[clang][CodeGen] Shift relink option implementation away from module cloning #81693
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…cloning We recently implemented a new option allowing relinking of bitcode modules via the "-mllvm -relink-builtin-bitcode-postop" option. This implementation relied on llvm::CloneModule() in order to pass copies to modules and preserve the original modules for later relinking. However, cloning modules has been found to be prohibitively expensive, significantly increasing compilation time for large bitcode libraries. In this patch, we shift the relink option implementation to instead link the original modules initially, and reload modules from the file system if relinking is requested. This approach results in significantly reduced overhead.
llvmbot
added
clang
|
@llvm/pr-subscribers-clang-codegen @llvm/pr-subscribers-clang Author: Jacob Lambert (lamb-j) ChangesWe recently implemented a new option allowing relinking of bitcode modules via the "-mllvm -relink-builtin-bitcode-postop" option. This implementation relied on llvm::CloneModule() in order to pass copies to modules and preserve the original modules for later relinking. However, cloning modules has been found to be prohibitively expensive, significantly increasing compilation time for large bitcode libraries. In this patch, we shift the relink option implementation to instead link the original modules initially, and reload modules from the file system if relinking is requested. This approach results in significantly reduced overhead. Full diff: https://github.com/llvm/llvm-project/pull/81693.diff 5 Files Affected:
diff --git a/clang/include/clang/Basic/FileManager.h b/clang/include/clang/Basic/FileManager.h
index 997c17a0ffcfcc..2245fd78bfc9f0 100644
--- a/clang/include/clang/Basic/FileManager.h
+++ b/clang/include/clang/Basic/FileManager.h
@@ -283,7 +283,7 @@ class FileManager : public RefCountedBase<FileManager> {
bool RequiresNullTerminator = true);
llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
getBufferForFile(StringRef Filename, bool isVolatile = false,
- bool RequiresNullTerminator = true) {
+ bool RequiresNullTerminator = true) const {
return getBufferForFileImpl(Filename, /*FileSize=*/-1, isVolatile,
RequiresNullTerminator);
}
@@ -291,7 +291,7 @@ class FileManager : public RefCountedBase<FileManager> {
private:
llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
getBufferForFileImpl(StringRef Filename, int64_t FileSize, bool isVolatile,
- bool RequiresNullTerminator);
+ bool RequiresNullTerminator) const;
public:
/// Get the 'stat' information for the given \p Path.
diff --git a/clang/lib/Basic/FileManager.cpp b/clang/lib/Basic/FileManager.cpp
index 6097a27e429d66..cd520a6375e07e 100644
--- a/clang/lib/Basic/FileManager.cpp
+++ b/clang/lib/Basic/FileManager.cpp
@@ -547,7 +547,7 @@ FileManager::getBufferForFile(FileEntryRef FE, bool isVolatile,
llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
FileManager::getBufferForFileImpl(StringRef Filename, int64_t FileSize,
bool isVolatile,
- bool RequiresNullTerminator) {
+ bool RequiresNullTerminator) const {
if (FileSystemOpts.WorkingDir.empty())
return FS->getBufferForFile(Filename, FileSize, RequiresNullTerminator,
isVolatile);
diff --git a/clang/lib/CodeGen/BackendConsumer.h b/clang/lib/CodeGen/BackendConsumer.h
index 72a814cd43d738..fd0f1984d6c0f7 100644
--- a/clang/lib/CodeGen/BackendConsumer.h
+++ b/clang/lib/CodeGen/BackendConsumer.h
@@ -34,6 +34,7 @@ class BackendConsumer : public ASTConsumer {
const CodeGenOptions &CodeGenOpts;
const TargetOptions &TargetOpts;
const LangOptions &LangOpts;
+ const FileManager &FileMgr;
std::unique_ptr<raw_pwrite_stream> AsmOutStream;
ASTContext *Context;
IntrusiveRefCntPtr<llvm::vfs::FileSystem> FS;
@@ -74,8 +75,8 @@ class BackendConsumer : public ASTConsumer {
const HeaderSearchOptions &HeaderSearchOpts,
const PreprocessorOptions &PPOpts,
const CodeGenOptions &CodeGenOpts,
- const TargetOptions &TargetOpts,
- const LangOptions &LangOpts, const std::string &InFile,
+ const TargetOptions &TargetOpts, const LangOptions &LangOpts,
+ const FileManager &FileMgr, const std::string &InFile,
SmallVector<LinkModule, 4> LinkModules,
std::unique_ptr<raw_pwrite_stream> OS, llvm::LLVMContext &C,
CoverageSourceInfo *CoverageInfo = nullptr);
@@ -88,8 +89,8 @@ class BackendConsumer : public ASTConsumer {
const HeaderSearchOptions &HeaderSearchOpts,
const PreprocessorOptions &PPOpts,
const CodeGenOptions &CodeGenOpts,
- const TargetOptions &TargetOpts,
- const LangOptions &LangOpts, llvm::Module *Module,
+ const TargetOptions &TargetOpts, const LangOptions &LangOpts,
+ const FileManager &FileMgr, llvm::Module *Module,
SmallVector<LinkModule, 4> LinkModules, llvm::LLVMContext &C,
CoverageSourceInfo *CoverageInfo = nullptr);
@@ -111,10 +112,13 @@ class BackendConsumer : public ASTConsumer {
void AssignInheritanceModel(CXXRecordDecl *RD) override;
void HandleVTable(CXXRecordDecl *RD) override;
-
- // Links each entry in LinkModules into our module. Returns true on error.
+ // Links each entry in LinkModules into our module. Returns true on error.
bool LinkInModules(llvm::Module *M, bool ShouldLinkFiles = true);
+ // Load a bitcode module from -mlink-builtin-bitcode option using
+ // methods from a BackendConsumer instead of CompilerInstance
+ bool ReloadModules(llvm::Module *M);
+
/// Get the best possible source location to represent a diagnostic that
/// may have associated debug info.
const FullSourceLoc getBestLocationFromDebugLoc(
diff --git a/clang/lib/CodeGen/CodeGenAction.cpp b/clang/lib/CodeGen/CodeGenAction.cpp
index f8038497d90a7b..88b6ce84b51105 100644
--- a/clang/lib/CodeGen/CodeGenAction.cpp
+++ b/clang/lib/CodeGen/CodeGenAction.cpp
@@ -109,56 +109,52 @@ static void reportOptRecordError(Error E, DiagnosticsEngine &Diags,
});
}
-BackendConsumer::BackendConsumer(BackendAction Action, DiagnosticsEngine &Diags,
- IntrusiveRefCntPtr<llvm::vfs::FileSystem> VFS,
- const HeaderSearchOptions &HeaderSearchOpts,
- const PreprocessorOptions &PPOpts,
- const CodeGenOptions &CodeGenOpts,
- const TargetOptions &TargetOpts,
- const LangOptions &LangOpts,
- const std::string &InFile,
- SmallVector<LinkModule, 4> LinkModules,
- std::unique_ptr<raw_pwrite_stream> OS,
- LLVMContext &C,
- CoverageSourceInfo *CoverageInfo)
- : Diags(Diags), Action(Action), HeaderSearchOpts(HeaderSearchOpts),
- CodeGenOpts(CodeGenOpts), TargetOpts(TargetOpts), LangOpts(LangOpts),
- AsmOutStream(std::move(OS)), Context(nullptr), FS(VFS),
- LLVMIRGeneration("irgen", "LLVM IR Generation Time"),
- LLVMIRGenerationRefCount(0),
- Gen(CreateLLVMCodeGen(Diags, InFile, std::move(VFS), HeaderSearchOpts,
- PPOpts, CodeGenOpts, C, CoverageInfo)),
- LinkModules(std::move(LinkModules)) {
- TimerIsEnabled = CodeGenOpts.TimePasses;
- llvm::TimePassesIsEnabled = CodeGenOpts.TimePasses;
- llvm::TimePassesPerRun = CodeGenOpts.TimePassesPerRun;
+BackendConsumer::BackendConsumer(
+ BackendAction Action, DiagnosticsEngine &Diags,
+ IntrusiveRefCntPtr<llvm::vfs::FileSystem> VFS,
+ const HeaderSearchOptions &HeaderSearchOpts,
+ const PreprocessorOptions &PPOpts, const CodeGenOptions &CodeGenOpts,
+ const TargetOptions &TargetOpts, const LangOptions &LangOpts,
+ const FileManager &FileMgr, const std::string &InFile,
+ SmallVector<LinkModule, 4> LinkModules,
+ std::unique_ptr<raw_pwrite_stream> OS, LLVMContext &C,
+ CoverageSourceInfo *CoverageInfo)
+ : Diags(Diags), Action(Action), HeaderSearchOpts(HeaderSearchOpts),
+ CodeGenOpts(CodeGenOpts), TargetOpts(TargetOpts), LangOpts(LangOpts),
+ FileMgr(FileMgr), AsmOutStream(std::move(OS)), Context(nullptr), FS(VFS),
+ LLVMIRGeneration("irgen", "LLVM IR Generation Time"),
+ LLVMIRGenerationRefCount(0),
+ Gen(CreateLLVMCodeGen(Diags, InFile, std::move(VFS), HeaderSearchOpts,
+ PPOpts, CodeGenOpts, C, CoverageInfo)),
+ LinkModules(std::move(LinkModules)) {
+ TimerIsEnabled = CodeGenOpts.TimePasses;
+ llvm::TimePassesIsEnabled = CodeGenOpts.TimePasses;
+ llvm::TimePassesPerRun = CodeGenOpts.TimePassesPerRun;
}
// This constructor is used in installing an empty BackendConsumer
// to use the clang diagnostic handler for IR input files. It avoids
// initializing the OS field.
-BackendConsumer::BackendConsumer(BackendAction Action, DiagnosticsEngine &Diags,
- IntrusiveRefCntPtr<llvm::vfs::FileSystem> VFS,
- const HeaderSearchOptions &HeaderSearchOpts,
- const PreprocessorOptions &PPOpts,
- const CodeGenOptions &CodeGenOpts,
- const TargetOptions &TargetOpts,
- const LangOptions &LangOpts,
- llvm::Module *Module,
- SmallVector<LinkModule, 4> LinkModules,
- LLVMContext &C,
- CoverageSourceInfo *CoverageInfo)
- : Diags(Diags), Action(Action), HeaderSearchOpts(HeaderSearchOpts),
- CodeGenOpts(CodeGenOpts), TargetOpts(TargetOpts), LangOpts(LangOpts),
- Context(nullptr), FS(VFS),
- LLVMIRGeneration("irgen", "LLVM IR Generation Time"),
- LLVMIRGenerationRefCount(0),
- Gen(CreateLLVMCodeGen(Diags, "", std::move(VFS), HeaderSearchOpts,
- PPOpts, CodeGenOpts, C, CoverageInfo)),
- LinkModules(std::move(LinkModules)), CurLinkModule(Module) {
- TimerIsEnabled = CodeGenOpts.TimePasses;
- llvm::TimePassesIsEnabled = CodeGenOpts.TimePasses;
- llvm::TimePassesPerRun = CodeGenOpts.TimePassesPerRun;
+BackendConsumer::BackendConsumer(
+ BackendAction Action, DiagnosticsEngine &Diags,
+ IntrusiveRefCntPtr<llvm::vfs::FileSystem> VFS,
+ const HeaderSearchOptions &HeaderSearchOpts,
+ const PreprocessorOptions &PPOpts, const CodeGenOptions &CodeGenOpts,
+ const TargetOptions &TargetOpts, const LangOptions &LangOpts,
+ const FileManager &FileMgr, llvm::Module *Module,
+ SmallVector<LinkModule, 4> LinkModules, LLVMContext &C,
+ CoverageSourceInfo *CoverageInfo)
+ : Diags(Diags), Action(Action), HeaderSearchOpts(HeaderSearchOpts),
+ CodeGenOpts(CodeGenOpts), TargetOpts(TargetOpts), LangOpts(LangOpts),
+ FileMgr(FileMgr), Context(nullptr), FS(VFS),
+ LLVMIRGeneration("irgen", "LLVM IR Generation Time"),
+ LLVMIRGenerationRefCount(0),
+ Gen(CreateLLVMCodeGen(Diags, "", std::move(VFS), HeaderSearchOpts, PPOpts,
+ CodeGenOpts, C, CoverageInfo)),
+ LinkModules(std::move(LinkModules)), CurLinkModule(Module) {
+ TimerIsEnabled = CodeGenOpts.TimePasses;
+ llvm::TimePassesIsEnabled = CodeGenOpts.TimePasses;
+ llvm::TimePassesPerRun = CodeGenOpts.TimePassesPerRun;
}
llvm::Module* BackendConsumer::getModule() const {
@@ -233,9 +229,37 @@ void BackendConsumer::HandleInterestingDecl(DeclGroupRef D) {
HandleTopLevelDecl(D);
}
+bool BackendConsumer::ReloadModules(llvm::Module *M) {
+ for (const CodeGenOptions::BitcodeFileToLink &F :
+ CodeGenOpts.LinkBitcodeFiles) {
+ auto BCBuf = FileMgr.getBufferForFile(F.Filename);
+ if (!BCBuf) {
+ Diags.Report(diag::err_cannot_open_file)
+ << F.Filename << BCBuf.getError().message();
+ LinkModules.clear();
+ return true;
+ }
+
+ LLVMContext &Ctx = getModule()->getContext();
+ Expected<std::unique_ptr<llvm::Module>> ModuleOrErr =
+ getOwningLazyBitcodeModule(std::move(*BCBuf), Ctx);
+
+ if (!ModuleOrErr) {
+ handleAllErrors(ModuleOrErr.takeError(), [&](ErrorInfoBase &EIB) {
+ Diags.Report(diag::err_cannot_open_file) << F.Filename << EIB.message();
+ });
+ LinkModules.clear();
+ return true;
+ }
+ LinkModules.push_back({std::move(ModuleOrErr.get()), F.PropagateAttrs,
+ F.Internalize, F.LinkFlags});
+ }
+
+ return false; // success
+}
+
// Links each entry in LinkModules into our module. Returns true on error.
bool BackendConsumer::LinkInModules(llvm::Module *M, bool ShouldLinkFiles) {
-
for (auto &LM : LinkModules) {
assert(LM.Module && "LinkModule does not actually have a module");
@@ -257,37 +281,19 @@ bool BackendConsumer::LinkInModules(llvm::Module *M, bool ShouldLinkFiles) {
CurLinkModule = LM.Module.get();
bool Err;
- auto DoLink = [&](auto &Mod) {
- if (LM.Internalize) {
- Err = Linker::linkModules(
- *M, std::move(Mod), LM.LinkFlags,
- [](llvm::Module &M, const llvm::StringSet<> &GVS) {
- internalizeModule(M, [&GVS](const llvm::GlobalValue &GV) {
- return !GV.hasName() || (GVS.count(GV.getName()) == 0);
- });
+ if (LM.Internalize) {
+ Err = Linker::linkModules(
+ *M, std::move(LM.Module), LM.LinkFlags,
+ [](llvm::Module &M, const llvm::StringSet<> &GVS) {
+ internalizeModule(M, [&GVS](const llvm::GlobalValue &GV) {
+ return !GV.hasName() || (GVS.count(GV.getName()) == 0);
});
- } else
- Err = Linker::linkModules(*M, std::move(Mod), LM.LinkFlags);
- };
-
- // Create a Clone to move to the linker, which preserves the original
- // linking modules, allowing them to be linked again in the future
- if (ClRelinkBuiltinBitcodePostop) {
- // TODO: If CloneModule() is updated to support cloning of unmaterialized
- // modules, we can remove this
- if (Error E = CurLinkModule->materializeAll())
- return false;
-
- std::unique_ptr<llvm::Module> Clone = llvm::CloneModule(*LM.Module);
-
- DoLink(Clone);
- }
- // Otherwise we can link (and clean up) the original modules
- else {
- DoLink(LM.Module);
- }
+ });
+ } else
+ Err = Linker::linkModules(*M, std::move(LM.Module), LM.LinkFlags);
}
+ LinkModules.clear();
return false; // success
}
@@ -350,7 +356,7 @@ void BackendConsumer::HandleTranslationUnit(ASTContext &C) {
}
// Link each LinkModule into our module.
- if (LinkInModules(getModule()))
+ if (LinkInModules(getModule(), true))
return;
for (auto &F : getModule()->functions()) {
@@ -1037,8 +1043,9 @@ CodeGenAction::CreateASTConsumer(CompilerInstance &CI, StringRef InFile) {
std::unique_ptr<BackendConsumer> Result(new BackendConsumer(
BA, CI.getDiagnostics(), &CI.getVirtualFileSystem(),
CI.getHeaderSearchOpts(), CI.getPreprocessorOpts(), CI.getCodeGenOpts(),
- CI.getTargetOpts(), CI.getLangOpts(), std::string(InFile),
- std::move(LinkModules), std::move(OS), *VMContext, CoverageInfo));
+ CI.getTargetOpts(), CI.getLangOpts(), CI.getFileManager(),
+ std::string(InFile), std::move(LinkModules), std::move(OS), *VMContext,
+ CoverageInfo));
BEConsumer = Result.get();
// Enable generating macro debug info only when debug info is not disabled and
@@ -1199,7 +1206,7 @@ void CodeGenAction::ExecuteAction() {
BackendConsumer Result(BA, CI.getDiagnostics(), &CI.getVirtualFileSystem(),
CI.getHeaderSearchOpts(), CI.getPreprocessorOpts(),
CI.getCodeGenOpts(), CI.getTargetOpts(),
- CI.getLangOpts(), TheModule.get(),
+ CI.getLangOpts(), CI.getFileManager(), TheModule.get(),
std::move(LinkModules), *VMContext, nullptr);
// Link in each pending link module.
diff --git a/clang/lib/CodeGen/LinkInModulesPass.cpp b/clang/lib/CodeGen/LinkInModulesPass.cpp
index 6ce2b94c1db82c..929539cc8f3346 100644
--- a/clang/lib/CodeGen/LinkInModulesPass.cpp
+++ b/clang/lib/CodeGen/LinkInModulesPass.cpp
@@ -14,6 +14,10 @@
#include "LinkInModulesPass.h"
#include "BackendConsumer.h"
+#include "clang/Basic/CodeGenOptions.h"
+#include "clang/Basic/FileManager.h"
+#include "clang/Basic/SourceManager.h"
+
using namespace llvm;
LinkInModulesPass::LinkInModulesPass(clang::BackendConsumer *BC,
@@ -21,9 +25,15 @@ LinkInModulesPass::LinkInModulesPass(clang::BackendConsumer *BC,
: BC(BC), ShouldLinkFiles(ShouldLinkFiles) {}
PreservedAnalyses LinkInModulesPass::run(Module &M, ModuleAnalysisManager &AM) {
+ if (!BC)
+ return PreservedAnalyses::all();
+
+ // Re-load bitcode modules from files
+ if (BC->ReloadModules(&M))
+ report_fatal_error("Bitcode module re-loading failed, aborted!");
- if (BC && BC->LinkInModules(&M, ShouldLinkFiles))
- report_fatal_error("Bitcode module linking failed, compilation aborted!");
+ if (BC->LinkInModules(&M, ShouldLinkFiles))
+ report_fatal_error("Bitcode module re-linking failed, aborted!");
return PreservedAnalyses::all();
}
|
AaronHLiu
reviewed
Feb 15, 2024
| @@ -257,37 +281,19 @@ bool BackendConsumer::LinkInModules(llvm::Module *M, bool ShouldLinkFiles) { | |||
| CurLinkModule = LM.Module.get(); | |||
| bool Err; | |||
There was a problem hiding this comment.
Got error:
clang/lib/CodeGen/CodeGenAction.cpp:282:10: error: variable 'Err' set but not used [-Werror,-Wunused-but-set-variable]
bool Err;
^
1 error generated.
There was a problem hiding this comment.
Oversight on my part. Should be fixed by #81777
searlmc1
pushed a commit
to ROCm/llvm-project
that referenced
this pull request
Apr 3, 2024
…cloning (llvm#81693) We recently implemented a new option allowing relinking of bitcode modules via the "-mllvm -relink-builtin-bitcode-postop" option. This implementation relied on llvm::CloneModule() in order to pass copies to modules and preserve the original modules for later relinking. However, cloning modules has been found to be prohibitively expensive, significantly increasing compilation time for large bitcode libraries. In this patch, we shift the relink option implementation to instead link the original modules initially, and reload modules from the file system if relinking is requested. This approach results in significantly reduced overhead. We accomplish this by creating a new ReloadModules() routine that can be called from a BackendConsumer class, to mimic the behavior of ASTConsumer's loadLinkModules(), but without access to the CompilerInstance. Because loading the bitcodes from the filesystem requires access to the FileManager class, we also forward a reference to the CompilerInstance class to the BackendConsumer. This mirrors what is already done for several CompilerInstance members, such as TargetOptions and CodeGenOptions. Finally, we needed to add a const specifier to the FileManager::getBufferForFile() routine to allow it to be called using the const reference returned from CompilerInstance::getFileManager() Change-Id: Ica6ff6cd2e5ec7c4724d28b0b686343ebd0b606e
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
We recently implemented a new option allowing relinking of bitcode modules via the "-mllvm -relink-builtin-bitcode-postop"
option.
This implementation relied on llvm::CloneModule() in order to pass copies to modules and preserve the original modules for later relinking. However, cloning modules has been found to be prohibitively expensive, significantly increasing compilation time for large bitcode libraries.
In this patch, we shift the relink option implementation to instead link the original modules initially, and reload modules from the file system if relinking is requested. This approach results in significantly reduced overhead.
We accomplish this by creating a new ReloadModules() routine that can be called from a BackendConsumer class, to mimic the behavior of ASTConsumer's loadLinkModules(), but without access to the CompilerInstance.
Because loading the bitcodes from the filesystem requires access to the FileManager class, we also forward a reference to the CompilerInstance class to the BackendConsumer. This mirrors what is already done for several CompilerInstance members, such as TargetOptions and CodeGenOptions.
Finally, we needed to add a const specifier to the FileManager::getBufferForFile() routine to allow it to be called using the const reference returned from CompilerInstance::getFileManager()