flang/lib/Optimizer/Transforms/AffineDemotion.cpp

   1 //===-- AffineDemotion.cpp -----------------------------------------------===//
   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 transformation is a prototype that demote affine dialects operations
  10 // after optimizations to FIR loops operations.
  11 // It is used after the AffinePromotion pass.
  12 // It is not part of the production pipeline and would need more work in order
  13 // to be used in production.
  14 // More information can be found in this presentation:
  15 // https://slides.com/rajanwalia/deck
  16 //
  17 //===----------------------------------------------------------------------===//
  18
  19 #include "flang/Optimizer/Dialect/FIRDialect.h"
  20 #include "flang/Optimizer/Dialect/FIROps.h"
  21 #include "flang/Optimizer/Dialect/FIRType.h"
  22 #include "flang/Optimizer/Transforms/Passes.h"
  23 #include "mlir/Dialect/Affine/IR/AffineOps.h"
  24 #include "mlir/Dialect/Affine/Utils.h"
  25 #include "mlir/Dialect/Func/IR/FuncOps.h"
  26 #include "mlir/Dialect/MemRef/IR/MemRef.h"
  27 #include "mlir/Dialect/SCF/IR/SCF.h"
  28 #include "mlir/IR/BuiltinAttributes.h"
  29 #include "mlir/IR/IntegerSet.h"
  30 #include "mlir/IR/Visitors.h"
  31 #include "mlir/Pass/Pass.h"
  32 #include "mlir/Transforms/DialectConversion.h"
  33 #include "llvm/ADT/DenseMap.h"
  34 #include "llvm/Support/CommandLine.h"
  35 #include "llvm/Support/Debug.h"
  36
  37 namespace fir {
  38 #define GEN_PASS_DEF_AFFINEDIALECTDEMOTION
  39 #include "flang/Optimizer/Transforms/Passes.h.inc"
  40 } // namespace fir
  41
  42 #define DEBUG_TYPE "flang-affine-demotion"
  43
  44 using namespace fir;
  45 using namespace mlir;
  46
  47 namespace {
  48
  49 class AffineLoadConversion : public OpConversionPattern<mlir::AffineLoadOp> {
  50 public:
  51   using OpConversionPattern<mlir::AffineLoadOp>::OpConversionPattern;
  52
  53   LogicalResult
  54   matchAndRewrite(mlir::AffineLoadOp op, OpAdaptor adaptor,
  55                   ConversionPatternRewriter &rewriter) const override {
  56     SmallVector<Value> indices(adaptor.getIndices());
  57     auto maybeExpandedMap =
  58         expandAffineMap(rewriter, op.getLoc(), op.getAffineMap(), indices);
  59     if (!maybeExpandedMap)
  60       return failure();
  61
  62     auto coorOp = rewriter.create<fir::CoordinateOp>(
  63         op.getLoc(), fir::ReferenceType::get(op.getResult().getType()),
  64         adaptor.getMemref(), *maybeExpandedMap);
  65
  66     rewriter.replaceOpWithNewOp<fir::LoadOp>(op, coorOp.getResult());
  67     return success();
  68   }
  69 };
  70
  71 class AffineStoreConversion : public OpConversionPattern<mlir::AffineStoreOp> {
  72 public:
  73   using OpConversionPattern<mlir::AffineStoreOp>::OpConversionPattern;
  74
  75   LogicalResult
  76   matchAndRewrite(mlir::AffineStoreOp op, OpAdaptor adaptor,
  77                   ConversionPatternRewriter &rewriter) const override {
  78     SmallVector<Value> indices(op.getIndices());
  79     auto maybeExpandedMap =
  80         expandAffineMap(rewriter, op.getLoc(), op.getAffineMap(), indices);
  81     if (!maybeExpandedMap)
  82       return failure();
  83
  84     auto coorOp = rewriter.create<fir::CoordinateOp>(
  85         op.getLoc(), fir::ReferenceType::get(op.getValueToStore().getType()),
  86         adaptor.getMemref(), *maybeExpandedMap);
  87     rewriter.replaceOpWithNewOp<fir::StoreOp>(op, adaptor.getValue(),
  88                                               coorOp.getResult());
  89     return success();
  90   }
  91 };
  92
  93 class ConvertConversion : public mlir::OpRewritePattern<fir::ConvertOp> {
  94 public:
  95   using OpRewritePattern::OpRewritePattern;
  96   mlir::LogicalResult
  97   matchAndRewrite(fir::ConvertOp op,
  98                   mlir::PatternRewriter &rewriter) const override {
  99     if (op.getRes().getType().isa<mlir::MemRefType>()) {
 100       // due to index calculation moving to affine maps we still need to
 101       // add converts for sequence types this has a side effect of losing
 102       // some information about arrays with known dimensions by creating:
 103       // fir.convert %arg0 : (!fir.ref<!fir.array<5xi32>>) ->
 104       // !fir.ref<!fir.array<?xi32>>
 105       if (auto refTy = op.getValue().getType().dyn_cast<fir::ReferenceType>())
 106         if (auto arrTy = refTy.getEleTy().dyn_cast<fir::SequenceType>()) {
 107           fir::SequenceType::Shape flatShape = {
 108               fir::SequenceType::getUnknownExtent()};
 109           auto flatArrTy = fir::SequenceType::get(flatShape, arrTy.getEleTy());
 110           auto flatTy = fir::ReferenceType::get(flatArrTy);
 111           rewriter.replaceOpWithNewOp<fir::ConvertOp>(op, flatTy,
 112                                                       op.getValue());
 113           return success();
 114         }
 115       rewriter.startRootUpdate(op->getParentOp());
 116       op.getResult().replaceAllUsesWith(op.getValue());
 117       rewriter.finalizeRootUpdate(op->getParentOp());
 118       rewriter.eraseOp(op);
 119     }
 120     return success();
 121   }
 122 };
 123
 124 mlir::Type convertMemRef(mlir::MemRefType type) {
 125   return fir::SequenceType::get(
 126       SmallVector<int64_t>(type.getShape().begin(), type.getShape().end()),
 127       type.getElementType());
 128 }
 129
 130 class StdAllocConversion : public mlir::OpRewritePattern<memref::AllocOp> {
 131 public:
 132   using OpRewritePattern::OpRewritePattern;
 133   mlir::LogicalResult
 134   matchAndRewrite(memref::AllocOp op,
 135                   mlir::PatternRewriter &rewriter) const override {
 136     rewriter.replaceOpWithNewOp<fir::AllocaOp>(op, convertMemRef(op.getType()),
 137                                                op.getMemref());
 138     return success();
 139   }
 140 };
 141
 142 class AffineDialectDemotion
 143     : public fir::impl::AffineDialectDemotionBase<AffineDialectDemotion> {
 144 public:
 145   void runOnOperation() override {
 146     auto *context = &getContext();
 147     auto function = getOperation();
 148     LLVM_DEBUG(llvm::dbgs() << "AffineDemotion: running on function:\n";
 149                function.print(llvm::dbgs()););
 150
 151     mlir::RewritePatternSet patterns(context);
 152     patterns.insert<ConvertConversion>(context);
 153     patterns.insert<AffineLoadConversion>(context);
 154     patterns.insert<AffineStoreConversion>(context);
 155     patterns.insert<StdAllocConversion>(context);
 156     mlir::ConversionTarget target(*context);
 157     target.addIllegalOp<memref::AllocOp>();
 158     target.addDynamicallyLegalOp<fir::ConvertOp>([](fir::ConvertOp op) {
 159       if (op.getRes().getType().isa<mlir::MemRefType>())
 160         return false;
 161       return true;
 162     });
 163     target
 164         .addLegalDialect<FIROpsDialect, mlir::scf::SCFDialect,
 165                          mlir::arith::ArithDialect, mlir::func::FuncDialect>();
 166
 167     if (mlir::failed(mlir::applyPartialConversion(function, target,
 168                                                   std::move(patterns)))) {
 169       mlir::emitError(mlir::UnknownLoc::get(context),
 170                       "error in converting affine dialect\n");
 171       signalPassFailure();
 172     }
 173   }
 174 };
 175
 176 } // namespace
 177
 178 std::unique_ptr<mlir::Pass> fir::createAffineDemotionPass() {
 179   return std::make_unique<AffineDialectDemotion>();
 180 }