brintos

brintos / llvm-project-archived public Read only

0
0
Text · 2.2 KiB · c816356 Raw
61 lines · cpp
1//===- SCFToGPUPass.cpp - Convert a loop nest to a GPU kernel -----------===//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-exception6//7//===----------------------------------------------------------------------===//8 9#include "mlir/Conversion/SCFToGPU/SCFToGPUPass.h"10 11#include "mlir/Conversion/SCFToGPU/SCFToGPU.h"12#include "mlir/Dialect/Affine/IR/AffineOps.h"13#include "mlir/Dialect/GPU/IR/GPUDialect.h"14#include "mlir/Transforms/DialectConversion.h"15 16namespace mlir {17#define GEN_PASS_DEF_CONVERTAFFINEFORTOGPUPASS18#define GEN_PASS_DEF_CONVERTPARALLELLOOPTOGPUPASS19#include "mlir/Conversion/Passes.h.inc"20} // namespace mlir21 22using namespace mlir;23using namespace mlir::scf;24 25namespace {26// A pass that traverses top-level loops in the function and converts them to27// GPU launch operations.  Nested launches are not allowed, so this does not28// walk the function recursively to avoid considering nested loops.29struct ForLoopMapper30    : public impl::ConvertAffineForToGPUPassBase<ForLoopMapper> {31  using Base::Base;32 33  void runOnOperation() override {34    for (Operation &op : llvm::make_early_inc_range(35             getOperation().getFunctionBody().getOps())) {36      if (auto forOp = dyn_cast<affine::AffineForOp>(&op)) {37        if (failed(convertAffineLoopNestToGPULaunch(forOp, numBlockDims,38                                                    numThreadDims)))39          signalPassFailure();40      }41    }42  }43};44 45struct ParallelLoopToGpuPass46    : public impl::ConvertParallelLoopToGpuPassBase<ParallelLoopToGpuPass> {47  void runOnOperation() override {48    RewritePatternSet patterns(&getContext());49    populateParallelLoopToGPUPatterns(patterns);50    ConversionTarget target(getContext());51    target.markUnknownOpDynamicallyLegal([](Operation *) { return true; });52    configureParallelLoopToGPULegality(target);53    if (failed(applyPartialConversion(getOperation(), target,54                                      std::move(patterns))))55      signalPassFailure();56    finalizeParallelLoopToGPUConversion(getOperation());57  }58};59 60} // namespace61