The existing implementation of the ExecutionEngine unconditionally runs a list
of "default" MLIR passes on the module upon creation. These passes include,
among others, dialect conversions from affine to standard and from standard to
LLVM IR dialects. In some cases, these conversions might have been performed
before ExecutionEngine is created. More advanced use cases may be performing
additional transformations that the "default" passes will conflict with.
Provide an overload for ExecutionEngine::create that takes a PassManager
configured with the passes to run on the module. If it is not provided, do not
run any passes. The engine will not be created if the input module, after the
pass manager, has any other dialect than the LLVM IR dialect.
--
PiperOrigin-RevId: 242127393
333 lines
12 KiB
C++
333 lines
12 KiB
C++
//===- ExecutionEngine.cpp - MLIR Execution engine and utils --------------===//
|
|
//
|
|
// Copyright 2019 The MLIR Authors.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
// =============================================================================
|
|
//
|
|
// This file implements the execution engine for MLIR modules based on LLVM Orc
|
|
// JIT engine.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
#include "mlir/ExecutionEngine/ExecutionEngine.h"
|
|
#include "mlir/IR/Function.h"
|
|
#include "mlir/IR/Module.h"
|
|
#include "mlir/LLVMIR/Transforms.h"
|
|
#include "mlir/Pass/Pass.h"
|
|
#include "mlir/Pass/PassManager.h"
|
|
#include "mlir/Target/LLVMIR.h"
|
|
#include "mlir/Transforms/Passes.h"
|
|
|
|
#include "llvm/ExecutionEngine/Orc/CompileUtils.h"
|
|
#include "llvm/ExecutionEngine/Orc/ExecutionUtils.h"
|
|
#include "llvm/ExecutionEngine/Orc/IRCompileLayer.h"
|
|
#include "llvm/ExecutionEngine/Orc/IRTransformLayer.h"
|
|
#include "llvm/ExecutionEngine/Orc/JITTargetMachineBuilder.h"
|
|
#include "llvm/ExecutionEngine/Orc/RTDyldObjectLinkingLayer.h"
|
|
#include "llvm/ExecutionEngine/SectionMemoryManager.h"
|
|
#include "llvm/IR/IRBuilder.h"
|
|
#include "llvm/Support/Error.h"
|
|
#include "llvm/Support/TargetRegistry.h"
|
|
|
|
using namespace mlir;
|
|
using llvm::Error;
|
|
using llvm::Expected;
|
|
|
|
namespace {
|
|
// Memory manager for the JIT's objectLayer. Its main goal is to fallback to
|
|
// resolving functions in the current process if they cannot be resolved in the
|
|
// JIT-compiled modules.
|
|
class MemoryManager : public llvm::SectionMemoryManager {
|
|
public:
|
|
MemoryManager(llvm::orc::ExecutionSession &execSession)
|
|
: session(execSession) {}
|
|
|
|
// Resolve the named symbol. First, try looking it up in the main library of
|
|
// the execution session. If there is no such symbol, try looking it up in
|
|
// the current process (for example, if it is a standard library function).
|
|
// Return `nullptr` if lookup fails.
|
|
llvm::JITSymbol findSymbol(const std::string &name) override {
|
|
auto mainLibSymbol = session.lookup({&session.getMainJITDylib()}, name);
|
|
if (mainLibSymbol)
|
|
return mainLibSymbol.get();
|
|
auto address = llvm::RTDyldMemoryManager::getSymbolAddressInProcess(name);
|
|
if (!address) {
|
|
llvm::errs() << "Could not look up: " << name << '\n';
|
|
return nullptr;
|
|
}
|
|
return llvm::JITSymbol(address, llvm::JITSymbolFlags::Exported);
|
|
}
|
|
|
|
private:
|
|
llvm::orc::ExecutionSession &session;
|
|
};
|
|
} // end anonymous namespace
|
|
|
|
namespace mlir {
|
|
namespace impl {
|
|
// Simple layered Orc JIT compilation engine.
|
|
class OrcJIT {
|
|
public:
|
|
using IRTransformer = std::function<Error(llvm::Module *)>;
|
|
|
|
// Construct a JIT engine for the target host defined by `machineBuilder`,
|
|
// using the data layout provided as `dataLayout`.
|
|
// Setup the object layer to use our custom memory manager in order to resolve
|
|
// calls to library functions present in the process.
|
|
OrcJIT(llvm::orc::JITTargetMachineBuilder machineBuilder,
|
|
llvm::DataLayout layout, IRTransformer transform)
|
|
: irTransformer(transform),
|
|
objectLayer(
|
|
session,
|
|
[this]() { return llvm::make_unique<MemoryManager>(session); }),
|
|
compileLayer(
|
|
session, objectLayer,
|
|
llvm::orc::ConcurrentIRCompiler(std::move(machineBuilder))),
|
|
transformLayer(session, compileLayer, makeIRTransformFunction()),
|
|
dataLayout(layout), mangler(session, this->dataLayout),
|
|
threadSafeCtx(llvm::make_unique<llvm::LLVMContext>()) {
|
|
session.getMainJITDylib().setGenerator(
|
|
cantFail(llvm::orc::DynamicLibrarySearchGenerator::GetForCurrentProcess(
|
|
layout)));
|
|
}
|
|
|
|
// Create a JIT engine for the current host.
|
|
static Expected<std::unique_ptr<OrcJIT>>
|
|
createDefault(IRTransformer transformer) {
|
|
auto machineBuilder = llvm::orc::JITTargetMachineBuilder::detectHost();
|
|
if (!machineBuilder)
|
|
return machineBuilder.takeError();
|
|
|
|
auto dataLayout = machineBuilder->getDefaultDataLayoutForTarget();
|
|
if (!dataLayout)
|
|
return dataLayout.takeError();
|
|
|
|
return llvm::make_unique<OrcJIT>(std::move(*machineBuilder),
|
|
std::move(*dataLayout), transformer);
|
|
}
|
|
|
|
// Add an LLVM module to the main library managed by the JIT engine.
|
|
Error addModule(std::unique_ptr<llvm::Module> M) {
|
|
return transformLayer.add(
|
|
session.getMainJITDylib(),
|
|
llvm::orc::ThreadSafeModule(std::move(M), threadSafeCtx));
|
|
}
|
|
|
|
// Lookup a symbol in the main library managed by the JIT engine.
|
|
Expected<llvm::JITEvaluatedSymbol> lookup(StringRef Name) {
|
|
return session.lookup({&session.getMainJITDylib()}, mangler(Name.str()));
|
|
}
|
|
|
|
private:
|
|
// Wrap the `irTransformer` into a function that can be called by the
|
|
// IRTranformLayer. If `irTransformer` is not set up, return the module as is
|
|
// without errors.
|
|
llvm::orc::IRTransformLayer::TransformFunction makeIRTransformFunction() {
|
|
return [this](llvm::orc::ThreadSafeModule module,
|
|
const llvm::orc::MaterializationResponsibility &resp)
|
|
-> Expected<llvm::orc::ThreadSafeModule> {
|
|
(void)resp;
|
|
if (!irTransformer)
|
|
return std::move(module);
|
|
if (Error err = irTransformer(module.getModule()))
|
|
return std::move(err);
|
|
return std::move(module);
|
|
};
|
|
}
|
|
|
|
IRTransformer irTransformer;
|
|
llvm::orc::ExecutionSession session;
|
|
llvm::orc::RTDyldObjectLinkingLayer objectLayer;
|
|
llvm::orc::IRCompileLayer compileLayer;
|
|
llvm::orc::IRTransformLayer transformLayer;
|
|
llvm::DataLayout dataLayout;
|
|
llvm::orc::MangleAndInterner mangler;
|
|
llvm::orc::ThreadSafeContext threadSafeCtx;
|
|
};
|
|
} // end namespace impl
|
|
} // namespace mlir
|
|
|
|
// Wrap a string into an llvm::StringError.
|
|
static inline Error make_string_error(const llvm::Twine &message) {
|
|
return llvm::make_error<llvm::StringError>(message.str(),
|
|
llvm::inconvertibleErrorCode());
|
|
}
|
|
|
|
// Given a list of PassRegistryEntry coming from a higher level, populates the
|
|
// given pass manager and appends the default set of required passes to lower to
|
|
// LLVMIR.
|
|
// Currently, these passes are:
|
|
// - constant folding
|
|
// - CSE
|
|
// - canonicalization
|
|
// - affine lowering
|
|
static void getDefaultPasses(
|
|
PassManager &manager,
|
|
const std::vector<const mlir::PassRegistryEntry *> &mlirPassRegistryList) {
|
|
// Run each of the passes that were selected.
|
|
for (const auto *passEntry : mlirPassRegistryList)
|
|
passEntry->addToPipeline(manager);
|
|
|
|
// Append the extra passes for lowering to MLIR.
|
|
manager.addPass(mlir::createCanonicalizerPass());
|
|
manager.addPass(mlir::createCSEPass());
|
|
manager.addPass(mlir::createCanonicalizerPass());
|
|
manager.addPass(mlir::createLowerAffinePass());
|
|
manager.addPass(mlir::createConvertToLLVMIRPass());
|
|
}
|
|
|
|
// Setup LLVM target triple from the current machine.
|
|
static bool setupTargetTriple(llvm::Module *llvmModule) {
|
|
// Setup the machine properties from the current architecture.
|
|
auto targetTriple = llvm::sys::getDefaultTargetTriple();
|
|
std::string errorMessage;
|
|
auto target = llvm::TargetRegistry::lookupTarget(targetTriple, errorMessage);
|
|
if (!target) {
|
|
llvm::errs() << "NO target: " << errorMessage << "\n";
|
|
return true;
|
|
}
|
|
auto machine =
|
|
target->createTargetMachine(targetTriple, "generic", "", {}, {});
|
|
llvmModule->setDataLayout(machine->createDataLayout());
|
|
llvmModule->setTargetTriple(targetTriple);
|
|
return false;
|
|
}
|
|
|
|
static std::string makePackedFunctionName(StringRef name) {
|
|
return "_mlir_" + name.str();
|
|
}
|
|
|
|
// For each function in the LLVM module, define an interface function that wraps
|
|
// all the arguments of the original function and all its results into an i8**
|
|
// pointer to provide a unified invocation interface.
|
|
void packFunctionArguments(llvm::Module *module) {
|
|
auto &ctx = module->getContext();
|
|
llvm::IRBuilder<> builder(ctx);
|
|
llvm::DenseSet<llvm::Function *> interfaceFunctions;
|
|
for (auto &func : module->getFunctionList()) {
|
|
if (func.isDeclaration()) {
|
|
continue;
|
|
}
|
|
if (interfaceFunctions.count(&func)) {
|
|
continue;
|
|
}
|
|
|
|
// Given a function `foo(<...>)`, define the interface function
|
|
// `mlir_foo(i8**)`.
|
|
auto newType = llvm::FunctionType::get(
|
|
builder.getVoidTy(), builder.getInt8PtrTy()->getPointerTo(),
|
|
/*isVarArg=*/false);
|
|
auto newName = makePackedFunctionName(func.getName());
|
|
auto funcCst = module->getOrInsertFunction(newName, newType);
|
|
llvm::Function *interfaceFunc =
|
|
llvm::cast<llvm::Function>(funcCst.getCallee());
|
|
interfaceFunctions.insert(interfaceFunc);
|
|
|
|
// Extract the arguments from the type-erased argument list and cast them to
|
|
// the proper types.
|
|
auto bb = llvm::BasicBlock::Create(ctx);
|
|
bb->insertInto(interfaceFunc);
|
|
builder.SetInsertPoint(bb);
|
|
llvm::Value *argList = interfaceFunc->arg_begin();
|
|
llvm::SmallVector<llvm::Value *, 8> args;
|
|
args.reserve(llvm::size(func.args()));
|
|
for (auto &indexedArg : llvm::enumerate(func.args())) {
|
|
llvm::Value *argIndex = llvm::Constant::getIntegerValue(
|
|
builder.getInt64Ty(), llvm::APInt(64, indexedArg.index()));
|
|
llvm::Value *argPtrPtr = builder.CreateGEP(argList, argIndex);
|
|
llvm::Value *argPtr = builder.CreateLoad(argPtrPtr);
|
|
argPtr = builder.CreateBitCast(
|
|
argPtr, indexedArg.value().getType()->getPointerTo());
|
|
llvm::Value *arg = builder.CreateLoad(argPtr);
|
|
args.push_back(arg);
|
|
}
|
|
|
|
// Call the implementation function with the extracted arguments.
|
|
llvm::Value *result = builder.CreateCall(&func, args);
|
|
|
|
// Assuming the result is one value, potentially of type `void`.
|
|
if (!result->getType()->isVoidTy()) {
|
|
llvm::Value *retIndex = llvm::Constant::getIntegerValue(
|
|
builder.getInt64Ty(), llvm::APInt(64, llvm::size(func.args())));
|
|
llvm::Value *retPtrPtr = builder.CreateGEP(argList, retIndex);
|
|
llvm::Value *retPtr = builder.CreateLoad(retPtrPtr);
|
|
retPtr = builder.CreateBitCast(retPtr, result->getType()->getPointerTo());
|
|
builder.CreateStore(result, retPtr);
|
|
}
|
|
|
|
// The interface function returns void.
|
|
builder.CreateRetVoid();
|
|
}
|
|
}
|
|
|
|
// Out of line for PIMPL unique_ptr.
|
|
ExecutionEngine::~ExecutionEngine() = default;
|
|
|
|
Expected<std::unique_ptr<ExecutionEngine>> ExecutionEngine::create(
|
|
Module *m, PassManager *pm,
|
|
std::function<llvm::Error(llvm::Module *)> transformer) {
|
|
auto engine = llvm::make_unique<ExecutionEngine>();
|
|
auto expectedJIT = impl::OrcJIT::createDefault(transformer);
|
|
if (!expectedJIT)
|
|
return expectedJIT.takeError();
|
|
|
|
if (pm && failed(pm->run(m)))
|
|
return make_string_error("passes failed");
|
|
|
|
auto llvmModule = translateModuleToLLVMIR(*m);
|
|
if (!llvmModule)
|
|
return make_string_error("could not convert to LLVM IR");
|
|
// FIXME: the triple should be passed to the translation or dialect conversion
|
|
// instead of this. Currently, the LLVM module created above has no triple
|
|
// associated with it.
|
|
setupTargetTriple(llvmModule.get());
|
|
packFunctionArguments(llvmModule.get());
|
|
|
|
if (auto err = (*expectedJIT)->addModule(std::move(llvmModule)))
|
|
return std::move(err);
|
|
engine->jit = std::move(*expectedJIT);
|
|
|
|
return std::move(engine);
|
|
}
|
|
|
|
Expected<std::unique_ptr<ExecutionEngine>> ExecutionEngine::create(
|
|
Module *m, std::function<llvm::Error(llvm::Module *)> transformer) {
|
|
// Construct and run the default MLIR pipeline.
|
|
PassManager manager;
|
|
getDefaultPasses(manager, {});
|
|
return create(m, &manager, transformer);
|
|
}
|
|
|
|
Expected<void (*)(void **)> ExecutionEngine::lookup(StringRef name) const {
|
|
auto expectedSymbol = jit->lookup(makePackedFunctionName(name));
|
|
if (!expectedSymbol)
|
|
return expectedSymbol.takeError();
|
|
auto rawFPtr = expectedSymbol->getAddress();
|
|
auto fptr = reinterpret_cast<void (*)(void **)>(rawFPtr);
|
|
if (!fptr)
|
|
return make_string_error("looked up function is null");
|
|
return fptr;
|
|
}
|
|
|
|
llvm::Error ExecutionEngine::invoke(StringRef name,
|
|
MutableArrayRef<void *> args) {
|
|
auto expectedFPtr = lookup(name);
|
|
if (!expectedFPtr)
|
|
return expectedFPtr.takeError();
|
|
auto fptr = *expectedFPtr;
|
|
|
|
(*fptr)(args.data());
|
|
|
|
return llvm::Error::success();
|
|
}
|