Files
llvm-project/mlir/lib/Target/SPIRV/Target.cpp
Zichen Lu fbffdaa174 [MLIR][GPU] Update serializeToObject to use SerializedObject wrapper and include ISA compiler logs (#176697)
This PR makes the compilation log from ISA compiler available to users
by returning it as part of the `gpu::ObjectAttr` properties, following
the existing pattern like `LLVMIRToISATimeInMs`.

Currently, the compiler log (which contains useful information such as
spill statistics when --verbose is passed) is only accessible in debug
builds via `LLVM_DEBUG`. However, there are good reasons to make this
information available in release builds as well:

1. Both `ptxas` and `libnvptxcompiler` are publicly available
tools/libraries distributed with the CUDA Toolkit. The `--verbose` flag
and its output are documented public features, not internal debug
information.
2. The verbose output provides valuable insights for users.

A new `SerializedObject` class is used to carry the metadata alongside
the binary when returning from `serializeObject`.
2026-01-30 12:56:20 +01:00

105 lines
3.7 KiB
C++

//===- Target.cpp - MLIR SPIR-V target compilation --------------*- C++ -*-===//
//
// 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 files defines SPIR-V target related functions including registration
// calls for the `#spirv.target_env` compilation attribute.
//
//===----------------------------------------------------------------------===//
#include "mlir/Target/SPIRV/Target.h"
#include "mlir/Dialect/GPU/IR/GPUDialect.h"
#include "mlir/Dialect/SPIRV/IR/SPIRVAttributes.h"
#include "mlir/Dialect/SPIRV/IR/SPIRVDialect.h"
#include "mlir/Dialect/SPIRV/IR/SPIRVOps.h"
#include "mlir/Target/SPIRV/Serialization.h"
#include <cstdlib>
#include <cstring>
using namespace mlir;
using namespace mlir::spirv;
namespace {
// SPIR-V implementation of the gpu:TargetAttrInterface.
class SPIRVTargetAttrImpl
: public gpu::TargetAttrInterface::FallbackModel<SPIRVTargetAttrImpl> {
public:
std::optional<mlir::gpu::SerializedObject>
serializeToObject(Attribute attribute, Operation *module,
const gpu::TargetOptions &options) const;
Attribute createObject(Attribute attribute, Operation *module,
const mlir::gpu::SerializedObject &object,
const gpu::TargetOptions &options) const;
};
} // namespace
// Register the SPIR-V dialect, the SPIR-V translation & the target interface.
void mlir::spirv::registerSPIRVTargetInterfaceExternalModels(
DialectRegistry &registry) {
registry.addExtension(+[](MLIRContext *ctx, spirv::SPIRVDialect *dialect) {
spirv::TargetEnvAttr::attachInterface<SPIRVTargetAttrImpl>(*ctx);
});
}
void mlir::spirv::registerSPIRVTargetInterfaceExternalModels(
MLIRContext &context) {
DialectRegistry registry;
registerSPIRVTargetInterfaceExternalModels(registry);
context.appendDialectRegistry(registry);
}
// Reuse from existing serializer
std::optional<mlir::gpu::SerializedObject>
SPIRVTargetAttrImpl::serializeToObject(
Attribute attribute, Operation *module,
const gpu::TargetOptions &options) const {
if (!module)
return std::nullopt;
auto gpuMod = dyn_cast<gpu::GPUModuleOp>(module);
if (!gpuMod) {
module->emitError("expected to be a gpu.module op");
return std::nullopt;
}
auto spvMods = gpuMod.getOps<spirv::ModuleOp>();
if (spvMods.empty())
return std::nullopt;
auto spvMod = *spvMods.begin();
llvm::SmallVector<uint32_t, 0> spvBinary;
spvBinary.clear();
// Serialize the spirv.module op to SPIR-V blob.
if (mlir::failed(spirv::serialize(spvMod, spvBinary))) {
spvMod.emitError() << "failed to serialize SPIR-V module";
return std::nullopt;
}
SmallVector<char, 0> spvData(spvBinary.size() * sizeof(uint32_t), 0);
std::memcpy(spvData.data(), spvBinary.data(), spvData.size());
spvMod.erase();
return gpu::SerializedObject{std::move(spvData)};
}
// Prepare Attribute for gpu.binary with serialized kernel object
Attribute
SPIRVTargetAttrImpl::createObject(Attribute attribute, Operation *module,
const mlir::gpu::SerializedObject &object,
const gpu::TargetOptions &options) const {
gpu::CompilationTarget format = options.getCompilationTarget();
DictionaryAttr objectProps;
Builder builder(attribute.getContext());
return builder.getAttr<gpu::ObjectAttr>(
attribute, format,
builder.getStringAttr(
StringRef(object.getObject().data(), object.getObject().size())),
objectProps, /*kernels=*/nullptr);
}