本文地址:https://www.cnblogs.com/wanger-sjtu/p/17573212.html
TVM在编译过程中,经历了
graph LR A[3rd IR] --> B[Relay IR] B --> C[TIR] C --> D[LLVM IR] C -->E[Source]这一系列的过程。其中在生成cpu、rocm、nvptx、hexagon等平台的相关代码的时候,会先由TVM的TIR
转换为LLVM IR
,在后续由LLVM生成相关的机器码。
这一步是由tvm::codegen::Build
调用转换的。
runtime::Module Build(IRModule mod, Target target) {
if (transform::PassContext::Current()
->GetConfig<Bool>("tir.disable_assert", Bool(false))
.value()) {
mod = tir::transform::SkipAssert()(mod);
}
auto target_attr_map = tvm::TargetKind::GetAttrMap<FTVMTIRToRuntime>("TIRToRuntime");
if (target_attr_map.count(target->kind)) {
return target_attr_map[target->kind](mod, target);
}
// the build function.
std::string build_f_name = "target.build." + target->kind->name;
const PackedFunc* bf = runtime::Registry::Get(build_f_name);
ICHECK(bf != nullptr) << build_f_name << " is not enabled";
return (*bf)(mod, target);
}
在LLVM相关的target时候,这里的build_f_name
就是target.build.llvm
这时候会走到
TVM_REGISTER_GLOBAL("target.build.llvm")
.set_body_typed([](IRModule mod, Target target) -> runtime::Module {
auto n = make_object<LLVMModuleNode>();
n->Init(mod, target);
return runtime::Module(n);
});
在Init
函数中创建codegen的具体类:
void LLVMModuleNode::Init(const IRModule& mod, const Target& target) {
llvm_instance_ = std::make_unique<LLVMInstance>();
With<LLVMTarget> llvm_target(*llvm_instance_, target);
llvm::TargetMachine* tm = llvm_target->GetOrCreateTargetMachine();
// 这里会根据target得到不同的codegen的实现类
std::unique_ptr<CodeGenLLVM> cg = CodeGenLLVM::Create(llvm_target.get());
std::string entry_func;
/*
skip crt/cpp systemlib options
*/
for (auto kv : mod->functions) {
if (!kv.second->IsInstance<PrimFuncNode>()) {
// (@jroesch): we relax constraints here, Relay functions will just be ignored.
DLOG(INFO) << "Can only lower IR Module with PrimFuncs, but got " << kv.second->GetTypeKey();
continue;
}
auto f = Downcast<PrimFunc>(kv.second);
auto global_symbol = f->GetAttr<String>(tvm::attr::kGlobalSymbol);
bool is_entry_func = f->HasNonzeroAttr(tir::attr::kIsEntryFunc);
if (global_symbol) {
function_names_.push_back(global_symbol.value());
if (is_entry_func) {
entry_func = global_symbol.value();
}
}
}
// 初始化CodeGenLLVM, 会产生builder_, module_等llvm 中codegen需要的基础数据结构
cg->Init("TVMMod", llvm_target.get(), system_lib_prefix,
system_lib_prefix.defined(),
target_c_runtime);
cg->SetFastMathFlags(llvm_target->GetFastMathFlags());
// 核心功能,tir 转化为llvm ir就在此
cg->AddFunctionsOrdered(mod->functions.begin(), mod->functions.end());
if (entry_func.length() != 0) {
cg->AddMainFunction(entry_func);
}
module_owning_ptr_ = cg->Finish();
module_ = module_owning_ptr_.get();
llvm_target->SetTargetMetadata(module_);
module_->addModuleFlag(llvm::Module::Override, "Debug Info Version",
llvm::DEBUG_METADATA_VERSION);
}
标签:代码生成,llvm,target,IR,module,TVM,LLVM,build,mod
From: https://www.cnblogs.com/wanger-sjtu/p/17573212.html