1 | //===- RuntimeOpVerification.cpp - Op Verification ------------------------===// |
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 | #include "mlir/Dialect/Linalg/Transforms/RuntimeOpVerification.h" |
10 | |
11 | #include "mlir/Dialect/Affine/IR/AffineOps.h" |
12 | #include "mlir/Dialect/Arith/IR/Arith.h" |
13 | #include "mlir/Dialect/Arith/Utils/Utils.h" |
14 | #include "mlir/Dialect/ControlFlow/IR/ControlFlowOps.h" |
15 | #include "mlir/Dialect/Index/IR/IndexAttrs.h" |
16 | #include "mlir/Dialect/Index/IR/IndexDialect.h" |
17 | #include "mlir/Dialect/Index/IR/IndexOps.h" |
18 | #include "mlir/Dialect/Linalg/IR/Linalg.h" |
19 | #include "mlir/Dialect/MemRef/IR/MemRef.h" |
20 | #include "mlir/Dialect/Tensor/IR/Tensor.h" |
21 | #include "mlir/Interfaces/RuntimeVerifiableOpInterface.h" |
22 | |
23 | namespace mlir { |
24 | namespace linalg { |
25 | namespace { |
26 | /// Verify that the runtime sizes of the operands to linalg structured ops are |
27 | /// compatible with the runtime sizes inferred by composing the loop ranges with |
28 | /// the linalg op's indexing maps. This is similar to the verifier except that |
29 | /// here we insert IR to perform the verification at runtime. |
30 | template <typename T> |
31 | struct StructuredOpInterface |
32 | : public RuntimeVerifiableOpInterface::ExternalModel< |
33 | StructuredOpInterface<T>, T> { |
34 | void generateRuntimeVerification(Operation *op, OpBuilder &builder, |
35 | Location loc) const { |
36 | auto linalgOp = llvm::cast<LinalgOp>(op); |
37 | |
38 | SmallVector<Range> loopRanges = linalgOp.createLoopRanges(builder, loc); |
39 | auto [starts, ends, _] = getOffsetsSizesAndStrides(loopRanges); |
40 | |
41 | auto zero = builder.create<arith::ConstantIndexOp>(loc, 0); |
42 | auto one = builder.create<arith::ConstantIndexOp>(loc, 1); |
43 | |
44 | // Subtract one from the loop ends before composing with the indexing map |
45 | transform(ends, ends.begin(), [&](OpFoldResult end) { |
46 | auto endValue = getValueOrCreateConstantIndexOp(b&: builder, loc, ofr: end); |
47 | return builder.createOrFold<index::SubOp>(loc, endValue, one); |
48 | }); |
49 | |
50 | for (OpOperand &opOperand : linalgOp->getOpOperands()) { |
51 | AffineMap indexingMap = linalgOp.getMatchingIndexingMap(&opOperand); |
52 | auto startIndices = affine::makeComposedFoldedMultiResultAffineApply( |
53 | builder, loc, indexingMap, starts); |
54 | auto endIndices = affine::makeComposedFoldedMultiResultAffineApply( |
55 | builder, loc, indexingMap, ends); |
56 | |
57 | for (auto dim : llvm::seq(linalgOp.getRank(&opOperand))) { |
58 | auto startIndex = |
59 | getValueOrCreateConstantIndexOp(builder, loc, startIndices[dim]); |
60 | auto endIndex = |
61 | getValueOrCreateConstantIndexOp(builder, loc, endIndices[dim]); |
62 | |
63 | // Generate: |
64 | // minIndex = min(startIndex, endIndex) |
65 | // assert(minIndex >= 0) |
66 | // To ensure we do not generate a negative index. We take the minimum of |
67 | // the start and end indices in order to handle reverse loops such as |
68 | // `affine_map<(i) -> (3 - i)>` |
69 | auto min = |
70 | builder.createOrFold<index::MinSOp>(loc, startIndex, endIndex); |
71 | auto cmpOp = builder.createOrFold<index::CmpOp>( |
72 | loc, index::IndexCmpPredicate::SGE, min, zero); |
73 | auto msg = RuntimeVerifiableOpInterface::generateErrorMessage( |
74 | linalgOp, "unexpected negative result on dimension #" + |
75 | std::to_string(dim) + " of input/output operand #" + |
76 | std::to_string(opOperand.getOperandNumber())); |
77 | builder.createOrFold<cf::AssertOp>(loc, cmpOp, msg); |
78 | |
79 | // Generate: |
80 | // inferredDimSize = max(startIndex, endIndex) + 1 |
81 | // actualDimSize = dim(operand) |
82 | // assert(inferredDimSize <= actualDimSize) |
83 | // To ensure that we do not index past the bounds of the operands. |
84 | auto max = |
85 | builder.createOrFold<index::MaxSOp>(loc, startIndex, endIndex); |
86 | |
87 | auto inferredDimSize = |
88 | builder.createOrFold<index::AddOp>(loc, max, one); |
89 | |
90 | auto actualDimSize = |
91 | createOrFoldDimOp(builder, loc, opOperand.get(), dim); |
92 | |
93 | // Similar to the verifier, when the affine expression in the indexing |
94 | // map is complicated, we just check that the inferred dimension sizes |
95 | // are in the boundary of the operands' size. Being more precise than |
96 | // that is difficult. |
97 | auto predicate = isa<AffineDimExpr>(indexingMap.getResult(dim)) |
98 | ? index::IndexCmpPredicate::EQ |
99 | : index::IndexCmpPredicate::SLE; |
100 | |
101 | cmpOp = builder.createOrFold<index::CmpOp>( |
102 | loc, predicate, inferredDimSize, actualDimSize); |
103 | msg = RuntimeVerifiableOpInterface::generateErrorMessage( |
104 | linalgOp, "dimension #" + std::to_string(dim) + |
105 | " of input/output operand #" + |
106 | std::to_string(opOperand.getOperandNumber()) + |
107 | " is incompatible with inferred dimension size" ); |
108 | builder.createOrFold<cf::AssertOp>(loc, cmpOp, msg); |
109 | } |
110 | } |
111 | } |
112 | }; |
113 | |
114 | template <typename... OpTs> |
115 | void attachInterface(MLIRContext *ctx) { |
116 | (OpTs::template attachInterface<StructuredOpInterface<OpTs>>(*ctx), ...); |
117 | } |
118 | } // namespace |
119 | } // namespace linalg |
120 | } // namespace mlir |
121 | |
122 | void mlir::linalg::registerRuntimeVerifiableOpInterfaceExternalModels( |
123 | DialectRegistry ®istry) { |
124 | registry.addExtension(+[](MLIRContext *ctx, LinalgDialect *) { |
125 | attachInterface< |
126 | #define GET_OP_LIST |
127 | #include "mlir/Dialect/Linalg/IR/LinalgStructuredOps.cpp.inc" |
128 | >(ctx); |
129 | |
130 | // Load additional dialects of which ops may get created. |
131 | ctx->loadDialect<affine::AffineDialect, arith::ArithDialect, |
132 | cf::ControlFlowDialect, index::IndexDialect, |
133 | tensor::TensorDialect>(); |
134 | }); |
135 | } |
136 | |