shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity
llvm-project/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp
Krzysztof Drewniak 51b65d0895 [mlir][AMDGPU] Improve BF16 handling through AMDGPU compilation 
Many previous sets of AMDGPU dialect code have been incorrect in the
presence of the bf16 type (when lowered to LLVM's bfloat) as they were
developed in a setting that run a custom bf16-to-i16 pass before LLVM
lowering.

An overall effect of this patch is that you should run
--arith-emulate-unsupported-floats="source-types=bf16 target-type=f32"
on your GPU module before calling --convert-gpu-to-rocdl if your code
performs bf16 arithmetic.

While LLVM now supports software bfloat, initial experiments showed
that using this support on AMDGPU inserted a large number of
conversions around loads and stores which had substantial performance
imparts. Furthermore, all of the native AMDGPU operations on bf16
types (like the WMMA operations) operate on 16-bit integers instead of
the bfloat type.

First, we make the following changes to preserve compatibility once
the LLVM bfloat type is reenabled.
1. The matrix multiplication operations (MFMA and WMMA) will bitcast
bfloat vectors to i16 vectors.
2. Buffer loads and stores will operate on the relevant integer
datatype and then cast to bfloat if needed.

Second, we add type conversions to convert bf16 and vectors of it to
equivalent i16 types.

Third, we add the bfloat <-> f32 expansion patterns to the set of
operations run before the main LLVM conversion so that MLIR's
implementation of these conversion routines is used.

Finally, we extend the "floats treated as integers" support in the
LLVM exporter to handle types other than fp8.

We also fix a bug in the unsupported floats emulation where it tried
to operate on `arith.bitcast` due to an oversight.

Reviewed By: rsuderman

Differential Revision: https://reviews.llvm.org/D156361
2023-08-17 18:31:28 +00:00

335 lines
15 KiB
C++
Raw Blame History

//===- LowerGpuOpsToROCDLOps.cpp - MLIR GPU to ROCDL 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 ROCDLIR operations for higher-level
// GPU operations.
//
//===----------------------------------------------------------------------===//
#include "mlir/Conversion/ControlFlowToLLVM/ControlFlowToLLVM.h"
#include "mlir/Conversion/GPUToROCDL/GPUToROCDLPass.h"
#include "mlir/Dialect/Arith/Transforms/Passes.h"
#include "mlir/Pass/Pass.h"
#include "mlir/Pass/PassManager.h"
#include "mlir/Transforms/Passes.h"
#include "mlir/Conversion/AMDGPUToROCDL/AMDGPUToROCDL.h"
#include "mlir/Conversion/ArithToLLVM/ArithToLLVM.h"
#include "mlir/Conversion/FuncToLLVM/ConvertFuncToLLVM.h"
#include "mlir/Conversion/GPUCommon/GPUCommonPass.h"
#include "mlir/Conversion/LLVMCommon/ConversionTarget.h"
#include "mlir/Conversion/LLVMCommon/LoweringOptions.h"
#include "mlir/Conversion/LLVMCommon/Pattern.h"
#include "mlir/Conversion/LLVMCommon/TypeConverter.h"
#include "mlir/Conversion/MemRefToLLVM/MemRefToLLVM.h"
#include "mlir/Conversion/VectorToLLVM/ConvertVectorToLLVM.h"
#include "mlir/Dialect/ControlFlow/IR/ControlFlow.h"
#include "mlir/Dialect/Func/IR/FuncOps.h"
#include "mlir/Dialect/GPU/IR/GPUDialect.h"
#include "mlir/Dialect/GPU/Transforms/Passes.h"
#include "mlir/Dialect/LLVMIR/LLVMDialect.h"
#include "mlir/Dialect/LLVMIR/ROCDLDialect.h"
#include "mlir/Dialect/Math/IR/Math.h"
#include "mlir/Dialect/MemRef/IR/MemRef.h"
#include "mlir/Dialect/Vector/IR/VectorOps.h"
#include "mlir/IR/BuiltinAttributes.h"
#include "mlir/Pass/Pass.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"
namespace mlir {
#define GEN_PASS_DEF_CONVERTGPUOPSTOROCDLOPS
#include "mlir/Conversion/Passes.h.inc"
} // namespace mlir
using namespace mlir;
/// Returns true if the given `gpu.func` can be safely called using the bare
/// pointer calling convention.
static bool canBeCalledWithBarePointers(gpu::GPUFuncOp func) {
bool canBeBare = true;
for (Type type : func.getArgumentTypes())
if (auto memrefTy = dyn_cast<BaseMemRefType>(type))
canBeBare &= LLVMTypeConverter::canConvertToBarePtr(memrefTy);
return canBeBare;
}
namespace {
struct GPULaneIdOpToROCDL : ConvertOpToLLVMPattern<gpu::LaneIdOp> {
using ConvertOpToLLVMPattern<gpu::LaneIdOp>::ConvertOpToLLVMPattern;
LogicalResult
matchAndRewrite(gpu::LaneIdOp op, gpu::LaneIdOp::Adaptor adaptor,
ConversionPatternRewriter &rewriter) const override {
auto loc = op->getLoc();
MLIRContext *context = rewriter.getContext();
// convert to: %mlo = call @llvm.amdgcn.mbcnt.lo(-1, 0)
// followed by: %lid = call @llvm.amdgcn.mbcnt.hi(-1, %mlo)
Type intTy = IntegerType::get(context, 32);
Value zero = rewriter.createOrFold<arith::ConstantIntOp>(loc, 0, 32);
Value minus1 = rewriter.createOrFold<arith::ConstantIntOp>(loc, -1, 32);
Value mbcntLo =
rewriter.create<ROCDL::MbcntLoOp>(loc, intTy, ValueRange{minus1, zero});
Value laneId = rewriter.create<ROCDL::MbcntHiOp>(
loc, intTy, ValueRange{minus1, mbcntLo});
// Truncate or extend the result depending on the index bitwidth specified
// by the LLVMTypeConverter options.
const unsigned indexBitwidth = getTypeConverter()->getIndexTypeBitwidth();
if (indexBitwidth > 32) {
laneId = rewriter.create<LLVM::SExtOp>(
loc, IntegerType::get(context, indexBitwidth), laneId);
} else if (indexBitwidth < 32) {
laneId = rewriter.create<LLVM::TruncOp>(
loc, IntegerType::get(context, indexBitwidth), laneId);
}
rewriter.replaceOp(op, {laneId});
return success();
}
};
/// Import the GPU Ops to ROCDL Patterns.
#include "GPUToROCDL.cpp.inc"
// A pass that replaces all occurrences of GPU device operations with their
// corresponding ROCDL equivalent.
//
// This pass only handles device code and is not meant to be run on GPU host
// code.
struct LowerGpuOpsToROCDLOpsPass
: public impl::ConvertGpuOpsToROCDLOpsBase<LowerGpuOpsToROCDLOpsPass> {
LowerGpuOpsToROCDLOpsPass() = default;
LowerGpuOpsToROCDLOpsPass(const std::string &chipset, unsigned indexBitwidth,
bool useBarePtrCallConv,
gpu::amd::Runtime runtime) {
if (this->chipset.getNumOccurrences() == 0)
this->chipset = chipset;
if (this->indexBitwidth.getNumOccurrences() == 0)
this->indexBitwidth = indexBitwidth;
if (this->useBarePtrCallConv.getNumOccurrences() == 0)
this->useBarePtrCallConv = useBarePtrCallConv;
if (this->runtime.getNumOccurrences() == 0)
this->runtime = runtime;
}
void runOnOperation() override {
gpu::GPUModuleOp m = getOperation();
MLIRContext *ctx = m.getContext();
// Request C wrapper emission.
for (auto func : m.getOps<func::FuncOp>()) {
func->setAttr(LLVM::LLVMDialect::getEmitCWrapperAttrName(),
UnitAttr::get(ctx));
}
FailureOr<amdgpu::Chipset> maybeChipset = amdgpu::Chipset::parse(chipset);
if (failed(maybeChipset)) {
emitError(UnknownLoc::get(ctx), "Invalid chipset name: " + chipset);
return signalPassFailure();
}
/// Customize the bitwidth used for the device side index computations.
LowerToLLVMOptions options(
ctx, DataLayout(cast<DataLayoutOpInterface>(m.getOperation())));
if (indexBitwidth != kDeriveIndexBitwidthFromDataLayout)
options.overrideIndexBitwidth(indexBitwidth);
options.useOpaquePointers = useOpaquePointers;
if (useBarePtrCallConv) {
options.useBarePtrCallConv = true;
WalkResult canUseBarePointers =
m.walk([](gpu::GPUFuncOp func) -> WalkResult {
if (canBeCalledWithBarePointers(func))
return WalkResult::advance();
return WalkResult::interrupt();
});
if (canUseBarePointers.wasInterrupted()) {
emitError(UnknownLoc::get(ctx),
"bare pointer calling convention requires all memrefs to "
"have static shape and use the identity map");
return signalPassFailure();
}
}
// Apply in-dialect lowering. In-dialect lowering will replace
// ops which need to be lowered further, which is not supported by a
// single conversion pass.
{
RewritePatternSet patterns(ctx);
populateGpuRewritePatterns(patterns);
arith::populateExpandBFloat16Patterns(patterns);
(void)applyPatternsAndFoldGreedily(m, std::move(patterns));
}
LLVMTypeConverter converter(ctx, options);
populateGpuMemorySpaceAttributeConversions(
converter, [](gpu::AddressSpace space) {
switch (space) {
case gpu::AddressSpace::Global:
return 1;
case gpu::AddressSpace::Workgroup:
return 3;
case gpu::AddressSpace::Private:
return 5;
}
llvm_unreachable("unknown address space enum value");
return 0;
});
RewritePatternSet llvmPatterns(ctx);
mlir::arith::populateArithToLLVMConversionPatterns(converter, llvmPatterns);
populateAMDGPUToROCDLConversionPatterns(converter, llvmPatterns,
*maybeChipset);
populateVectorToLLVMConversionPatterns(converter, llvmPatterns);
cf::populateControlFlowToLLVMConversionPatterns(converter, llvmPatterns);
populateFuncToLLVMConversionPatterns(converter, llvmPatterns);
populateFinalizeMemRefToLLVMConversionPatterns(converter, llvmPatterns);
populateGpuToROCDLConversionPatterns(converter, llvmPatterns, runtime);
LLVMConversionTarget target(getContext());
configureGpuToROCDLConversionLegality(target);
if (failed(applyPartialConversion(m, target, std::move(llvmPatterns))))
signalPassFailure();
// Manually rewrite known block size attributes so the LLVMIR translation
// infrastructure can pick them up.
m.walk([ctx](LLVM::LLVMFuncOp op) {
if (auto blockSizes = dyn_cast_or_null<DenseI32ArrayAttr>(
op->removeAttr(gpu::GPUFuncOp::getKnownBlockSizeAttrName()))) {
op->setAttr(ROCDL::ROCDLDialect::getReqdWorkGroupSizeAttrName(),
blockSizes);
// Also set up the rocdl.flat_work_group_size attribute to prevent
// conflicting metadata.
uint32_t flatSize = 1;
for (uint32_t size : blockSizes.asArrayRef()) {
flatSize *= size;
}
StringAttr flatSizeAttr =
StringAttr::get(ctx, Twine(flatSize) + "," + Twine(flatSize));
op->setAttr(ROCDL::ROCDLDialect::getFlatWorkGroupSizeAttrName(),
flatSizeAttr);
}
});
}
};
} // namespace
void mlir::configureGpuToROCDLConversionLegality(ConversionTarget &target) {
target.addIllegalOp<func::FuncOp>();
target.addLegalDialect<::mlir::LLVM::LLVMDialect>();
target.addLegalDialect<ROCDL::ROCDLDialect>();
target.addIllegalDialect<gpu::GPUDialect>();
target.addIllegalOp<LLVM::CosOp, LLVM::ExpOp, LLVM::Exp2Op, 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>();
}
template <typename OpTy>
static void populateOpPatterns(LLVMTypeConverter &converter,
RewritePatternSet &patterns, StringRef f32Func,
StringRef f64Func) {
patterns.add<ScalarizeVectorOpLowering<OpTy>>(converter);
patterns.add<OpToFuncCallLowering<OpTy>>(converter, f32Func, f64Func);
}
void mlir::populateGpuToROCDLConversionPatterns(
LLVMTypeConverter &converter, RewritePatternSet &patterns,
mlir::gpu::amd::Runtime runtime) {
using mlir::gpu::amd::Runtime;
populateWithGenerated(patterns);
patterns
.add<GPUIndexIntrinsicOpLowering<gpu::ThreadIdOp, ROCDL::ThreadIdXOp,
ROCDL::ThreadIdYOp, ROCDL::ThreadIdZOp>>(
converter, gpu::GPUFuncOp::getKnownBlockSizeAttrName());
patterns.add<GPUIndexIntrinsicOpLowering<
gpu::BlockIdOp, ROCDL::BlockIdXOp, ROCDL::BlockIdYOp, ROCDL::BlockIdZOp>>(
converter, gpu::GPUFuncOp::getKnownGridSizeAttrName());
patterns
.add<GPUIndexIntrinsicOpLowering<gpu::BlockDimOp, ROCDL::BlockDimXOp,
ROCDL::BlockDimYOp, ROCDL::BlockDimZOp>,
GPUIndexIntrinsicOpLowering<gpu::GridDimOp, ROCDL::GridDimXOp,
ROCDL::GridDimYOp, ROCDL::GridDimZOp>,
GPUReturnOpLowering>(converter);
patterns.add<GPUFuncOpLowering>(
converter,
/*allocaAddrSpace=*/ROCDL::ROCDLDialect::kPrivateMemoryAddressSpace,
/*workgroupAddrSpace=*/ROCDL::ROCDLDialect::kSharedMemoryAddressSpace,
StringAttr::get(&converter.getContext(),
ROCDL::ROCDLDialect::getKernelFuncAttrName()));
if (Runtime::HIP == runtime) {
patterns.add<GPUPrintfOpToHIPLowering>(converter);
} else if (Runtime::OpenCL == runtime) {
// Use address space = 4 to match the OpenCL definition of printf()
patterns.add<GPUPrintfOpToLLVMCallLowering>(converter, /*addressSpace=*/4);
}
patterns.add<GPULaneIdOpToROCDL>(converter);
populateOpPatterns<math::AbsFOp>(converter, patterns, "__ocml_fabs_f32",
"__ocml_fabs_f64");
populateOpPatterns<math::AtanOp>(converter, patterns, "__ocml_atan_f32",
"__ocml_atan_f64");
populateOpPatterns<math::Atan2Op>(converter, patterns, "__ocml_atan2_f32",
"__ocml_atan2_f64");
populateOpPatterns<math::CbrtOp>(converter, patterns, "__ocml_cbrt_f32",
"__ocml_cbrt_f64");
populateOpPatterns<math::CeilOp>(converter, patterns, "__ocml_ceil_f32",
"__ocml_ceil_f64");
populateOpPatterns<math::CosOp>(converter, patterns, "__ocml_cos_f32",
"__ocml_cos_f64");
populateOpPatterns<math::ExpOp>(converter, patterns, "__ocml_exp_f32",
"__ocml_exp_f64");
populateOpPatterns<math::Exp2Op>(converter, patterns, "__ocml_exp2_f32",
"__ocml_exp2_f64");
populateOpPatterns<math::ExpM1Op>(converter, patterns, "__ocml_expm1_f32",
"__ocml_expm1_f64");
populateOpPatterns<math::FloorOp>(converter, patterns, "__ocml_floor_f32",
"__ocml_floor_f64");
populateOpPatterns<math::LogOp>(converter, patterns, "__ocml_log_f32",
"__ocml_log_f64");
populateOpPatterns<math::Log10Op>(converter, patterns, "__ocml_log10_f32",
"__ocml_log10_f64");
populateOpPatterns<math::Log1pOp>(converter, patterns, "__ocml_log1p_f32",
"__ocml_log1p_f64");
populateOpPatterns<math::Log2Op>(converter, patterns, "__ocml_log2_f32",
"__ocml_log2_f64");
populateOpPatterns<math::PowFOp>(converter, patterns, "__ocml_pow_f32",
"__ocml_pow_f64");
populateOpPatterns<math::RsqrtOp>(converter, patterns, "__ocml_rsqrt_f32",
"__ocml_rsqrt_f64");
populateOpPatterns<math::SinOp>(converter, patterns, "__ocml_sin_f32",
"__ocml_sin_f64");
populateOpPatterns<math::SqrtOp>(converter, patterns, "__ocml_sqrt_f32",
"__ocml_sqrt_f64");
populateOpPatterns<math::TanhOp>(converter, patterns, "__ocml_tanh_f32",
"__ocml_tanh_f64");
populateOpPatterns<math::TanOp>(converter, patterns, "__ocml_tan_f32",
"__ocml_tan_f64");
populateOpPatterns<math::ErfOp>(converter, patterns, "__ocml_erf_f32",
"__ocml_erf_f64");
}
std::unique_ptr<OperationPass<gpu::GPUModuleOp>>
mlir::createLowerGpuOpsToROCDLOpsPass(const std::string &chipset,
unsigned indexBitwidth,
bool useBarePtrCallConv,
gpu::amd::Runtime runtime) {
return std::make_unique<LowerGpuOpsToROCDLOpsPass>(
chipset, indexBitwidth, useBarePtrCallConv, runtime);
}
Reference in New Issue View Git Blame Copy Permalink