| 1 | //===- LowerVectorScam.cpp - Lower 'vector.scan' operation ----------------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | // |
| 9 | // This file implements target-independent rewrites and utilities to lower the |
| 10 | // 'vector.scan' operation. |
| 11 | // |
| 12 | //===----------------------------------------------------------------------===// |
| 13 | |
| 14 | #include "mlir/Dialect/Arith/IR/Arith.h" |
| 15 | #include "mlir/Dialect/MemRef/IR/MemRef.h" |
| 16 | #include "mlir/Dialect/Utils/IndexingUtils.h" |
| 17 | #include "mlir/Dialect/Vector/IR/VectorOps.h" |
| 18 | #include "mlir/Dialect/Vector/Transforms/LoweringPatterns.h" |
| 19 | #include "mlir/Dialect/Vector/Utils/VectorUtils.h" |
| 20 | #include "mlir/IR/BuiltinTypes.h" |
| 21 | #include "mlir/IR/Location.h" |
| 22 | #include "mlir/IR/PatternMatch.h" |
| 23 | #include "mlir/IR/TypeUtilities.h" |
| 24 | |
| 25 | #define DEBUG_TYPE "vector-broadcast-lowering" |
| 26 | |
| 27 | using namespace mlir; |
| 28 | using namespace mlir::vector; |
| 29 | |
| 30 | /// This function checks to see if the vector combining kind |
| 31 | /// is consistent with the integer or float element type. |
| 32 | static bool isValidKind(bool isInt, vector::CombiningKind kind) { |
| 33 | using vector::CombiningKind; |
| 34 | enum class KindType { FLOAT, INT, INVALID }; |
| 35 | KindType type{KindType::INVALID}; |
| 36 | switch (kind) { |
| 37 | case CombiningKind::MINNUMF: |
| 38 | case CombiningKind::MINIMUMF: |
| 39 | case CombiningKind::MAXNUMF: |
| 40 | case CombiningKind::MAXIMUMF: |
| 41 | case CombiningKind::MAXIMUMNUMF: |
| 42 | case CombiningKind::MINIMUMNUMF: |
| 43 | type = KindType::FLOAT; |
| 44 | break; |
| 45 | case CombiningKind::MINUI: |
| 46 | case CombiningKind::MINSI: |
| 47 | case CombiningKind::MAXUI: |
| 48 | case CombiningKind::MAXSI: |
| 49 | case CombiningKind::AND: |
| 50 | case CombiningKind::OR: |
| 51 | case CombiningKind::XOR: |
| 52 | type = KindType::INT; |
| 53 | break; |
| 54 | case CombiningKind::ADD: |
| 55 | case CombiningKind::MUL: |
| 56 | type = isInt ? KindType::INT : KindType::FLOAT; |
| 57 | break; |
| 58 | } |
| 59 | bool isValidIntKind = (type == KindType::INT) && isInt; |
| 60 | bool isValidFloatKind = (type == KindType::FLOAT) && (!isInt); |
| 61 | return (isValidIntKind || isValidFloatKind); |
| 62 | } |
| 63 | |
| 64 | namespace { |
| 65 | /// Convert vector.scan op into arith ops and vector.insert_strided_slice / |
| 66 | /// vector.extract_strided_slice. |
| 67 | /// |
| 68 | /// Example: |
| 69 | /// |
| 70 | /// ``` |
| 71 | /// %0:2 = vector.scan <add>, %arg0, %arg1 |
| 72 | /// {inclusive = true, reduction_dim = 1} : |
| 73 | /// (vector<2x3xi32>, vector<2xi32>) to (vector<2x3xi32>, vector<2xi32>) |
| 74 | /// ``` |
| 75 | /// |
| 76 | /// is converted to: |
| 77 | /// |
| 78 | /// ``` |
| 79 | /// %cst = arith.constant dense<0> : vector<2x3xi32> |
| 80 | /// %0 = vector.extract_strided_slice %arg0 |
| 81 | /// {offsets = [0, 0], sizes = [2, 1], strides = [1, 1]} |
| 82 | /// : vector<2x3xi32> to vector<2x1xi32> |
| 83 | /// %1 = vector.insert_strided_slice %0, %cst |
| 84 | /// {offsets = [0, 0], strides = [1, 1]} |
| 85 | /// : vector<2x1xi32> into vector<2x3xi32> |
| 86 | /// %2 = vector.extract_strided_slice %arg0 |
| 87 | /// {offsets = [0, 1], sizes = [2, 1], strides = [1, 1]} |
| 88 | /// : vector<2x3xi32> to vector<2x1xi32> |
| 89 | /// %3 = arith.muli %0, %2 : vector<2x1xi32> |
| 90 | /// %4 = vector.insert_strided_slice %3, %1 |
| 91 | /// {offsets = [0, 1], strides = [1, 1]} |
| 92 | /// : vector<2x1xi32> into vector<2x3xi32> |
| 93 | /// %5 = vector.extract_strided_slice %arg0 |
| 94 | /// {offsets = [0, 2], sizes = [2, 1], strides = [1, 1]} |
| 95 | /// : vector<2x3xi32> to vector<2x1xi32> |
| 96 | /// %6 = arith.muli %3, %5 : vector<2x1xi32> |
| 97 | /// %7 = vector.insert_strided_slice %6, %4 |
| 98 | /// {offsets = [0, 2], strides = [1, 1]} |
| 99 | /// : vector<2x1xi32> into vector<2x3xi32> |
| 100 | /// %8 = vector.shape_cast %6 : vector<2x1xi32> to vector<2xi32> |
| 101 | /// return %7, %8 : vector<2x3xi32>, vector<2xi32> |
| 102 | /// ``` |
| 103 | struct ScanToArithOps : public OpRewritePattern<vector::ScanOp> { |
| 104 | using Base::Base; |
| 105 | |
| 106 | LogicalResult matchAndRewrite(vector::ScanOp scanOp, |
| 107 | PatternRewriter &rewriter) const override { |
| 108 | auto loc = scanOp.getLoc(); |
| 109 | VectorType destType = scanOp.getDestType(); |
| 110 | ArrayRef<int64_t> destShape = destType.getShape(); |
| 111 | auto elType = destType.getElementType(); |
| 112 | bool isInt = elType.isIntOrIndex(); |
| 113 | if (!isValidKind(isInt, kind: scanOp.getKind())) |
| 114 | return failure(); |
| 115 | |
| 116 | int64_t reductionDim = scanOp.getReductionDim(); |
| 117 | bool inclusive = scanOp.getInclusive(); |
| 118 | int64_t destRank = destType.getRank(); |
| 119 | VectorType initialValueType = scanOp.getInitialValueType(); |
| 120 | int64_t initialValueRank = initialValueType.getRank(); |
| 121 | |
| 122 | SmallVector<int64_t> reductionShape(destShape); |
| 123 | SmallVector<bool> reductionScalableDims(destType.getScalableDims()); |
| 124 | |
| 125 | // Check before creating any IR so that returning failure() does not |
| 126 | // violate the pattern API contract. |
| 127 | if (reductionScalableDims[reductionDim]) |
| 128 | return rewriter.notifyMatchFailure( |
| 129 | arg&: scanOp, msg: "Trying to reduce scalable dimension - not yet supported!" ); |
| 130 | |
| 131 | VectorType resType = destType; |
| 132 | Value result = arith::ConstantOp::create(builder&: rewriter, location: loc, result: resType, |
| 133 | value: rewriter.getZeroAttr(type: resType)); |
| 134 | |
| 135 | // The reduction dimension, after reducing, becomes 1. It's a fixed-width |
| 136 | // dimension - no need to touch the scalability flag. |
| 137 | reductionShape[reductionDim] = 1; |
| 138 | VectorType reductionType = |
| 139 | VectorType::get(shape: reductionShape, elementType: elType, scalableDims: reductionScalableDims); |
| 140 | |
| 141 | SmallVector<int64_t> offsets(destRank, 0); |
| 142 | SmallVector<int64_t> strides(destRank, 1); |
| 143 | SmallVector<int64_t> sizes(destShape); |
| 144 | sizes[reductionDim] = 1; |
| 145 | ArrayAttr scanSizes = rewriter.getI64ArrayAttr(values: sizes); |
| 146 | ArrayAttr scanStrides = rewriter.getI64ArrayAttr(values: strides); |
| 147 | |
| 148 | Value lastOutput, lastInput; |
| 149 | for (int i = 0; i < destShape[reductionDim]; i++) { |
| 150 | offsets[reductionDim] = i; |
| 151 | ArrayAttr scanOffsets = rewriter.getI64ArrayAttr(values: offsets); |
| 152 | Value input = vector::ExtractStridedSliceOp::create( |
| 153 | builder&: rewriter, location: loc, resultType0: reductionType, source: scanOp.getSource(), offsets: scanOffsets, |
| 154 | sizes: scanSizes, strides: scanStrides); |
| 155 | Value output; |
| 156 | if (i == 0) { |
| 157 | if (inclusive) { |
| 158 | output = input; |
| 159 | } else { |
| 160 | if (initialValueRank == 0) { |
| 161 | // ShapeCastOp cannot handle 0-D vectors |
| 162 | output = vector::BroadcastOp::create(builder&: rewriter, location: loc, result: input.getType(), |
| 163 | source: scanOp.getInitialValue()); |
| 164 | } else { |
| 165 | output = vector::ShapeCastOp::create(builder&: rewriter, location: loc, result: input.getType(), |
| 166 | source: scanOp.getInitialValue()); |
| 167 | } |
| 168 | } |
| 169 | } else { |
| 170 | Value y = inclusive ? input : lastInput; |
| 171 | output = vector::makeArithReduction(b&: rewriter, loc, kind: scanOp.getKind(), |
| 172 | v1: lastOutput, acc: y); |
| 173 | } |
| 174 | result = vector::InsertStridedSliceOp::create(builder&: rewriter, location: loc, valueToStore: output, |
| 175 | dest: result, offsets, strides); |
| 176 | lastOutput = output; |
| 177 | lastInput = input; |
| 178 | } |
| 179 | |
| 180 | Value reduction; |
| 181 | if (initialValueRank == 0) { |
| 182 | Value v = vector::ExtractOp::create(builder&: rewriter, location: loc, source: lastOutput, position: 0); |
| 183 | reduction = |
| 184 | vector::BroadcastOp::create(builder&: rewriter, location: loc, result: initialValueType, source: v); |
| 185 | } else { |
| 186 | reduction = vector::ShapeCastOp::create(builder&: rewriter, location: loc, result: initialValueType, |
| 187 | source: lastOutput); |
| 188 | } |
| 189 | |
| 190 | rewriter.replaceOp(op: scanOp, newValues: {result, reduction}); |
| 191 | return success(); |
| 192 | } |
| 193 | }; |
| 194 | } // namespace |
| 195 | |
| 196 | void mlir::vector::populateVectorScanLoweringPatterns( |
| 197 | RewritePatternSet &patterns, PatternBenefit benefit) { |
| 198 | patterns.add<ScanToArithOps>(arg: patterns.getContext(), args&: benefit); |
| 199 | } |
| 200 | |