Files
clang-p2996/mlir/lib/Conversion/StandardToSPIRV/LegalizeStandardForSPIRV.cpp
Nicolas Vasilache a8de412f51 [mlir] NFC - Expose an OffsetSizeAndStrideOpInterface
This revision will make it easier to create new ops base on the strided memref abstraction outside of the std dialect.

OffsetSizeAndStrideOpInterface is an interface for ops that allow specifying mixed dynamic and static offsets, sizes and strides variadic operands.
    Ops that implement this interface need to expose the following methods:
      1. `getArrayAttrRanks` to specify the length of static integer
          attributes.
      2. `offsets`, `sizes` and `strides` variadic operands.
      3. `static_offsets`, resp. `static_sizes` and `static_strides` integer
          array attributes.

    The invariants of this interface are:
      1. `static_offsets`, `static_sizes` and `static_strides` have length
          exactly `getArrayAttrRanks()`[0] (resp. [1], [2]).
      2. `offsets`, `sizes` and `strides` have each length at most
         `getArrayAttrRanks()`[0] (resp. [1], [2]).
      3. if an entry of `static_offsets` (resp. `static_sizes`,
         `static_strides`) is equal to a special sentinel value, namely
         `ShapedType::kDynamicStrideOrOffset` (resp. `ShapedType::kDynamicSize`,
         `ShapedType::kDynamicStrideOrOffset`), then the corresponding entry is
         a dynamic offset (resp. size, stride).
      4. a variadic `offset` (resp. `sizes`, `strides`) operand  must be present
         for each dynamic offset (resp. size, stride).

    This interface is useful to factor out common behavior and provide support
    for carrying or injecting static behavior through the use of the static
    attributes.

Differential Revision: https://reviews.llvm.org/D92011
2020-11-24 14:42:47 +00:00

212 lines
8.3 KiB
C++

//===- LegalizeStandardForSPIRV.cpp - Legalize ops for SPIR-V lowering ----===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// This transformation pass legalizes operations before the conversion to SPIR-V
// dialect to handle ops that cannot be lowered directly.
//
//===----------------------------------------------------------------------===//
#include "../PassDetail.h"
#include "mlir/Conversion/StandardToSPIRV/ConvertStandardToSPIRV.h"
#include "mlir/Conversion/StandardToSPIRV/ConvertStandardToSPIRVPass.h"
#include "mlir/Dialect/SPIRV/SPIRVDialect.h"
#include "mlir/Dialect/StandardOps/IR/Ops.h"
#include "mlir/Dialect/Vector/VectorOps.h"
#include "mlir/IR/StandardTypes.h"
#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
using namespace mlir;
namespace {
/// Merges subview operation with load/transferRead operation.
template <typename OpTy>
class LoadOpOfSubViewFolder final : public OpRewritePattern<OpTy> {
public:
using OpRewritePattern<OpTy>::OpRewritePattern;
LogicalResult matchAndRewrite(OpTy loadOp,
PatternRewriter &rewriter) const override;
private:
void replaceOp(OpTy loadOp, SubViewOp subViewOp,
ArrayRef<Value> sourceIndices,
PatternRewriter &rewriter) const;
};
/// Merges subview operation with store/transferWriteOp operation.
template <typename OpTy>
class StoreOpOfSubViewFolder final : public OpRewritePattern<OpTy> {
public:
using OpRewritePattern<OpTy>::OpRewritePattern;
LogicalResult matchAndRewrite(OpTy storeOp,
PatternRewriter &rewriter) const override;
private:
void replaceOp(OpTy StoreOp, SubViewOp subViewOp,
ArrayRef<Value> sourceIndices,
PatternRewriter &rewriter) const;
};
template <>
void LoadOpOfSubViewFolder<LoadOp>::replaceOp(LoadOp loadOp,
SubViewOp subViewOp,
ArrayRef<Value> sourceIndices,
PatternRewriter &rewriter) const {
rewriter.replaceOpWithNewOp<LoadOp>(loadOp, subViewOp.source(),
sourceIndices);
}
template <>
void LoadOpOfSubViewFolder<vector::TransferReadOp>::replaceOp(
vector::TransferReadOp loadOp, SubViewOp subViewOp,
ArrayRef<Value> sourceIndices, PatternRewriter &rewriter) const {
rewriter.replaceOpWithNewOp<vector::TransferReadOp>(
loadOp, loadOp.getVectorType(), subViewOp.source(), sourceIndices,
loadOp.permutation_map(), loadOp.padding(), loadOp.maskedAttr());
}
template <>
void StoreOpOfSubViewFolder<StoreOp>::replaceOp(
StoreOp storeOp, SubViewOp subViewOp, ArrayRef<Value> sourceIndices,
PatternRewriter &rewriter) const {
rewriter.replaceOpWithNewOp<StoreOp>(storeOp, storeOp.value(),
subViewOp.source(), sourceIndices);
}
template <>
void StoreOpOfSubViewFolder<vector::TransferWriteOp>::replaceOp(
vector::TransferWriteOp tranferWriteOp, SubViewOp subViewOp,
ArrayRef<Value> sourceIndices, PatternRewriter &rewriter) const {
rewriter.replaceOpWithNewOp<vector::TransferWriteOp>(
tranferWriteOp, tranferWriteOp.vector(), subViewOp.source(),
sourceIndices, tranferWriteOp.permutation_map(),
tranferWriteOp.maskedAttr());
}
} // namespace
//===----------------------------------------------------------------------===//
// Utility functions for op legalization.
//===----------------------------------------------------------------------===//
/// Given the 'indices' of an load/store operation where the memref is a result
/// of a subview op, returns the indices w.r.t to the source memref of the
/// subview op. For example
///
/// %0 = ... : memref<12x42xf32>
/// %1 = subview %0[%arg0, %arg1][][%stride1, %stride2] : memref<12x42xf32> to
/// memref<4x4xf32, offset=?, strides=[?, ?]>
/// %2 = load %1[%i1, %i2] : memref<4x4xf32, offset=?, strides=[?, ?]>
///
/// could be folded into
///
/// %2 = load %0[%arg0 + %i1 * %stride1][%arg1 + %i2 * %stride2] :
/// memref<12x42xf32>
static LogicalResult
resolveSourceIndices(Location loc, PatternRewriter &rewriter,
SubViewOp subViewOp, ValueRange indices,
SmallVectorImpl<Value> &sourceIndices) {
// TODO: Aborting when the offsets are static. There might be a way to fold
// the subview op with load even if the offsets have been canonicalized
// away.
SmallVector<Range, 4> opRanges = subViewOp.getOrCreateRanges(rewriter, loc);
auto opOffsets = llvm::map_range(opRanges, [](Range r) { return r.offset; });
auto opStrides = llvm::map_range(opRanges, [](Range r) { return r.stride; });
assert(opRanges.size() == indices.size() &&
"expected as many indices as rank of subview op result type");
// New indices for the load are the current indices * subview_stride +
// subview_offset.
sourceIndices.resize(indices.size());
for (auto index : llvm::enumerate(indices)) {
auto offset = *(opOffsets.begin() + index.index());
auto stride = *(opStrides.begin() + index.index());
auto mul = rewriter.create<MulIOp>(loc, index.value(), stride);
sourceIndices[index.index()] =
rewriter.create<AddIOp>(loc, offset, mul).getResult();
}
return success();
}
//===----------------------------------------------------------------------===//
// Folding SubViewOp and LoadOp/TransferReadOp.
//===----------------------------------------------------------------------===//
template <typename OpTy>
LogicalResult
LoadOpOfSubViewFolder<OpTy>::matchAndRewrite(OpTy loadOp,
PatternRewriter &rewriter) const {
auto subViewOp = loadOp.memref().template getDefiningOp<SubViewOp>();
if (!subViewOp) {
return failure();
}
SmallVector<Value, 4> sourceIndices;
if (failed(resolveSourceIndices(loadOp.getLoc(), rewriter, subViewOp,
loadOp.indices(), sourceIndices)))
return failure();
replaceOp(loadOp, subViewOp, sourceIndices, rewriter);
return success();
}
//===----------------------------------------------------------------------===//
// Folding SubViewOp and StoreOp/TransferWriteOp.
//===----------------------------------------------------------------------===//
template <typename OpTy>
LogicalResult
StoreOpOfSubViewFolder<OpTy>::matchAndRewrite(OpTy storeOp,
PatternRewriter &rewriter) const {
auto subViewOp = storeOp.memref().template getDefiningOp<SubViewOp>();
if (!subViewOp) {
return failure();
}
SmallVector<Value, 4> sourceIndices;
if (failed(resolveSourceIndices(storeOp.getLoc(), rewriter, subViewOp,
storeOp.indices(), sourceIndices)))
return failure();
replaceOp(storeOp, subViewOp, sourceIndices, rewriter);
return success();
}
//===----------------------------------------------------------------------===//
// Hook for adding patterns.
//===----------------------------------------------------------------------===//
void mlir::populateStdLegalizationPatternsForSPIRVLowering(
MLIRContext *context, OwningRewritePatternList &patterns) {
patterns.insert<LoadOpOfSubViewFolder<LoadOp>,
LoadOpOfSubViewFolder<vector::TransferReadOp>,
StoreOpOfSubViewFolder<StoreOp>,
StoreOpOfSubViewFolder<vector::TransferWriteOp>>(context);
}
//===----------------------------------------------------------------------===//
// Pass for testing just the legalization patterns.
//===----------------------------------------------------------------------===//
namespace {
struct SPIRVLegalization final
: public LegalizeStandardForSPIRVBase<SPIRVLegalization> {
void runOnOperation() override;
};
} // namespace
void SPIRVLegalization::runOnOperation() {
OwningRewritePatternList patterns;
auto *context = &getContext();
populateStdLegalizationPatternsForSPIRVLowering(context, patterns);
applyPatternsAndFoldGreedily(getOperation()->getRegions(),
std::move(patterns));
}
std::unique_ptr<Pass> mlir::createLegalizeStdOpsForSPIRVLoweringPass() {
return std::make_unique<SPIRVLegalization>();
}