llvm/mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp

//===- EmulateAtomics.cpp - Emulate unsupported AMDGPU atomics ------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//

#include "mlir/Dialect/AMDGPU/Transforms/Passes.h"

#include "mlir/Dialect/AMDGPU/IR/AMDGPUDialect.h"
#include "mlir/Dialect/Arith/IR/Arith.h"
#include "mlir/Dialect/ControlFlow/IR/ControlFlow.h"
#include "mlir/Dialect/ControlFlow/IR/ControlFlowOps.h"
#include "mlir/IR/BuiltinAttributes.h"
#include "mlir/Transforms/DialectConversion.h"
#include "mlir/Transforms/GreedyPatternRewriteDriver.h"

namespace mlir::amdgpu {
#define GEN_PASS_DEF_AMDGPUEMULATEATOMICSPASS
#include "mlir/Dialect/AMDGPU/Transforms/Passes.h.inc"
} // namespace mlir::amdgpu

usingnamespacemlir;
usingnamespacemlir::amdgpu;

namespace {
struct AmdgpuEmulateAtomicsPass
    : public amdgpu::impl::AmdgpuEmulateAtomicsPassBase<
          AmdgpuEmulateAtomicsPass> { … };

template <typename AtomicOp, typename ArithOp>
struct RawBufferAtomicByCasPattern : public OpConversionPattern<AtomicOp> { … };
} // namespace

namespace {
enum class DataArgAction : unsigned char { … };
} // namespace

// Fix up the fact that, when we're migrating from a general bugffer atomic
// to a load or to a CAS, the number of openrands, and thus the number of
// entries needed in operandSegmentSizes, needs to change. We use this method
// because we'd like to preserve unknown attributes on the atomic instead of
// discarding them.
static void patchOperandSegmentSizes(ArrayRef<NamedAttribute> attrs,
                                     SmallVectorImpl<NamedAttribute> &newAttrs,
                                     DataArgAction action) { … }

template <typename AtomicOp, typename ArithOp>
LogicalResult RawBufferAtomicByCasPattern<AtomicOp, ArithOp>::matchAndRewrite(
    AtomicOp atomicOp, Adaptor adaptor,
    ConversionPatternRewriter &rewriter) const { … }

void mlir::amdgpu::populateAmdgpuEmulateAtomicsPatterns(
    ConversionTarget &target, RewritePatternSet &patterns, Chipset chipset) { … }

void AmdgpuEmulateAtomicsPass::runOnOperation() { … }