Files
clang-p2996/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
thomasraoux b44007bec2 [mlir][gpu] Relax restriction on MMA store op to allow chain of mma ops.
In order to allow large matmul operations using the MMA ops we need to chain
operations this is not possible unless "DOp" and "COp" type have matching
layout so remove the "DOp" layout and force accumulator and result type to
match.
Added a test for the case where the MMA value is accumulated.

Differential Revision: https://reviews.llvm.org/D103023
2021-05-27 09:13:51 -07:00

255 lines
12 KiB
C++

//===- LowerGpuOpsToNVVMOps.cpp - MLIR GPU to NVVM lowering passes --------===//
//
// 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 file implements a pass to generate NVVMIR operations for higher-level
// GPU operations.
//
//===----------------------------------------------------------------------===//
#include "mlir/Conversion/GPUToNVVM/GPUToNVVMPass.h"
#include "mlir/Conversion/StandardToLLVM/ConvertStandardToLLVMPass.h"
#include "mlir/Dialect/GPU/GPUDialect.h"
#include "mlir/Dialect/GPU/Passes.h"
#include "mlir/Dialect/LLVMIR/NVVMDialect.h"
#include "mlir/Dialect/Math/IR/Math.h"
#include "mlir/IR/BlockAndValueMapping.h"
#include "mlir/Transforms/DialectConversion.h"
#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
#include "llvm/Support/FormatVariadic.h"
#include "../GPUCommon/GPUOpsLowering.h"
#include "../GPUCommon/IndexIntrinsicsOpLowering.h"
#include "../GPUCommon/OpToFuncCallLowering.h"
#include "../PassDetail.h"
using namespace mlir;
namespace {
struct GPUShuffleOpLowering : public ConvertOpToLLVMPattern<gpu::ShuffleOp> {
using ConvertOpToLLVMPattern<gpu::ShuffleOp>::ConvertOpToLLVMPattern;
/// Lowers a shuffle to the corresponding NVVM op.
///
/// Convert the `width` argument into an activeMask (a bitmask which specifies
/// which threads participate in the shuffle) and a maskAndClamp (specifying
/// the highest lane which participates in the shuffle).
///
/// %one = llvm.constant(1 : i32) : i32
/// %shl = llvm.shl %one, %width : i32
/// %active_mask = llvm.sub %shl, %one : i32
/// %mask_and_clamp = llvm.sub %width, %one : i32
/// %shfl = nvvm.shfl.sync.bfly %active_mask, %value, %offset,
/// %mask_and_clamp : !llvm<"{ float, i1 }">
/// %shfl_value = llvm.extractvalue %shfl[0 : index] :
/// !llvm<"{ float, i1 }">
/// %shfl_pred = llvm.extractvalue %shfl[1 : index] :
/// !llvm<"{ float, i1 }">
LogicalResult
matchAndRewrite(gpu::ShuffleOp op, ArrayRef<Value> operands,
ConversionPatternRewriter &rewriter) const override {
Location loc = op->getLoc();
gpu::ShuffleOpAdaptor adaptor(operands);
auto valueTy = adaptor.value().getType();
auto int32Type = IntegerType::get(rewriter.getContext(), 32);
auto predTy = IntegerType::get(rewriter.getContext(), 1);
auto resultTy = LLVM::LLVMStructType::getLiteral(rewriter.getContext(),
{valueTy, predTy});
Value one = rewriter.create<LLVM::ConstantOp>(
loc, int32Type, rewriter.getI32IntegerAttr(1));
// Bit mask of active lanes: `(1 << activeWidth) - 1`.
Value activeMask = rewriter.create<LLVM::SubOp>(
loc, int32Type,
rewriter.create<LLVM::ShlOp>(loc, int32Type, one, adaptor.width()),
one);
// Clamp lane: `activeWidth - 1`
Value maskAndClamp =
rewriter.create<LLVM::SubOp>(loc, int32Type, adaptor.width(), one);
auto returnValueAndIsValidAttr = rewriter.getUnitAttr();
Value shfl = rewriter.create<NVVM::ShflBflyOp>(
loc, resultTy, activeMask, adaptor.value(), adaptor.offset(),
maskAndClamp, returnValueAndIsValidAttr);
Value shflValue = rewriter.create<LLVM::ExtractValueOp>(
loc, valueTy, shfl, rewriter.getIndexArrayAttr(0));
Value isActiveSrcLane = rewriter.create<LLVM::ExtractValueOp>(
loc, predTy, shfl, rewriter.getIndexArrayAttr(1));
rewriter.replaceOp(op, {shflValue, isActiveSrcLane});
return success();
}
};
/// Import the GPU Ops to NVVM Patterns.
#include "GPUToNVVM.cpp.inc"
/// A pass that replaces all occurrences of GPU device operations with their
/// corresponding NVVM equivalent.
///
/// This pass only handles device code and is not meant to be run on GPU host
/// code.
struct LowerGpuOpsToNVVMOpsPass
: public ConvertGpuOpsToNVVMOpsBase<LowerGpuOpsToNVVMOpsPass> {
LowerGpuOpsToNVVMOpsPass() = default;
LowerGpuOpsToNVVMOpsPass(unsigned indexBitwidth) {
this->indexBitwidth = indexBitwidth;
}
void runOnOperation() override {
gpu::GPUModuleOp m = getOperation();
/// Customize the bitwidth used for the device side index computations.
LowerToLLVMOptions options(
m.getContext(),
DataLayout(cast<DataLayoutOpInterface>(m.getOperation())));
options.emitCWrappers = true;
if (indexBitwidth != kDeriveIndexBitwidthFromDataLayout)
options.overrideIndexBitwidth(indexBitwidth);
/// MemRef conversion for GPU to NVVM lowering. The GPU dialect uses memory
/// space 5 for private memory attributions, but NVVM represents private
/// memory allocations as local `alloca`s in the default address space. This
/// converter drops the private memory space to support the use case above.
LLVMTypeConverter converter(m.getContext(), options);
converter.addConversion([&](MemRefType type) -> Optional<Type> {
if (type.getMemorySpaceAsInt() !=
gpu::GPUDialect::getPrivateAddressSpace())
return llvm::None;
return converter.convertType(MemRefType::Builder(type).setMemorySpace(0));
});
// Lowering for MMAMatrixType.
converter.addConversion([&](gpu::MMAMatrixType type) -> Type {
// The number of items in structToReturn are dependent on the the dataType
// and the MMA operand that this operation is associated with.
llvm::DenseMap<StringRef, int64_t> numElemsPerThreadF16,
numElemsPerThreadF32;
numElemsPerThreadF16["AOp"] = 8;
numElemsPerThreadF16["BOp"] = 8;
numElemsPerThreadF16["COp"] = 4;
numElemsPerThreadF32["AOp"] = 8;
numElemsPerThreadF32["BOp"] = 8;
numElemsPerThreadF32["COp"] = 8;
Type structToReturn;
if (type.getElementType().isF16()) {
// Number of f16's in 32-bit.
unsigned vecSize = 2;
Type vec = VectorType::get(vecSize, FloatType::getF16(&getContext()));
unsigned size = numElemsPerThreadF16[type.getOperand()];
SmallVector<Type> elements(size, vec);
structToReturn =
LLVM::LLVMStructType::getLiteral(&getContext(), elements);
} else if (type.getElementType().isF32()) {
unsigned size = numElemsPerThreadF32[type.getOperand()];
SmallVector<Type> elements(size, FloatType::getF32(&getContext()));
structToReturn =
LLVM::LLVMStructType::getLiteral(&getContext(), elements);
}
return structToReturn;
});
RewritePatternSet patterns(m.getContext());
RewritePatternSet llvmPatterns(m.getContext());
// Apply in-dialect lowering first. In-dialect lowering will replace ops
// which need to be lowered further, which is not supported by a single
// conversion pass.
populateGpuRewritePatterns(patterns);
(void)applyPatternsAndFoldGreedily(m, std::move(patterns));
populateStdToLLVMConversionPatterns(converter, llvmPatterns);
populateGpuToNVVMConversionPatterns(converter, llvmPatterns);
populateGpuWMMAToNVVMConversionPatterns(converter, llvmPatterns);
LLVMConversionTarget target(getContext());
configureGpuToNVVMConversionLegality(target);
if (failed(applyPartialConversion(m, target, std::move(llvmPatterns))))
signalPassFailure();
}
};
} // anonymous namespace
void mlir::configureGpuToNVVMConversionLegality(ConversionTarget &target) {
target.addIllegalOp<FuncOp>();
target.addLegalDialect<::mlir::LLVM::LLVMDialect>();
target.addLegalDialect<::mlir::NVVM::NVVMDialect>();
target.addIllegalDialect<gpu::GPUDialect>();
target.addIllegalOp<LLVM::CosOp, LLVM::ExpOp, LLVM::FAbsOp, LLVM::FCeilOp,
LLVM::FFloorOp, LLVM::LogOp, LLVM::Log10Op, LLVM::Log2Op,
LLVM::PowOp, LLVM::SinOp, LLVM::SqrtOp>();
// TODO: Remove once we support replacing non-root ops.
target.addLegalOp<gpu::YieldOp, gpu::GPUModuleOp, gpu::ModuleEndOp>();
}
void mlir::populateGpuToNVVMConversionPatterns(LLVMTypeConverter &converter,
RewritePatternSet &patterns) {
populateWithGenerated(patterns);
patterns
.add<GPUIndexIntrinsicOpLowering<gpu::ThreadIdOp, NVVM::ThreadIdXOp,
NVVM::ThreadIdYOp, NVVM::ThreadIdZOp>,
GPUIndexIntrinsicOpLowering<gpu::BlockDimOp, NVVM::BlockDimXOp,
NVVM::BlockDimYOp, NVVM::BlockDimZOp>,
GPUIndexIntrinsicOpLowering<gpu::BlockIdOp, NVVM::BlockIdXOp,
NVVM::BlockIdYOp, NVVM::BlockIdZOp>,
GPUIndexIntrinsicOpLowering<gpu::GridDimOp, NVVM::GridDimXOp,
NVVM::GridDimYOp, NVVM::GridDimZOp>,
GPUShuffleOpLowering, GPUReturnOpLowering>(converter);
// Explicitly drop memory space when lowering private memory
// attributions since NVVM models it as `alloca`s in the default
// memory space and does not support `alloca`s with addrspace(5).
patterns.add<GPUFuncOpLowering>(
converter, /*allocaAddrSpace=*/0,
Identifier::get(NVVM::NVVMDialect::getKernelFuncAttrName(),
&converter.getContext()));
patterns.add<OpToFuncCallLowering<AbsFOp>>(converter, "__nv_fabsf",
"__nv_fabs");
patterns.add<OpToFuncCallLowering<math::AtanOp>>(converter, "__nv_atanf",
"__nv_atan");
patterns.add<OpToFuncCallLowering<math::Atan2Op>>(converter, "__nv_atan2f",
"__nv_atan2");
patterns.add<OpToFuncCallLowering<CeilFOp>>(converter, "__nv_ceilf",
"__nv_ceil");
patterns.add<OpToFuncCallLowering<math::CosOp>>(converter, "__nv_cosf",
"__nv_cos");
patterns.add<OpToFuncCallLowering<math::ExpOp>>(converter, "__nv_expf",
"__nv_exp");
patterns.add<OpToFuncCallLowering<math::ExpM1Op>>(converter, "__nv_expm1f",
"__nv_expm1");
patterns.add<OpToFuncCallLowering<FloorFOp>>(converter, "__nv_floorf",
"__nv_floor");
patterns.add<OpToFuncCallLowering<math::LogOp>>(converter, "__nv_logf",
"__nv_log");
patterns.add<OpToFuncCallLowering<math::Log1pOp>>(converter, "__nv_log1pf",
"__nv_log1p");
patterns.add<OpToFuncCallLowering<math::Log10Op>>(converter, "__nv_log10f",
"__nv_log10");
patterns.add<OpToFuncCallLowering<math::Log2Op>>(converter, "__nv_log2f",
"__nv_log2");
patterns.add<OpToFuncCallLowering<math::PowFOp>>(converter, "__nv_powf",
"__nv_pow");
patterns.add<OpToFuncCallLowering<math::RsqrtOp>>(converter, "__nv_rsqrtf",
"__nv_rsqrt");
patterns.add<OpToFuncCallLowering<math::SinOp>>(converter, "__nv_sinf",
"__nv_sin");
patterns.add<OpToFuncCallLowering<math::SqrtOp>>(converter, "__nv_sqrtf",
"__nv_sqrt");
patterns.add<OpToFuncCallLowering<math::TanhOp>>(converter, "__nv_tanhf",
"__nv_tanh");
}
std::unique_ptr<OperationPass<gpu::GPUModuleOp>>
mlir::createLowerGpuOpsToNVVMOpsPass(unsigned indexBitwidth) {
return std::make_unique<LowerGpuOpsToNVVMOpsPass>(indexBitwidth);
}