Try these specialized patterns first before resorting to the generic one.
   patterns.add<PadOpVectorizationWithTransferReadPattern,
                PadOpVectorizationWithTransferWritePattern,
                PadOpVectorizationWithInsertSlicePattern>(
diff --git a/mlir/test/Dialect/Linalg/generalize-pad-tensor.mlir b/mlir/test/Dialect/Linalg/decompose-pad-tensor.mlir
similarity index 98%
rename from mlir/test/Dialect/Linalg/generalize-pad-tensor.mlir
rename to mlir/test/Dialect/Linalg/decompose-pad-tensor.mlir
index 2beab31b613d54..184361dfb30dfd 100644
--- a/mlir/test/Dialect/Linalg/generalize-pad-tensor.mlir
+++ b/mlir/test/Dialect/Linalg/decompose-pad-tensor.mlir
@@ -1,4 +1,4 @@
-// RUN: mlir-opt -split-input-file --test-linalg-transform-patterns="test-generalize-pad-tensor"  %s | FileCheck %s
+// RUN: mlir-opt -split-input-file --test-linalg-transform-patterns="test-decompose-pad-tensor"  %s | FileCheck %s
 
 // CHECK-LABEL:   func @generalize_pad_tensor_static_shape(
 // CHECK-SAME:                                             %[[IN:.*]]: tensor<1x28x28x1xf32>) -> tensor<1x32x32x1xf32> {
diff --git a/mlir/test/Dialect/Linalg/vectorization-pad-patterns.mlir b/mlir/test/Dialect/Linalg/vectorization-pad-patterns.mlir
index 640de85cc5f12e..41e480648177f5 100644
--- a/mlir/test/Dialect/Linalg/vectorization-pad-patterns.mlir
+++ b/mlir/test/Dialect/Linalg/vectorization-pad-patterns.mlir
@@ -202,6 +202,8 @@ module attributes {transform.with_named_sequence} {
     %func_op = transform.structured.match ops{["func.func"]} in %arg1 : (!transform.any_op) -> !transform.op<"func.func">
 
     transform.apply_patterns to %func_op {
+      // TODO: Split into two tests, one for each pattern
+      transform.apply_patterns.linalg.decompose_pad
       transform.apply_patterns.linalg.pad_vectorization
     } : !transform.op<"func.func">
     transform.yield
@@ -236,6 +238,8 @@ module attributes {transform.with_named_sequence} {
     %func_op = transform.structured.match ops{["func.func"]} in %arg1 : (!transform.any_op) -> !transform.op<"func.func">
 
     transform.apply_patterns to %func_op {
+      // TODO: Split into two tests, one for each pattern
+      transform.apply_patterns.linalg.decompose_pad
       transform.apply_patterns.linalg.pad_vectorization
     } : !transform.op<"func.func">
     transform.yield
@@ -270,6 +274,8 @@ module attributes {transform.with_named_sequence} {
     %func_op = transform.structured.match ops{["func.func"]} in %arg1 : (!transform.any_op) -> !transform.op<"func.func">
 
     transform.apply_patterns to %func_op {
+      // TODO: Split into two tests, one for each pattern
+      transform.apply_patterns.linalg.decompose_pad
       transform.apply_patterns.linalg.pad_vectorization
     } : !transform.op<"func.func">
     transform.yield
diff --git a/mlir/test/lib/Dialect/Linalg/TestLinalgTransforms.cpp b/mlir/test/lib/Dialect/Linalg/TestLinalgTransforms.cpp
index c65e68eaf31f09..25aec75c3c14ad 100644
--- a/mlir/test/lib/Dialect/Linalg/TestLinalgTransforms.cpp
+++ b/mlir/test/lib/Dialect/Linalg/TestLinalgTransforms.cpp
@@ -70,8 +70,8 @@ struct TestLinalgTransforms
       llvm::cl::desc("Test a set of patterns that rewrite a linalg contraction "
                      "in vector.contract form"),
       llvm::cl::init(false)};
-  Option<bool> testGeneralizePadTensor{
-      *this, "test-generalize-pad-tensor",
+  Option<bool> testDecomposePadTensor{
+      *this, "test-decompose-pad-tensor",
       llvm::cl::desc("Test transform pad tensor by copying with generic ops"),
       llvm::cl::init(false)};
   Option<bool> testDecomposeTensorPackOp{
@@ -166,9 +166,9 @@ static void applyLinalgToVectorPatterns(func::FuncOp funcOp) {
   (void)applyPatternsAndFoldGreedily(funcOp, std::move(patterns));
 }
 
-static void applyGeneralizePadTensorPatterns(func::FuncOp funcOp) {
+static void applyDecomposePadPatterns(func::FuncOp funcOp) {
   RewritePatternSet patterns(funcOp.getContext());
-  patterns.add<GeneralizePadOpPattern>(funcOp.getContext());
+  patterns.add<DecomposePadOpPattern>(funcOp.getContext());
   (void)applyPatternsAndFoldGreedily(funcOp, std::move(patterns));
 }
 
@@ -235,8 +235,8 @@ void TestLinalgTransforms::runOnOperation() {
     return applyVectorTransferForwardingPatterns(getOperation());
   if (testGenericToVectorPattern)
     return applyLinalgToVectorPatterns(getOperation());
-  if (testGeneralizePadTensor)
-    return applyGeneralizePadTensorPatterns(getOperation());
+  if (testDecomposePadTensor)
+    return applyDecomposePadPatterns(getOperation());
   if (testDecomposeTensorPackOp)
     return applyDecomposeTensorPackPatterns(getOperation());
   if (testDecomposeTensorUnPackOp)

>From d0e7294e7674ba0391e05d298a0a4c6c51da400d Mon Sep 17 00:00:00 2001
From: Andrzej Warzynski <andrzej.warzynski at arm.com>
Date: Mon, 25 Nov 2024 09:37:56 +0000
Subject: [PATCH 2/2] [mlir][linalg][nfc] Update "pack-dynamic-inner-tile.mlir"
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

[mlir][linalg][nfc] Update pack-dynamic-inner-tile.mlir

Builds on:
  * #117329: Extract GeneralizePadOpPattern into a standalone transformation.
  * #116373: Update pack-dynamic-inner-tile.mlir.

This update adds vectorization to the "pack-dynamic-inner-tile.mlir"
pipeline.

The pipeline first decomposes `tensor.pack` into `tensor.pad` and then
into `linalg.fill` (#117329). Next, `linalg.fill` is vectorized, with
vector sizes matching the inner tile sizes of the original
`tensor.pack`.

••NOTE:** Depends on #117329 - please only review the top commit!
---
 .../Linalg/CPU/pack-dynamic-inner-tile.mlir   | 41 ++++++++++++++-----
 1 file changed, 30 insertions(+), 11 deletions(-)

diff --git a/mlir/test/Integration/Dialect/Linalg/CPU/pack-dynamic-inner-tile.mlir b/mlir/test/Integration/Dialect/Linalg/CPU/pack-dynamic-inner-tile.mlir
index 32b7247e60d622..0d2fd977c8d557 100644
--- a/mlir/test/Integration/Dialect/Linalg/CPU/pack-dynamic-inner-tile.mlir
+++ b/mlir/test/Integration/Dialect/Linalg/CPU/pack-dynamic-inner-tile.mlir
@@ -10,10 +10,6 @@
 
 /// End-to-end test for tensor.pack where one of the inner tile sizes is
 /// dynamic.
-///
-/// Note, ATM this is a relatively simple example, with no vectorization and
-/// the dynamic tile size being a compile-time constant. The intention is to
-/// incrementally expand the config to something much more complex.
 
 func.func @main() {
   // Allocate and initialise the inputs
@@ -89,26 +85,49 @@ module @transforms attributes { transform.with_named_sequence } {
     %tiled_pack_op_p, %loops:2 = transform.structured.tile_using_for %pack tile_sizes [1, 1]
        : (!transform.any_op) -> (!transform.any_op, !transform.any_op, !transform.any_op)
 
-    // 2. Decompose the tiled Op into (trimmed for brevity):
+    // 2. Decompose the tiled pack Op into (trimmed for brevity):
     //
     //  %padded = tensor.pad %slice_of_A (..) :
     //      tensor<?x?xi32> to tensor<8x1xi32>
     //  %inserted_slice = tensor.insert_slice %padded into %slice_of_A_pack (...) :
     //      tensor<8x1xi32> into tensor<1x1x?x1xi32>
     //
-    // NOTE: no tile is transposed, hence no linalg.transpose
-    %func_1 = transform.get_parent_op %tiled_pack_op_p {isolated_from_above} : (!transform.any_op) -> !transform.any_op
-    transform.apply_patterns to %func_1 {
+    // (NOTE: no tile is transposed, hence no linalg.transpose)
+    //
+    // This is followed by this decomposition of the pad Op:
+    //
+    //  %c123_i32 = arith.constant 123 : i32
+    //  %slice_of_A = tensor.extract_slice %A[%3, %arg3] [%4, %5] [1, 1] :
+    //    tensor<7x16xi32> to tensor<?x?xi32>
+    //  %empty = tensor.empty() : tensor<8x1xi32>
+    //  %fill = linalg.fill ins(%c123_i32 : i32) outs(%empty :
+    //    tensor<8x1xi32>) -> tensor<8x1xi32>
+    //  %inserted_slice = tensor.insert_slice %slice_of_A into %fill[0, 0] [%4, %5] [1, 1] :
+    //    tensor<?x?xi32> into tensor<8x1xi32>
+    //
+    %func_op = transform.get_parent_op %tiled_pack_op_p {isolated_from_above} : (!transform.any_op) -> !transform.op<"func.func">
+    transform.apply_patterns to %func_op {
       transform.apply_patterns.linalg.decompose_pack_unpack
-    } : !transform.any_op
+      transform.apply_patterns.linalg.decompose_pad
+    } : !transform.op<"func.func">
+
+    // 3. Vectorize linalg.fill.
+    // Vector sizes match the inner tiles in the payload IR.
+    %fill = transform.structured.match ops{["linalg.fill"]} in %func_op : (!transform.op<"func.func">) -> !transform.any_op
+    transform.structured.vectorize %fill vector_sizes [8, 1] : !transform.any_op
+
+    transform.apply_patterns to %func_op {
+      transform.apply_patterns.tensor.fold_tensor_subset_ops
+      transform.apply_patterns.canonicalization
+    } : !transform.op<"func.func">
 
     // 3. Bufferize before lowering to LLVM
     %bufferize = transform.bufferization.one_shot_bufferize %module
       {bufferize_function_boundaries=true} : (!transform.any_op) -> !transform.any_op
 
     // 4. Canonicalize
-    %func_2 = transform.structured.match ops{["func.func"]} in %bufferize : (!transform.any_op) -> !transform.op<"func.func">
-    transform.apply_patterns to %func_2 {
+    %func_op_bufferized = transform.structured.match ops{["func.func"]} in %bufferize : (!transform.any_op) -> !transform.op<"func.func">
+    transform.apply_patterns to %func_op_bufferized {
       transform.apply_patterns.canonicalization
     } : !transform.op<"func.func">
 



More information about the Mlir-commits mailing list