| 1 | //===- InlineElementals.cpp - Inline chained hlfir.elemental ops ----------===// |
| 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 | // Chained elemental operations like a + b + c can inline the first elemental |
| 9 | // at the hlfir.apply in the body of the second one (as described in |
| 10 | // docs/HighLevelFIR.md). This has to be done in a pass rather than in lowering |
| 11 | // so that it happens after the HLFIR intrinsic simplification pass. |
| 12 | //===----------------------------------------------------------------------===// |
| 13 | |
| 14 | #include "flang/Optimizer/Builder/FIRBuilder.h" |
| 15 | #include "flang/Optimizer/Builder/HLFIRTools.h" |
| 16 | #include "flang/Optimizer/Dialect/Support/FIRContext.h" |
| 17 | #include "flang/Optimizer/HLFIR/HLFIROps.h" |
| 18 | #include "flang/Optimizer/HLFIR/Passes.h" |
| 19 | #include "mlir/Dialect/Func/IR/FuncOps.h" |
| 20 | #include "mlir/IR/IRMapping.h" |
| 21 | #include "mlir/IR/PatternMatch.h" |
| 22 | #include "mlir/Pass/Pass.h" |
| 23 | #include "mlir/Support/LLVM.h" |
| 24 | #include "mlir/Transforms/DialectConversion.h" |
| 25 | #include "mlir/Transforms/GreedyPatternRewriteDriver.h" |
| 26 | #include "llvm/ADT/TypeSwitch.h" |
| 27 | #include <iterator> |
| 28 | |
| 29 | namespace hlfir { |
| 30 | #define GEN_PASS_DEF_INLINEELEMENTALS |
| 31 | #include "flang/Optimizer/HLFIR/Passes.h.inc" |
| 32 | } // namespace hlfir |
| 33 | |
| 34 | /// If the elemental has only two uses and those two are an apply operation and |
| 35 | /// a destroy operation, return those two, otherwise return {} |
| 36 | static std::optional<std::pair<hlfir::ApplyOp, hlfir::DestroyOp>> |
| 37 | getTwoUses(hlfir::ElementalOp elemental) { |
| 38 | mlir::Operation::user_range users = elemental->getUsers(); |
| 39 | // don't inline anything with more than one use (plus hfir.destroy) |
| 40 | if (std::distance(users.begin(), users.end()) != 2) { |
| 41 | return std::nullopt; |
| 42 | } |
| 43 | |
| 44 | // If the ElementalOp must produce a temporary (e.g. for |
| 45 | // finalization purposes), then we cannot inline it. |
| 46 | if (hlfir::elementalOpMustProduceTemp(elemental)) |
| 47 | return std::nullopt; |
| 48 | |
| 49 | hlfir::ApplyOp apply; |
| 50 | hlfir::DestroyOp destroy; |
| 51 | for (mlir::Operation *user : users) |
| 52 | mlir::TypeSwitch<mlir::Operation *, void>(user) |
| 53 | .Case([&](hlfir::ApplyOp op) { apply = op; }) |
| 54 | .Case([&](hlfir::DestroyOp op) { destroy = op; }); |
| 55 | |
| 56 | if (!apply || !destroy) |
| 57 | return std::nullopt; |
| 58 | |
| 59 | // we can't inline if the return type of the yield doesn't match the return |
| 60 | // type of the apply |
| 61 | auto yield = mlir::dyn_cast_or_null<hlfir::YieldElementOp>( |
| 62 | elemental.getRegion().back().back()); |
| 63 | assert(yield && "hlfir.elemental should always end with a yield" ); |
| 64 | if (apply.getResult().getType() != yield.getElementValue().getType()) |
| 65 | return std::nullopt; |
| 66 | |
| 67 | return std::pair{apply, destroy}; |
| 68 | } |
| 69 | |
| 70 | namespace { |
| 71 | class InlineElementalConversion |
| 72 | : public mlir::OpRewritePattern<hlfir::ElementalOp> { |
| 73 | public: |
| 74 | using mlir::OpRewritePattern<hlfir::ElementalOp>::OpRewritePattern; |
| 75 | |
| 76 | llvm::LogicalResult |
| 77 | matchAndRewrite(hlfir::ElementalOp elemental, |
| 78 | mlir::PatternRewriter &rewriter) const override { |
| 79 | std::optional<std::pair<hlfir::ApplyOp, hlfir::DestroyOp>> maybeTuple = |
| 80 | getTwoUses(elemental); |
| 81 | if (!maybeTuple) |
| 82 | return rewriter.notifyMatchFailure( |
| 83 | elemental, "hlfir.elemental does not have two uses" ); |
| 84 | |
| 85 | if (elemental.isOrdered()) { |
| 86 | // We can only inline the ordered elemental into a loop-like |
| 87 | // construct that processes the indices in-order and does not |
| 88 | // have the side effects itself. Adhere to conservative behavior |
| 89 | // for the time being. |
| 90 | return rewriter.notifyMatchFailure(elemental, |
| 91 | "hlfir.elemental is ordered" ); |
| 92 | } |
| 93 | auto [apply, destroy] = *maybeTuple; |
| 94 | |
| 95 | assert(elemental.getRegion().hasOneBlock() && |
| 96 | "expect elemental region to have one block" ); |
| 97 | |
| 98 | fir::FirOpBuilder builder{rewriter, elemental.getOperation()}; |
| 99 | builder.setInsertionPointAfter(apply); |
| 100 | hlfir::YieldElementOp yield = hlfir::inlineElementalOp( |
| 101 | elemental.getLoc(), builder, elemental, apply.getIndices()); |
| 102 | |
| 103 | // remove the old elemental and all of the bookkeeping |
| 104 | rewriter.replaceAllUsesWith(apply.getResult(), yield.getElementValue()); |
| 105 | rewriter.eraseOp(yield); |
| 106 | rewriter.eraseOp(apply); |
| 107 | rewriter.eraseOp(destroy); |
| 108 | rewriter.eraseOp(elemental); |
| 109 | |
| 110 | return mlir::success(); |
| 111 | } |
| 112 | }; |
| 113 | |
| 114 | class InlineElementalsPass |
| 115 | : public hlfir::impl::InlineElementalsBase<InlineElementalsPass> { |
| 116 | public: |
| 117 | void runOnOperation() override { |
| 118 | mlir::MLIRContext *context = &getContext(); |
| 119 | |
| 120 | mlir::GreedyRewriteConfig config; |
| 121 | // Prevent the pattern driver from merging blocks. |
| 122 | config.setRegionSimplificationLevel( |
| 123 | mlir::GreedySimplifyRegionLevel::Disabled); |
| 124 | |
| 125 | mlir::RewritePatternSet patterns(context); |
| 126 | patterns.insert<InlineElementalConversion>(context); |
| 127 | |
| 128 | if (mlir::failed(mlir::applyPatternsGreedily( |
| 129 | getOperation(), std::move(patterns), config))) { |
| 130 | mlir::emitError(getOperation()->getLoc(), |
| 131 | "failure in HLFIR elemental inlining" ); |
| 132 | signalPassFailure(); |
| 133 | } |
| 134 | } |
| 135 | }; |
| 136 | } // namespace |
| 137 | |