use shape_cast as canonical type for extract broadcast and transpose

Author: newling

mlir/lib/Dialect/Vector/IR/VectorOps.cpp

@@ -2351,11 +2351,45 @@ LogicalResult foldExtractFromFromElements(ExtractOp extractOp,
   return success();
 }
 
+/// For example,
+/// ```
+/// %0 = vector.extract %arg0[0] : vector<4xf32> from vector<1x4xf32>
+/// ```
+/// becomes
+/// ```
+/// %0 = vector.shape_cast %arg0 : vector<1x4xf32> to vector<4xf32>
+/// ```
+struct ExtractToShapeCast final : public OpRewritePattern<vector::ExtractOp> {
+  using OpRewritePattern::OpRewritePattern;
+  LogicalResult matchAndRewrite(vector::ExtractOp extractOp,
+                                PatternRewriter &rewriter) const override {
+    VectorType sourceType = extractOp.getSourceVectorType();
+    VectorType outType = dyn_cast<VectorType>(extractOp.getType());
+    if (!outType)
+      return failure();
+
+    // Negative values in `position` indicates poison, cannot convert to
+    // shape_cast
+    if (llvm::any_of(extractOp.getMixedPosition(),
+                     [](OpFoldResult v) { return !isConstantIntValue(v, 0); }))
+      return failure();
+
+    if (sourceType.getNumElements() != outType.getNumElements())
+      return failure();
+
+    rewriter.replaceOpWithNewOp<vector::ShapeCastOp>(extractOp, outType,
+                                                     extractOp.getVector());
+    return success();
+  }
+};
+
 } // namespace
 
 void ExtractOp::getCanonicalizationPatterns(RewritePatternSet &results,
                                             MLIRContext *context) {
-  results.add<ExtractOpFromBroadcast, ExtractOpFromCreateMask>(context);
+  results
+      .add<ExtractOpFromBroadcast, ExtractOpFromCreateMask, ExtractToShapeCast>(
+          context);
   results.add(foldExtractFromShapeCastToShapeCast);
   results.add(foldExtractFromFromElements);
 }
@@ -2867,13 +2901,40 @@ struct BroadcastFolder : public OpRewritePattern<BroadcastOp> {
     return success();
   }
 };
+
+/// For example,
+/// ```
+/// %0 = vector.broadcast %arg0 : vector<4xi8> to vector<1x1x4xi8>
+/// ```
+/// becomes
+/// ```
+/// %0 = vector.shape_cast %arg0 : vector<4xi8> to vector<1x1x4xi8>
+/// ```
+struct BroadcastToShapeCast final
+    : public OpRewritePattern<vector::BroadcastOp> {
+  using OpRewritePattern::OpRewritePattern;
+  LogicalResult matchAndRewrite(vector::BroadcastOp broadcast,
+                                PatternRewriter &rewriter) const override {
+    auto sourceType = dyn_cast<VectorType>(broadcast.getSourceType());
+    if (!sourceType) {
+      return rewriter.notifyMatchFailure(
+          broadcast, "source is a scalar, shape_cast doesn't support scalar");
+    }
+
+    VectorType outType = broadcast.getType();
+    if (sourceType.getNumElements() != outType.getNumElements())
+      return failure();
+
+    rewriter.replaceOpWithNewOp<vector::ShapeCastOp>(broadcast, outType,
+                                                     broadcast.getSource());
+    return success();
+  }
+};
 } // namespace
 
 void BroadcastOp::getCanonicalizationPatterns(RewritePatternSet &results,
                                               MLIRContext *context) {
-  // BroadcastToShapeCast is not a default canonicalization, it is opt-in by
-  // calling `populateCastAwayVectorLeadingOneDimPatterns`
-  results.add<BroadcastFolder>(context);
+  results.add<BroadcastFolder, BroadcastToShapeCast>(context);
 }
 
 //===----------------------------------------------------------------------===//
@@ -5816,30 +5877,6 @@ LogicalResult ShapeCastOp::verify() {
   return success();
 }
 
-/// Return true if `transpose` does not permute a pair of non-unit dims.
-/// By `order preserving` we mean that the flattened versions of the input and
-/// output vectors are (numerically) identical. In other words `transpose` is
-/// effectively a shape cast.
-static bool isOrderPreserving(TransposeOp transpose) {
-  ArrayRef<int64_t> permutation = transpose.getPermutation();
-  VectorType sourceType = transpose.getSourceVectorType();
-  ArrayRef<int64_t> inShape = sourceType.getShape();
-  ArrayRef<bool> inDimIsScalable = sourceType.getScalableDims();
-  auto isNonScalableUnitDim = [&](int64_t dim) {
-    return inShape[dim] == 1 && !inDimIsScalable[dim];
-  };
-  int64_t current = 0;
-  for (auto p : permutation) {
-    if (!isNonScalableUnitDim(p)) {
-      if (p < current) {
-        return false;
-      }
-      current = p;
-    }
-  }
-  return true;
-}
-
 OpFoldResult ShapeCastOp::fold(FoldAdaptor adaptor) {
 
   VectorType resultType = getType();
@@ -5854,22 +5891,6 @@ OpFoldResult ShapeCastOp::fold(FoldAdaptor adaptor) {
     return getResult();
   }
 
-  // shape_cast(transpose(x)) -> shape_cast(x)
-  if (auto transpose = getSource().getDefiningOp<TransposeOp>()) {
-    if (isOrderPreserving(transpose)) {
-      setOperand(transpose.getVector());
-      return getResult();
-    }
-    return {};
-  }
-
-  // Y = shape_cast(broadcast(X))
-  //      -> X, if X and Y have same type
-  if (auto bcastOp = getSource().getDefiningOp<BroadcastOp>()) {
-    if (bcastOp.getSourceType() == resultType)
-      return bcastOp.getSource();
-  }
-
   // shape_cast(constant) -> constant
   if (auto splatAttr =
           llvm::dyn_cast_if_present<SplatElementsAttr>(adaptor.getSource()))
@@ -5991,10 +6012,7 @@ class ShapeCastCreateMaskFolderTrailingOneDim final
   }
 };
 
-/// Pattern to rewrite Y = ShapeCast(Broadcast(X)) as either
-///   i) Y = ShapeCast(X), or
-///  ii) Y = Broadcast(X)
-/// If both (i) and (ii) are possible, (i) is chosen.
+/// Pattern to rewrite Y = ShapeCast(Broadcast(X)) as Y = Broadcast(X)
 class ShapeCastBroadcastFolder final : public OpRewritePattern<ShapeCastOp> {
 public:
   using OpRewritePattern::OpRewritePattern;
@@ -6009,22 +6027,6 @@ class ShapeCastBroadcastFolder final : public OpRewritePattern<ShapeCastOp> {
     auto srcVectorType = dyn_cast<VectorType>(broadcastOp.getSourceType());
     bool srcIsScalar = !srcVectorType;
 
-    // Replace Y = ShapeCast(Broadcast(X)) with Y = ShapeCast(X).
-    // Example:
-    // %0 = vector.broadcast %in : vector<3x4xf32> to vector<1x3x4xf32>
-    // %1 = vector.shape_cast %0 : vector<1x3x4xf32> to vector<12xf32>
-    // to
-    // %1 = vector.shape_cast %in : vector<3x4xf32> to vector<12xf32>
-    if (srcVectorType) {
-      if (srcVectorType.getNumElements() ==
-          shapeCastOp.getResultVectorType().getNumElements()) {
-        rewriter.replaceOpWithNewOp<vector::ShapeCastOp>(
-            shapeCastOp, shapeCastOp.getResultVectorType(),
-            broadcastOp.getSource());
-        return success();
-      }
-    }
-
     // Replace Y = ShapeCast(Broadcast(X)) with Y = Broadcast(X)
     // Example
     // %0 = vector.broadcast %in : vector<3xf32> to vector<2x4x3xf32>
@@ -6225,21 +6227,6 @@ OpFoldResult vector::TransposeOp::fold(FoldAdaptor adaptor) {
   if (llvm::dyn_cast_if_present<ub::PoisonAttr>(adaptor.getVector()))
     return ub::PoisonAttr::get(getContext());
 
-  // Eliminate identity transposes, and more generally any transposes that
-  // preserves the shape without permuting elements.
-  //
-  // Examples of what to fold:
-  // %0 = vector.transpose %arg, [0, 1] : vector<1x1xi8> to vector<1x1xi8>
-  // %0 = vector.transpose %arg, [0, 1] : vector<2x2xi8> to vector<2x2xi8>
-  // %0 = vector.transpose %arg, [1, 0] : vector<1x1xi8> to vector<1x1xi8>
-  //
-  // Example of what NOT to fold:
-  // %0 = vector.transpose %arg, [1, 0] : vector<2x2xi8> to vector<2x2xi8>
-  //
-  if (getSourceVectorType() == getResultVectorType() &&
-      isOrderPreserving(*this))
-    return getVector();
-
   return {};
 }
 
@@ -6359,32 +6346,6 @@ class FoldTransposeCreateMask final : public OpRewritePattern<TransposeOp> {
   }
 };
 
-/// Folds transpose(shape_cast) into a new shape_cast.
-class FoldTransposeShapeCast final : public OpRewritePattern<TransposeOp> {
-public:
-  using OpRewritePattern::OpRewritePattern;
-
-  LogicalResult matchAndRewrite(TransposeOp transposeOp,
-                                PatternRewriter &rewriter) const override {
-    auto shapeCastOp =
-        transposeOp.getVector().getDefiningOp<vector::ShapeCastOp>();
-    if (!shapeCastOp)
-      return failure();
-    if (!isOrderPreserving(transposeOp))
-      return failure();
-
-    VectorType resultType = transposeOp.getType();
-
-    // We don't need to check isValidShapeCast at this point, because it is
-    // guaranteed that merging the transpose into the the shape_cast is a valid
-    // shape_cast, because the transpose just inserts/removes ones.
-
-    rewriter.replaceOpWithNewOp<vector::ShapeCastOp>(transposeOp, resultType,
-                                                     shapeCastOp.getSource());
-    return success();
-  }
-};
-
 /// Folds transpose(broadcast(x)) to broadcast(x) if the transpose is
 /// 'order preserving', where 'order preserving' means the flattened
 /// inputs and outputs of the transpose have identical (numerical) values.
@@ -6480,12 +6441,73 @@ class FoldTransposeBroadcast : public OpRewritePattern<vector::TransposeOp> {
   }
 };
 
+/// Return true if `transpose` does not permute a pair of non-unit dims.
+/// By `order preserving` we mean that the flattened versions of the input and
+/// output vectors are (numerically) identical. In other words `transpose` is
+/// effectively a shape cast.
+static bool isOrderPreserving(TransposeOp transpose) {
+  ArrayRef<int64_t> permutation = transpose.getPermutation();
+  VectorType sourceType = transpose.getSourceVectorType();
+  ArrayRef<int64_t> inShape = sourceType.getShape();
+  ArrayRef<bool> inDimIsScalable = sourceType.getScalableDims();
+  auto isNonScalableUnitDim = [&](int64_t dim) {
+    return inShape[dim] == 1 && !inDimIsScalable[dim];
+  };
+  int64_t current = 0;
+  for (auto p : permutation) {
+    if (!isNonScalableUnitDim(p)) {
+      if (p < current) {
+        return false;
+      }
+      current = p;
+    }
+  }
+  return true;
+}
+
+/// For example,
+/// ```
+/// %0 = vector.transpose %arg0, [0, 2, 1] :
+///                   vector<2x1x2xf32> to vector<2x2x1xf32>
+/// ```
+/// becomes
+/// ```
+/// %0 = vector.shape_cast %arg0 :
+///                   vector<2x1x2xf32> to vector<2x2x1xf32>
+/// ```
+struct TransposeToShapeCast final
+    : public OpRewritePattern<vector::TransposeOp> {
+  using OpRewritePattern::OpRewritePattern;
+  LogicalResult matchAndRewrite(vector::TransposeOp transpose,
+                                PatternRewriter &rewriter) const override {
+
+    // This folder does
+    //    shape_cast(transpose) -> shape_cast
+    // But another pattern, ConvertIllegalShapeCastOpsToTransposes, does
+    //    shape_cast -> shape_cast(transpose)
+    // i.e. the complete opposite. When paired, these 2 patterns can cause
+    // infinite cycles in pattern rewriting.
+    // ConvertIllegalShapeCastOpsToTransposes only matches on scalable
+    // vectors, so by disabling this folder for scalable vectors the
+    // cycle is avoided.
+    // TODO: Check if ConvertIllegalShapeCastOpsToTransposes is
+    // still needed. If it's not, then we can fold here.
+    if (!isOrderPreserving(transpose) || transpose.getType().isScalable()) {
+      return rewriter.notifyMatchFailure(
+          transpose, "not order preserving, so not semantically a 'copy'");
+    }
+    rewriter.replaceOpWithNewOp<vector::ShapeCastOp>(
+        transpose, transpose.getType(), transpose.getVector());
+    return success();
+  }
+};
+
 } // namespace
 
 void vector::TransposeOp::getCanonicalizationPatterns(
     RewritePatternSet &results, MLIRContext *context) {
-  results.add<FoldTransposeCreateMask, FoldTransposeShapeCast, TransposeFolder,
-              FoldTransposeSplat, FoldTransposeBroadcast>(context);
+  results.add<FoldTransposeCreateMask, TransposeFolder, FoldTransposeSplat,
+              FoldTransposeBroadcast, TransposeToShapeCast>(context);
 }
 
 //===----------------------------------------------------------------------===//

mlir/lib/Dialect/Vector/Transforms/LowerVectorTranspose.cpp

@@ -382,63 +382,6 @@ class TransposeOpLowering : public OpRewritePattern<vector::TransposeOp> {
   vector::VectorTransposeLowering vectorTransposeLowering;
 };
 
-/// Rewrites vector.transpose as vector.shape_cast. This pattern is only applied
-/// to 2D vectors with at least one unit dim. For example:
-///
-/// Replace:
-///   vector.transpose %0, [1, 0] : vector<4x1xi32>> to
-///                                 vector<1x4xi32>
-/// with:
-///   vector.shape_cast %0 : vector<4x1xi32> to vector<1x4xi32>
-///
-/// Source with leading unit dim (inverse) is also replaced. Unit dim must
-/// be fixed. Non-unit dim can be scalable.
-///
-/// TODO: This pattern was introduced specifically to help lower scalable
-/// vectors. In hindsight, a more specialised canonicalization (for shape_cast's
-/// to cancel out) would be preferable:
-///
-///  BEFORE:
-///     %0 = some_op
-///     %1 = vector.shape_cast %0 : vector<[4]xf32> to vector<[4]x1xf32>
-///     %2 = vector.transpose %1 [1, 0] : vector<[4]x1xf32> to vector<1x[4]xf32>
-///  AFTER:
-///     %0 = some_op
-///     %1 = vector.shape_cast %0 : vector<[4]xf32> to vector<1x[4]xf32>
-///
-/// Given the context above, we may want to consider (re-)moving this pattern
-/// at some later time. I am leaving it for now in case there are other users
-/// that I am not aware of.
-class Transpose2DWithUnitDimToShapeCast
-    : public OpRewritePattern<vector::TransposeOp> {
-public:
-  using OpRewritePattern::OpRewritePattern;
-
-  Transpose2DWithUnitDimToShapeCast(MLIRContext *context,
-                                    PatternBenefit benefit = 1)
-      : OpRewritePattern<vector::TransposeOp>(context, benefit) {}
-
-  LogicalResult matchAndRewrite(vector::TransposeOp op,
-                                PatternRewriter &rewriter) const override {
-    Value input = op.getVector();
-    VectorType resType = op.getResultVectorType();
-
-    // Set up convenience transposition table.
-    ArrayRef<int64_t> transp = op.getPermutation();
-
-    if (resType.getRank() == 2 &&
-        ((resType.getShape().front() == 1 &&
-          !resType.getScalableDims().front()) ||
-         (resType.getShape().back() == 1 &&
-          !resType.getScalableDims().back())) &&
-        transp == ArrayRef<int64_t>({1, 0})) {
-      rewriter.replaceOpWithNewOp<vector::ShapeCastOp>(op, resType, input);
-      return success();
-    }
-
-    return failure();
-  }
-};
 
 /// Rewrite a 2-D vector.transpose as a sequence of shuffle ops.
 /// If the strategy is Shuffle1D, it will be lowered to:
@@ -511,8 +454,7 @@ class TransposeOp2DToShuffleLowering
 void mlir::vector::populateVectorTransposeLoweringPatterns(
     RewritePatternSet &patterns,
     VectorTransposeLowering vectorTransposeLowering, PatternBenefit benefit) {
-  patterns.add<Transpose2DWithUnitDimToShapeCast>(patterns.getContext(),
-                                                  benefit);
+  BroadcastOp::getCanonicalizationPatterns(patterns, patterns.getContext());
   patterns.add<TransposeOpLowering, TransposeOp2DToShuffleLowering>(
       vectorTransposeLowering, patterns.getContext(), benefit);
 }