first commit of new LTO system. It is not hooked up in the llvm/tools/Makefile, so no one will build it be default yet
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47621 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/tools/lto2/LTOCodeGenerator.cpp b/tools/lto2/LTOCodeGenerator.cpp
new file mode 100644
index 0000000..795aca0
--- /dev/null
+++ b/tools/lto2/LTOCodeGenerator.cpp
@@ -0,0 +1,465 @@
+//===-LTOCodeGenerator.cpp - LLVM Link Time Optimizer ---------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the Link Time Optimization library. This library is
+// intended to be used by linker to optimize code at link time.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Module.h"
+#include "llvm/PassManager.h"
+#include "llvm/Linker.h"
+#include "llvm/Constants.h"
+#include "llvm/DerivedTypes.h"
+#include "llvm/ModuleProvider.h"
+#include "llvm/Bitcode/ReaderWriter.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/FileUtilities.h"
+#include "llvm/Support/SystemUtils.h"
+#include "llvm/Support/Mangler.h"
+#include "llvm/Support/MemoryBuffer.h"
+#include "llvm/System/Program.h"
+#include "llvm/System/Signals.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/LoopPass.h"
+#include "llvm/Analysis/Verifier.h"
+#include "llvm/CodeGen/FileWriters.h"
+#include "llvm/Target/SubtargetFeature.h"
+#include "llvm/Target/TargetOptions.h"
+#include "llvm/Target/TargetData.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetMachineRegistry.h"
+#include "llvm/Target/TargetAsmInfo.h"
+#include "llvm/Transforms/IPO.h"
+#include "llvm/Transforms/Scalar.h"
+#include "llvm/Analysis/LoadValueNumbering.h"
+#include "llvm/Support/MathExtras.h"
+#include "llvm/Config/config.h"
+
+#include "LTOModule.h"
+#include "LTOCodeGenerator.h"
+
+#include <fstream>
+#include <unistd.h>
+#include <stdlib.h>
+#include <fcntl.h>
+
+
+using namespace llvm;
+
+
+
+const char* LTOCodeGenerator::getVersionString()
+{
+#ifdef LLVM_VERSION_INFO
+ return PACKAGE_NAME " version " PACKAGE_VERSION ", " LLVM_VERSION_INFO;
+#else
+ return PACKAGE_NAME " version " PACKAGE_VERSION;
+#endif
+}
+
+
+LTOCodeGenerator::LTOCodeGenerator()
+ : _linker("LinkTimeOptimizer", "ld-temp.o"), _target(NULL),
+ _emitDwarfDebugInfo(false), _scopeRestrictionsDone(false),
+ _codeModel(LTO_CODEGEN_PIC_MODEL_DYNAMIC)
+{
+
+}
+
+LTOCodeGenerator::~LTOCodeGenerator()
+{
+ // FIXME
+}
+
+
+
+bool LTOCodeGenerator::addModule(LTOModule* mod, std::string& errMsg)
+{
+ return _linker.LinkInModule(mod->getLLVVMModule(), &errMsg);
+}
+
+
+bool LTOCodeGenerator::setDebugInfo(lto_debug_model debug, std::string& errMsg)
+{
+ switch (debug) {
+ case LTO_DEBUG_MODEL_NONE:
+ _emitDwarfDebugInfo = false;
+ return false;
+
+ case LTO_DEBUG_MODEL_DWARF:
+ _emitDwarfDebugInfo = true;
+ return false;
+ }
+ errMsg = "unknown debug format";
+ return true;
+}
+
+
+bool LTOCodeGenerator::setCodePICModel(lto_codegen_model model,
+ std::string& errMsg)
+{
+ switch (model) {
+ case LTO_CODEGEN_PIC_MODEL_STATIC:
+ case LTO_CODEGEN_PIC_MODEL_DYNAMIC:
+ case LTO_CODEGEN_PIC_MODEL_DYNAMIC_NO_PIC:
+ _codeModel = model;
+ return false;
+ }
+ errMsg = "unknown pic model";
+ return true;
+}
+
+
+void LTOCodeGenerator::addMustPreserveSymbol(const char* sym)
+{
+ _mustPreserveSymbols[sym] = 1;
+}
+
+
+bool LTOCodeGenerator::writeMergedModules(const char* path, std::string& errMsg)
+{
+ if ( this->determineTarget(errMsg) )
+ return true;
+
+ // mark which symbols can not be internalized
+ this->applyScopeRestrictions();
+
+ // create output file
+ std::ofstream out(path, std::ios_base::out|std::ios::trunc|std::ios::binary);
+ if ( out.fail() ) {
+ errMsg = "could not open bitcode file for writing: ";
+ errMsg += path;
+ return true;
+ }
+
+ // write bitcode to it
+ WriteBitcodeToFile(_linker.getModule(), out);
+ if ( out.fail() ) {
+ errMsg = "could not write bitcode file: ";
+ errMsg += path;
+ return true;
+ }
+
+ return false;
+}
+
+
+void* LTOCodeGenerator::compile(size_t* length, std::string& errMsg)
+{
+ // make unqiue temp .s file to put generated assembly code
+ sys::Path uniqueAsmPath("lto-llvm.s");
+ if ( uniqueAsmPath.createTemporaryFileOnDisk(true, &errMsg) )
+ return NULL;
+ sys::RemoveFileOnSignal(uniqueAsmPath);
+
+ // generate assembly code
+ std::ofstream asmFile(uniqueAsmPath.c_str());
+ bool genResult = this->generateAssemblyCode(asmFile, errMsg);
+ asmFile.close();
+ if ( genResult ) {
+ if ( uniqueAsmPath.exists() )
+ uniqueAsmPath.eraseFromDisk();
+ return NULL;
+ }
+
+ // make unqiue temp .o file to put generated object file
+ sys::PathWithStatus uniqueObjPath("lto-llvm.o");
+ if ( uniqueObjPath.createTemporaryFileOnDisk(true, &errMsg) ) {
+ if ( uniqueAsmPath.exists() )
+ uniqueAsmPath.eraseFromDisk();
+ return NULL;
+ }
+ sys::RemoveFileOnSignal(uniqueObjPath);
+
+ // assemble the assembly code
+ void* buffer = NULL;
+ bool asmResult = this->assemble(uniqueAsmPath.toString(),
+ uniqueObjPath.toString(), errMsg);
+ if ( !asmResult ) {
+ // read .o file into memory buffer
+ const sys::FileStatus* objStatus;
+ objStatus = uniqueObjPath.getFileStatus(false, &errMsg);
+ if ( objStatus != NULL ) {
+ *length = objStatus->getSize();
+ // use malloc() because caller will own this buffer and free() it
+ buffer = ::malloc(*length);
+ if ( buffer != NULL ) {
+ int fd = ::open(uniqueObjPath.c_str(), O_RDONLY, 0);
+ if ( fd != -1 ) {
+ // read object file contents into buffer
+ if ( ::read(fd, buffer, *length) != (ssize_t)*length ) {
+ errMsg = "error reading object file";
+ free(buffer);
+ buffer = NULL;
+ }
+ close(fd);
+ }
+ else {
+ errMsg = "error opening object file";
+ free(buffer);
+ buffer = NULL;
+ }
+ }
+ else {
+ errMsg = "error mallocing space for object file";
+ }
+ }
+ else {
+ errMsg = "error stat'ing object file";
+ }
+ }
+ // clean up temp files
+ uniqueAsmPath.eraseFromDisk();
+ uniqueObjPath.eraseFromDisk();
+ return buffer;
+}
+
+
+bool LTOCodeGenerator::assemble(const std::string& asmPath,
+ const std::string& objPath, std::string& errMsg)
+{
+ // find compiler driver
+ const sys::Path gcc = sys::Program::FindProgramByName("gcc");
+ if ( gcc.isEmpty() ) {
+ errMsg = "can't locate gcc";
+ return true;
+ }
+
+ // build argument list
+ std::vector<const char*> args;
+ std::string targetTriple = _linker.getModule()->getTargetTriple();
+ args.push_back(gcc.c_str());
+ if ( targetTriple.find("darwin") != targetTriple.size() ) {
+ if (strncmp(targetTriple.c_str(), "i686-apple-", 11) == 0) {
+ args.push_back("-arch");
+ args.push_back("i386");
+ }
+ else if (strncmp(targetTriple.c_str(), "x86_64-apple-", 13) == 0) {
+ args.push_back("-arch");
+ args.push_back("x86_64");
+ }
+ else if (strncmp(targetTriple.c_str(), "powerpc-apple-", 14) == 0) {
+ args.push_back("-arch");
+ args.push_back("ppc");
+ }
+ else if (strncmp(targetTriple.c_str(), "powerpc64-apple-", 16) == 0) {
+ args.push_back("-arch");
+ args.push_back("ppc64");
+ }
+ }
+ args.push_back("-c");
+ args.push_back("-x");
+ args.push_back("assembler");
+ args.push_back("-o");
+ args.push_back(objPath.c_str());
+ args.push_back(asmPath.c_str());
+ args.push_back(0);
+
+ // invoke assembler
+ if ( sys::Program::ExecuteAndWait(gcc, &args[0], 0, 0, 0, 0, &errMsg) ) {
+ errMsg = "error in assembly";
+ return true;
+ }
+ return false; // success
+}
+
+
+
+bool LTOCodeGenerator::determineTarget(std::string& errMsg)
+{
+ if ( _target == NULL ) {
+ // create target machine from info for merged modules
+ Module* mergedModule = _linker.getModule();
+ const TargetMachineRegistry::entry* march =
+ TargetMachineRegistry::getClosestStaticTargetForModule(
+ *mergedModule, errMsg);
+ if ( march == NULL )
+ return true;
+ std::string features;
+ _target = march->CtorFn(*mergedModule, features);
+ }
+ return false;
+}
+
+void LTOCodeGenerator::applyScopeRestrictions()
+{
+ if ( !_scopeRestrictionsDone ) {
+ Module* mergedModule = _linker.getModule();
+
+ // Start off with a verification pass.
+ PassManager passes;
+ passes.add(createVerifierPass());
+
+ // mark which symbols can not be internalized
+ if ( !_mustPreserveSymbols.empty() ) {
+ Mangler mangler(*mergedModule,
+ _target->getTargetAsmInfo()->getGlobalPrefix());
+ std::vector<const char*> mustPreserveList;
+ for (Module::iterator f = mergedModule->begin(),
+ e = mergedModule->end(); f != e; ++f) {
+ if ( !f->isDeclaration()
+ && _mustPreserveSymbols.count(mangler.getValueName(f)) )
+ mustPreserveList.push_back(::strdup(f->getName().c_str()));
+ }
+ for (Module::global_iterator v = mergedModule->global_begin(),
+ e = mergedModule->global_end(); v != e; ++v) {
+ if ( !v->isDeclaration()
+ && _mustPreserveSymbols.count(mangler.getValueName(v)) )
+ mustPreserveList.push_back(::strdup(v->getName().c_str()));
+ }
+ passes.add(createInternalizePass(mustPreserveList));
+ }
+ // apply scope restrictions
+ passes.run(*mergedModule);
+
+ _scopeRestrictionsDone = true;
+ }
+}
+
+
+/// Optimize merged modules using various IPO passes
+bool LTOCodeGenerator::generateAssemblyCode(std::ostream& out, std::string& errMsg)
+{
+ if ( this->determineTarget(errMsg) )
+ return true;
+
+ // mark which symbols can not be internalized
+ this->applyScopeRestrictions();
+
+ Module* mergedModule = _linker.getModule();
+
+ // If target supports exception handling then enable it now.
+ if ( _target->getTargetAsmInfo()->doesSupportExceptionHandling() )
+ llvm::ExceptionHandling = true;
+
+ // set codegen model
+ switch( _codeModel ) {
+ case LTO_CODEGEN_PIC_MODEL_STATIC:
+ _target->setRelocationModel(Reloc::Static);
+ break;
+ case LTO_CODEGEN_PIC_MODEL_DYNAMIC:
+ _target->setRelocationModel(Reloc::PIC_);
+ break;
+ case LTO_CODEGEN_PIC_MODEL_DYNAMIC_NO_PIC:
+ _target->setRelocationModel(Reloc::DynamicNoPIC);
+ break;
+ }
+
+ // Instantiate the pass manager to organize the passes.
+ PassManager passes;
+
+ // Start off with a verification pass.
+ passes.add(createVerifierPass());
+
+ // Add an appropriate TargetData instance for this module...
+ passes.add(new TargetData(*_target->getTargetData()));
+
+ // Now that we internalized some globals, see if we can hack on them!
+ passes.add(createGlobalOptimizerPass());
+
+ // Linking modules together can lead to duplicated global constants, only
+ // keep one copy of each constant...
+ passes.add(createConstantMergePass());
+
+ // If the -s command line option was specified, strip the symbols out of the
+ // resulting program to make it smaller. -s is a GLD option that we are
+ // supporting.
+ if( !llvm::ExceptionHandling ) {
+ // FIXME : This causes multiple nameless _.eh symbols on
+ // darwin when EH is ON.
+ passes.add(createStripSymbolsPass());
+ }
+
+ // Propagate constants at call sites into the functions they call.
+ passes.add(createIPConstantPropagationPass());
+
+ // Remove unused arguments from functions...
+ passes.add(createDeadArgEliminationPass());
+
+ passes.add(createFunctionInliningPass()); // Inline small functions
+
+ passes.add(createPruneEHPass()); // Remove dead EH info
+
+ passes.add(createGlobalDCEPass()); // Remove dead functions
+
+ // If we didn't decide to inline a function, check to see if we can
+ // transform it to pass arguments by value instead of by reference.
+ passes.add(createArgumentPromotionPass());
+
+ // The IPO passes may leave cruft around. Clean up after them.
+ passes.add(createInstructionCombiningPass());
+
+ passes.add(createScalarReplAggregatesPass()); // Break up allocas
+
+ // Run a few AA driven optimizations here and now, to cleanup the code.
+ passes.add(createGlobalsModRefPass()); // IP alias analysis
+
+ passes.add(createLICMPass()); // Hoist loop invariants
+ passes.add(createLoadValueNumberingPass()); // GVN for load instrs
+ passes.add(createGCSEPass()); // Remove common subexprs
+ passes.add(createDeadStoreEliminationPass()); // Nuke dead stores
+
+ // Cleanup and simplify the code after the scalar optimizations.
+ passes.add(createInstructionCombiningPass());
+
+ // Delete basic blocks, which optimization passes may have killed...
+ passes.add(createCFGSimplificationPass());
+
+ // Now that we have optimized the program, discard unreachable functions...
+ passes.add(createGlobalDCEPass());
+
+ // Make sure everything is still good.
+ passes.add(createVerifierPass());
+
+ FunctionPassManager* codeGenPasses =
+ new FunctionPassManager(new ExistingModuleProvider(mergedModule));
+
+ codeGenPasses->add(new TargetData(*_target->getTargetData()));
+
+ MachineCodeEmitter* mce = NULL;
+
+ switch (_target->addPassesToEmitFile(*codeGenPasses, out,
+ TargetMachine::AssemblyFile, true)) {
+ case FileModel::MachOFile:
+ mce = AddMachOWriter(*codeGenPasses, out, *_target);
+ break;
+ case FileModel::ElfFile:
+ mce = AddELFWriter(*codeGenPasses, out, *_target);
+ break;
+ case FileModel::AsmFile:
+ break;
+ case FileModel::Error:
+ case FileModel::None:
+ errMsg = "target file type not supported";
+ return true;
+ }
+
+ if (_target->addPassesToEmitFileFinish(*codeGenPasses, mce, true)) {
+ errMsg = "target does not support generation of this file type";
+ return true;
+ }
+
+ // Run our queue of passes all at once now, efficiently.
+ passes.run(*mergedModule);
+
+ // Run the code generator, and write assembly file
+ codeGenPasses->doInitialization();
+ for (Module::iterator it = mergedModule->begin(),
+ e = mergedModule->end(); it != e; ++it) {
+ if (!it->isDeclaration())
+ codeGenPasses->run(*it);
+ }
+ codeGenPasses->doFinalization();
+
+ return false; // success
+}
+
+
+