[MLIR] Add pattern to bubble up tensor.extract_slice (llvm#126898)

Add a pattern that bubbles up tensor.extract_slice through
tensor.expand_shape, and add a transform op to tensor dialect
to directly use this pattern.
This pattern enables tiling and fusing op chains which contain
tensor.expand_shape if added as a cleanup pattern of tile and fuse
utility.
Without this pattern that would not be possible, as
tensor.expand_shape does not implement the tiling interface.
In addition, registering this pattern as a cleanup pattern for
transform.structured.fuse.
The pattern was first implement in IREE project by
Quinn Dawkins and is being upstreamed.

---------

Co-authored-by: Quinn Dawkins <[email protected]>

Author: ofri-frishman

mlir/include/mlir/Dialect/Tensor/TransformOps/TensorTransformOps.td

@@ -101,6 +101,17 @@ def ApplyReassociativeReshapeFoldingPatternsOp : Op<Transform_Dialect,
   let assemblyFormat = "attr-dict";
 }
 
+def ApplyBubbleUpExtractSlicePatternsOp : Op<Transform_Dialect,
+    "apply_patterns.tensor.bubble_up_extract_slice",
+    [DeclareOpInterfaceMethods<PatternDescriptorOpInterface>]> {
+  let description = [{
+    Indicates that producers of tensor.extract_slice should swap and operate on 
+    the result of the slice.
+  }];
+
+  let assemblyFormat = "attr-dict";
+}
+
 def ApplyRewriteTensorOpsAsConstantPatternsOp : Op<Transform_Dialect,
     "apply_patterns.tensor.rewrite_as_constant",
     [DeclareOpInterfaceMethods<PatternDescriptorOpInterface>]> {

mlir/include/mlir/Dialect/Tensor/Transforms/Transforms.h

@@ -58,6 +58,12 @@ void populateFoldTensorSubsetIntoVectorTransferPatterns(
 void populateMergeConsecutiveInsertExtractSlicePatterns(
     RewritePatternSet &patterns);
 
+/// Appends patterns that are used to bubble up tensor.extract slice op above
+/// its producer. When used as cleanup patterns of tile and fuse, enables fusing
+/// the producer with the consumer even if the producer does not implement the
+/// tiling interface.
+void populateBubbleUpExtractSliceOpPatterns(RewritePatternSet &patterns);
+
 /// Populates `patterns` with patterns that drop redundant tensor.insert_slice
 /// rank expansions.
 void populateDropRedundantInsertSliceRankExpansionPatterns(

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

@@ -592,6 +592,7 @@ transform::FuseOp::apply(transform::TransformRewriter &rewriter,
     RewritePatternSet patterns(context);
     tensor::ExtractSliceOp::getCanonicalizationPatterns(patterns, context);
     tensor::populateMergeConsecutiveInsertExtractSlicePatterns(patterns);
+    tensor::populateBubbleUpExtractSliceOpPatterns(patterns);
     tileAndFuseOptions.cleanupPatterns = std::move(patterns);
   }
 

mlir/lib/Dialect/Tensor/TransformOps/TensorTransformOps.cpp

@@ -120,6 +120,11 @@ void transform::ApplyReassociativeReshapeFoldingPatternsOp::populatePatterns(
   tensor::populateReassociativeReshapeFoldingPatterns(patterns);
 }
 
+void transform::ApplyBubbleUpExtractSlicePatternsOp::populatePatterns(
+    RewritePatternSet &patterns) {
+  tensor::populateBubbleUpExtractSliceOpPatterns(patterns);
+}
+
 void transform::ApplyRewriteTensorOpsAsConstantPatternsOp::populatePatterns(
     RewritePatternSet &patterns) {
   ControlFoldFn defaultControlFn = [](OpOperand *fusedOperand) {

mlir/lib/Dialect/Tensor/Transforms/ReshapePatterns.cpp

@@ -6,10 +6,14 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "mlir/Dialect/Affine/IR/AffineOps.h"
+#include "mlir/Dialect/Arith/Utils/Utils.h"
 #include "mlir/Dialect/Tensor/IR/Tensor.h"
 #include "mlir/Dialect/Tensor/Transforms/Transforms.h"
 #include "mlir/IR/PatternMatch.h"
+#include "mlir/Interfaces/ValueBoundsOpInterface.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/LogicalResult.h"
 
 using namespace mlir;
 using namespace mlir::tensor;
@@ -210,6 +214,214 @@ struct BubbleUpExpandThroughParallelCollapse
   }
 };
 
+/// Converts `tensor.extract_slice(tensor.expand_shape)` to
+/// `tensor.expand_shape(tensor.extract_slice)`.
+///
+/// For this transformation to be possible, the slice must be fully contiguous
+/// within each reassociation group of the expand_shape. A slice is defined as
+/// fully contiguous within a reassociation group if after flattening the
+/// reassociation group to a single 1D range, then the slice taken out of the
+/// group could be defined as a single contiguous subrange within that range.
+///
+/// Rank reducing slices are not supported.
+///
+/// Example:
+/// The transformation is possible because each reassociation group has a
+/// contiguous slice (i.e., [2x4->2x4], [2x8->1x5], [4x2x4->1x1x4]).
+/// ```
+/// BEFORE:
+/// %reshape = tensor.expand_shape %in [[0, 1], [2, 3], [4, 5, 6]]
+///     tensor<8x16x32xf32> to tensor<2x4x2x8x4x2x4xf32>
+/// %slice = tensor.extract_slice %reshape ...
+///     tensor<2x4x2x8x4x2x4xf32> to tensor<2x4x1x5x1x1x4xf32>
+///
+/// AFTER:
+/// %slice = tensor.extract_slice %in ...
+///     tensor<8x16x32xf32> to tensor<8x5x4xf32>
+/// %reshape = tensor.expand_shape %slice [[0, 1], [2, 3], [4, 5, 6]]
+///     tensor<8x5x4xf32> to tensor<2x4x1x5x1x1x4xf32>
+/// ```
+///
+/// Note - this pattern could be extended to be a swap pattern between
+/// `tensor.expand_shape` and `tensor.extract_slice`, but is currently
+/// implemented only as a bubble up pattern for `tensor.extract_slice`.
+struct BubbleUpExpandShapeThroughExtractSlice
+    : public OpRewritePattern<tensor::ExtractSliceOp> {
+  using OpRewritePattern<tensor::ExtractSliceOp>::OpRewritePattern;
+
+  LogicalResult matchAndRewrite(tensor::ExtractSliceOp sliceOp,
+                                PatternRewriter &rewriter) const override {
+    auto expandShapeOp =
+        sliceOp.getSource().getDefiningOp<tensor::ExpandShapeOp>();
+
+    if (checkPreconditionForBubbleUpExtractSlice(sliceOp, expandShapeOp,
+                                                 rewriter)
+            .failed())
+      return failure();
+
+    // The tensor.extract_slice before applying the pattern works on the result
+    // of the tensor.expand_shape, so variables (i.e. inputs for ExtractSliceOp)
+    // referring to the state before applying the pattern are named with the
+    // prefix "expanded", and ones referring to the state after applying the
+    // pattern are named with the prefix "collapsed".
+    SmallVector<OpFoldResult> expandedOffsets = sliceOp.getMixedOffsets();
+    SmallVector<OpFoldResult> expandedSizes = sliceOp.getMixedSizes();
+    SmallVector<OpFoldResult> expandedShape =
+        getMixedValues(expandShapeOp.getStaticOutputShape(),
+                       expandShapeOp.getOutputShape(), rewriter);
+
+    // Helper variables and function for accumulating the size values.
+    Location loc = expandShapeOp->getLoc();
+    AffineExpr d0, d1, d2;
+    bindDims(rewriter.getContext(), d0, d1, d2);
+    // Multiply two integers.
+    auto mul = [&](OpFoldResult v1, OpFoldResult v2) {
+      auto mulMap = AffineMap::get(2, 0, {d0 * d1});
+      return affine::makeComposedFoldedAffineApply(rewriter, loc, mulMap,
+                                                   {v1, v2});
+    };
+
+    // Compute new offsets, sizes, and strides for tensor.extract_slice.
+    // The new tensor.extract_slice will work on a tensor that has has a rank of
+    // ReassociationIndices.size(). In the loop a single offset, size, and
+    // stride value is computed per reassociation group.
+    SmallVector<OpFoldResult> collapsedOffsets, collapsedSizes,
+        collapsedStrides;
+    for (const ReassociationIndices &indices :
+         expandShapeOp.getReassociationIndices()) {
+      // collapsedSize will hold the size of the single dim that represents the
+      // reassociation group in the non expanded tensor.
+      OpFoldResult collapsedSize = rewriter.getIndexAttr(1);
+      // The reassocGroupSizes and reassocGroupOffsets are used to create an
+      // affine.linearize_index op to linearize the single offset value required
+      // for this reassociation group.
+      SmallVector<OpFoldResult> reassocGroupSizes, reassocGroupOffsets;
+
+      for (long expandedDim : indices) {
+        // reassocGroupSizes and reassocGroupOffsets can be obtained directly
+        // from the expanded state, but the collapsed size requires calculation
+        // as it did not previously exist.
+        reassocGroupSizes.push_back(expandedShape[expandedDim]);
+        reassocGroupOffsets.push_back(expandedOffsets[expandedDim]);
+        collapsedSize = mul(collapsedSize, expandedSizes[expandedDim]);
+      }
+
+      SmallVector<Value> offsetVals =
+          llvm::map_to_vector(reassocGroupOffsets, [&](OpFoldResult ofr) {
+            return getValueOrCreateConstantIndexOp(rewriter, loc, ofr);
+          });
+      OpFoldResult collapsedOffset =
+          rewriter
+              .create<affine::AffineLinearizeIndexOp>(loc, offsetVals,
+                                                      reassocGroupSizes,
+                                                      /*disjoint=*/true)
+              .getResult();
+      collapsedOffsets.push_back(collapsedOffset);
+      collapsedSizes.push_back(collapsedSize);
+
+      // Only unit stride is supported.
+      collapsedStrides.push_back(rewriter.getIndexAttr(1));
+    }
+
+    // The shape of the result can be obtained from the sizes passed in.
+    SmallVector<Value> dynDims;
+    SmallVector<int64_t> shape;
+    dispatchIndexOpFoldResults(expandedSizes, dynDims, shape);
+    RankedTensorType resultType = RankedTensorType::get(
+        shape, expandShapeOp.getResultType().getElementType());
+
+    // Create a new ExtractSliceOp and ExpandShapeOp.
+    Value newSliceOp = rewriter.create<tensor::ExtractSliceOp>(
+        loc, expandShapeOp.getSrc(), collapsedOffsets, collapsedSizes,
+        collapsedStrides);
+    rewriter.replaceOpWithNewOp<tensor::ExpandShapeOp>(
+        sliceOp, resultType, newSliceOp,
+        expandShapeOp.getReassociationIndices(), expandedSizes);
+    return success();
+  }
+
+  // Helper function to check if all the required conditions for the
+  // tensor.extract_slice to be bubbled up through the tensor.expand_shape are
+  // met.
+  LogicalResult
+  checkPreconditionForBubbleUpExtractSlice(tensor::ExtractSliceOp sliceOp,
+                                           tensor::ExpandShapeOp expandShapeOp,
+                                           PatternRewriter &rewriter) const {
+
+    if (!expandShapeOp) {
+      return rewriter.notifyMatchFailure(
+          sliceOp, "tensor.extract_slice source not produced by expand_shape");
+    }
+
+    if (!sliceOp.hasUnitStride()) {
+      return rewriter.notifyMatchFailure(
+          sliceOp, "unsupported: non-unit stride. Only contiguous slices can "
+                   "be supported in this transformation.");
+    }
+
+    SmallVector<OpFoldResult> offsets = sliceOp.getMixedOffsets();
+    SmallVector<OpFoldResult> sizes = sliceOp.getMixedSizes();
+
+    if (static_cast<size_t>(sliceOp.getResultType().getRank()) !=
+        sizes.size()) {
+      return rewriter.notifyMatchFailure(sliceOp,
+                                         "unimplemented: rank reducing slice");
+    }
+
+    SmallVector<OpFoldResult> outputShape =
+        getMixedValues(expandShapeOp.getStaticOutputShape(),
+                       expandShapeOp.getOutputShape(), rewriter);
+
+    std::function<bool(OpFoldResult, OpFoldResult, OpFoldResult)>
+        isZeroOffsetAndFullSize =
+            [](OpFoldResult offset, OpFoldResult sliceSize, OpFoldResult size) {
+              if (!isConstantIntValue(offset, 0))
+                return false;
+              FailureOr<bool> maybeEqual =
+                  ValueBoundsConstraintSet::areEqual(sliceSize, size);
+              return llvm::succeeded(maybeEqual) && maybeEqual.value();
+            };
+
+    // Check that the slice is contiguous within each reassociation group.
+    // The slice is contiguous only if after the first dimension where a non
+    // unit slice is taken, the slice size on all subsequent dimensions of the
+    // group is equal to the entire size of the dimension.
+    // Examples of contiguous slices:
+    //   full sizes: [8, 8, 10] slice offsets: [0, 0, 0] slice sizes: [1, 1, 10]
+    //   full sizes: [5, 10] slice offsets: [3, 0] slice sizes: [2, 10]
+    // Examples of non contiguous slices:
+    //   full sizes: [8, 8, 10] slice offsets: [0, 0, 0] slice sizes: [1, 2, 5]
+    //   full sizes: [5, 10] slice offsets: [0, 4] slice sizes: [2, 5]
+    for (const ReassociationIndices &indices :
+         expandShapeOp.getReassociationIndices()) {
+      int64_t i = 0;
+      int64_t e = indices.size();
+      // Find the first expanded dim after the first dim with non-unit extracted
+      // size.
+      for (; i < e; ++i) {
+        if (!isConstantIntValue(sizes[indices[i]], 1)) {
+          // +1 to skip the first non-unit size dim.
+          i++;
+          break;
+        }
+      }
+
+      // Verify that all subsequent dimensions extract the full size of the
+      // source tensor.
+      for (; i < e; ++i) {
+        int64_t expandedDim = indices[i];
+        if (!isZeroOffsetAndFullSize(offsets[expandedDim], sizes[expandedDim],
+                                     outputShape[expandedDim])) {
+          return rewriter.notifyMatchFailure(
+              sliceOp, "Not a contiguous slice of the expanded tensor.");
+        }
+      }
+    }
+
+    return success();
+  }
+};
+
 } // namespace
 
 void mlir::tensor::populateReassociativeReshapeFoldingPatterns(
@@ -227,3 +439,8 @@ void mlir::tensor::populateBubbleUpExpandShapePatterns(
     RewritePatternSet &patterns) {
   patterns.add<BubbleUpExpandThroughParallelCollapse>(patterns.getContext());
 }
+
+void mlir::tensor::populateBubbleUpExtractSliceOpPatterns(
+    RewritePatternSet &patterns) {
+  patterns.add<BubbleUpExpandShapeThroughExtractSlice>(patterns.getContext());
+}