[lld] [llvm] [DRAFT] "Builtins world" for LTO (PR #164916)
Daniel Thornburgh via llvm-commits
llvm-commits at lists.llvm.org
Thu Nov 13 15:59:00 PST 2025
https://github.com/mysterymath updated https://github.com/llvm/llvm-project/pull/164916
>From 63ec1bb2553bfd00601a37c877188f4bc19952cc Mon Sep 17 00:00:00 2001
From: Daniel Thornburgh <dthorn at google.com>
Date: Thu, 7 Aug 2025 16:05:09 -0700
Subject: [PATCH 1/7] Builtins world prototype
- Preserve all libc symbols in LTO
- Disable libfuncs that aren't present in the module
This is overly-disably:
- Libfuncs not in bitcode are disabled too
- Libfuncs defined in other ThinLTO modules are disabled unless they are
imported into the module. Not sure when importing happens; maybe that
means this disables all libfuncs for ThinLTO entirely?
---
llvm/lib/LTO/LTOBackend.cpp | 12 ++++++++++++
llvm/lib/Object/CMakeLists.txt | 1 +
llvm/lib/Object/IRSymtab.cpp | 8 +++++++-
3 files changed, 20 insertions(+), 1 deletion(-)
diff --git a/llvm/lib/LTO/LTOBackend.cpp b/llvm/lib/LTO/LTOBackend.cpp
index 93118becedbac..db7c449cca75c 100644
--- a/llvm/lib/LTO/LTOBackend.cpp
+++ b/llvm/lib/LTO/LTOBackend.cpp
@@ -281,6 +281,18 @@ static void runNewPMPasses(const Config &Conf, Module &Mod, TargetMachine *TM,
new TargetLibraryInfoImpl(TM->getTargetTriple()));
if (Conf.Freestanding)
TLII->disableAllFunctions();
+
+ TargetLibraryInfo TLI(*TLII);
+ for (unsigned I = 0, E = static_cast<unsigned>(LibFunc::NumLibFuncs);
+ I != E; ++I) {
+ LibFunc F = static_cast<LibFunc>(I);
+ GlobalValue *Val = Mod.getNamedValue(TLI.getName(F));
+ // TODO: Non-lto functions should not be disabled, nor should functions that
+ // are somewhere in a ThinLTO link (just not imported in this module).
+ if (!Val || Val->isDeclaration())
+ TLII->setUnavailable(F);
+ }
+
FAM.registerPass([&] { return TargetLibraryAnalysis(*TLII); });
// Parse a custom AA pipeline if asked to.
diff --git a/llvm/lib/Object/CMakeLists.txt b/llvm/lib/Object/CMakeLists.txt
index 0f6d2f7c59a5c..c48d251249488 100644
--- a/llvm/lib/Object/CMakeLists.txt
+++ b/llvm/lib/Object/CMakeLists.txt
@@ -51,6 +51,7 @@ add_llvm_component_library(LLVMObject
BinaryFormat
MCParser
Support
+ Target
TargetParser
TextAPI
)
diff --git a/llvm/lib/Object/IRSymtab.cpp b/llvm/lib/Object/IRSymtab.cpp
index a45b34eb2e706..92ed072a80267 100644
--- a/llvm/lib/Object/IRSymtab.cpp
+++ b/llvm/lib/Object/IRSymtab.cpp
@@ -13,6 +13,7 @@
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/ADT/StringSet.h"
+#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/Bitcode/BitcodeReader.h"
#include "llvm/Config/llvm-config.h"
#include "llvm/IR/Comdat.h"
@@ -75,12 +76,14 @@ struct Builder {
Builder(SmallVector<char, 0> &Symtab, StringTableBuilder &StrtabBuilder,
BumpPtrAllocator &Alloc, const Triple &TT)
: Symtab(Symtab), StrtabBuilder(StrtabBuilder), Saver(Alloc), TT(TT),
- Libcalls(TT) {}
+ TLII(TT), TLI(TLII), Libcalls(TT) {}
DenseMap<const Comdat *, int> ComdatMap;
Mangler Mang;
const Triple &TT;
+ TargetLibraryInfoImpl TLII;
+ TargetLibraryInfo TLI;
// FIXME: This shouldn't be here.
RTLIB::RuntimeLibcallsInfo Libcalls;
@@ -95,6 +98,9 @@ struct Builder {
std::vector<storage::Str> DependentLibraries;
bool isPreservedName(StringRef Name) {
+ LibFunc F;
+ if (TLI.getLibFunc(Name, F) && TLI.has(F))
+ return true;
return Libcalls.getSupportedLibcallImpl(Name) != RTLIB::Unsupported;
}
>From 148a1a71efb5722d8b5941a435bac90741a3148b Mon Sep 17 00:00:00 2001
From: Daniel Thornburgh <dthorn at google.com>
Date: Mon, 20 Oct 2025 15:41:29 -0700
Subject: [PATCH 2/7] Add an LTO interface for querying all libfunc names
---
llvm/include/llvm/LTO/LTO.h | 8 ++++++++
llvm/lib/LTO/LTO.cpp | 15 +++++++++++++++
2 files changed, 23 insertions(+)
diff --git a/llvm/include/llvm/LTO/LTO.h b/llvm/include/llvm/LTO/LTO.h
index 3a4dc5a3dfcf8..2e98b2a304174 100644
--- a/llvm/include/llvm/LTO/LTO.h
+++ b/llvm/include/llvm/LTO/LTO.h
@@ -420,6 +420,14 @@ class LTO {
LLVM_ABI static SmallVector<const char *>
getRuntimeLibcallSymbols(const Triple &TT);
+ /// Static method that returns a list of library function symbols that can be
+ /// generated by LTO but might not be visible from bitcode symbol table.
+ /// Unlike the runtime libcalls, the linker can report to the code generator
+ /// which of these are actually available in the link, and the code generator
+ /// can then only reference that set of symbols.
+ LLVM_ABI static SmallVector<const char *>
+ getLibFuncSymbols(const Triple &TT, llvm::StringSaver &Saver);
+
private:
Config Conf;
diff --git a/llvm/lib/LTO/LTO.cpp b/llvm/lib/LTO/LTO.cpp
index fefc733fa7697..1d38120c6e642 100644
--- a/llvm/lib/LTO/LTO.cpp
+++ b/llvm/lib/LTO/LTO.cpp
@@ -1406,6 +1406,21 @@ SmallVector<const char *> LTO::getRuntimeLibcallSymbols(const Triple &TT) {
return LibcallSymbols;
}
+SmallVector<const char *> LTO::getLibFuncSymbols(const Triple &TT,
+ StringSaver &Saver) {
+ auto TLII = std::make_unique<TargetLibraryInfoImpl>(TT);
+ TargetLibraryInfo TLI(*TLII);
+ SmallVector<const char *> LibFuncSymbols;
+ LibFuncSymbols.reserve(LibFunc::NumLibFuncs);
+ for (unsigned I = 0, E = static_cast<unsigned>(LibFunc::NumLibFuncs); I != E;
+ ++I) {
+ LibFunc F = static_cast<LibFunc>(I);
+ if (TLI.has(F))
+ LibFuncSymbols.push_back(Saver.save(TLI.getName(F)).data());
+ }
+ return LibFuncSymbols;
+}
+
Error ThinBackendProc::emitFiles(
const FunctionImporter::ImportMapTy &ImportList, llvm::StringRef ModulePath,
const std::string &NewModulePath) const {
>From ebf8576efc25fe73ed8c08e1f102e1f43a2988df Mon Sep 17 00:00:00 2001
From: Daniel Thornburgh <dthorn at google.com>
Date: Tue, 21 Oct 2025 14:47:17 -0700
Subject: [PATCH 3/7] Collect bitcode libfuncs
---
lld/ELF/Driver.cpp | 14 +++++++++++++-
lld/ELF/LTO.cpp | 5 ++++-
lld/ELF/LTO.h | 3 ++-
3 files changed, 19 insertions(+), 3 deletions(-)
diff --git a/lld/ELF/Driver.cpp b/lld/ELF/Driver.cpp
index 8647752be31fe..b48e9f29d7b54 100644
--- a/lld/ELF/Driver.cpp
+++ b/lld/ELF/Driver.cpp
@@ -2709,7 +2709,19 @@ void LinkerDriver::compileBitcodeFiles(bool skipLinkedOutput) {
if (!ctx.bitcodeFiles.empty())
markBuffersAsDontNeed(ctx, skipLinkedOutput);
- ltoObjectFiles = lto->compile();
+ llvm::Triple tt(ctx.bitcodeFiles.front()->obj->getTargetTriple());
+ llvm::BumpPtrAllocator alloc;
+ llvm::StringSaver saver(alloc);
+ SmallVector<const char *> bitcodeLibFuncs;
+ for (const char *libFunc : lto::LTO::getLibFuncSymbols(tt, saver)) {
+ Symbol *sym = ctx.symtab->find(libFunc);
+ if (!sym)
+ continue;
+ if (isa<BitcodeFile>(sym->file))
+ bitcodeLibFuncs.push_back(libFunc);
+ }
+
+ ltoObjectFiles = lto->compile(bitcodeLibFuncs);
for (auto &file : ltoObjectFiles) {
auto *obj = cast<ObjFile<ELFT>>(file.get());
obj->parse(/*ignoreComdats=*/true);
diff --git a/lld/ELF/LTO.cpp b/lld/ELF/LTO.cpp
index 80c6d2482f9fa..06af36a117d7e 100644
--- a/lld/ELF/LTO.cpp
+++ b/lld/ELF/LTO.cpp
@@ -235,6 +235,8 @@ void BitcodeCompiler::add(BitcodeFile &f) {
const lto::InputFile::Symbol &objSym = objSyms[i];
lto::SymbolResolution &r = resols[i];
+ dbgs() << sym->
+
// Ideally we shouldn't check for SF_Undefined but currently IRObjectFile
// reports two symbols for module ASM defined. Without this check, lld
// flags an undefined in IR with a definition in ASM as prevailing.
@@ -311,7 +313,8 @@ static void thinLTOCreateEmptyIndexFiles(Ctx &ctx) {
// Merge all the bitcode files we have seen, codegen the result
// and return the resulting ObjectFile(s).
-SmallVector<std::unique_ptr<InputFile>, 0> BitcodeCompiler::compile() {
+SmallVector<std::unique_ptr<InputFile>, 0>
+BitcodeCompiler::compile(const SmallVector<const char *> &bitcodeLibfuncs) {
unsigned maxTasks = ltoObj->getMaxTasks();
buf.resize(maxTasks);
files.resize(maxTasks);
diff --git a/lld/ELF/LTO.h b/lld/ELF/LTO.h
index acf3bcff7f2f1..6f568d13a3182 100644
--- a/lld/ELF/LTO.h
+++ b/lld/ELF/LTO.h
@@ -42,7 +42,8 @@ class BitcodeCompiler {
~BitcodeCompiler();
void add(BitcodeFile &f);
- SmallVector<std::unique_ptr<InputFile>, 0> compile();
+ SmallVector<std::unique_ptr<InputFile>, 0>
+ compile(const SmallVector<const char *> &bitcodeLibfuncs);
private:
Ctx &ctx;
>From b52feaaa0c66211f5693e30c90b84013413f560f Mon Sep 17 00:00:00 2001
From: Daniel Thornburgh <dthorn at google.com>
Date: Tue, 21 Oct 2025 14:48:52 -0700
Subject: [PATCH 4/7] Remove hax
---
lld/ELF/LTO.cpp | 2 --
1 file changed, 2 deletions(-)
diff --git a/lld/ELF/LTO.cpp b/lld/ELF/LTO.cpp
index 06af36a117d7e..e25b57a1c4166 100644
--- a/lld/ELF/LTO.cpp
+++ b/lld/ELF/LTO.cpp
@@ -235,8 +235,6 @@ void BitcodeCompiler::add(BitcodeFile &f) {
const lto::InputFile::Symbol &objSym = objSyms[i];
lto::SymbolResolution &r = resols[i];
- dbgs() << sym->
-
// Ideally we shouldn't check for SF_Undefined but currently IRObjectFile
// reports two symbols for module ASM defined. Without this check, lld
// flags an undefined in IR with a definition in ASM as prevailing.
>From 0434173b4dbd51596f96f65f864c05817805a565 Mon Sep 17 00:00:00 2001
From: Daniel Thornburgh <dthorn at google.com>
Date: Tue, 21 Oct 2025 14:53:13 -0700
Subject: [PATCH 5/7] Pass through to LTO
---
lld/ELF/LTO.cpp | 4 ++-
lld/ELF/LTO.h | 2 +-
llvm/include/llvm/LTO/LTO.h | 12 ++++++--
llvm/include/llvm/LTO/LTOBackend.h | 7 +++--
llvm/lib/LTO/LTO.cpp | 44 +++++++++++++++++++-----------
llvm/lib/LTO/LTOBackend.cpp | 14 ++++++----
llvm/lib/LTO/LTOCodeGenerator.cpp | 4 +--
7 files changed, 57 insertions(+), 30 deletions(-)
diff --git a/lld/ELF/LTO.cpp b/lld/ELF/LTO.cpp
index e25b57a1c4166..c661780302a43 100644
--- a/lld/ELF/LTO.cpp
+++ b/lld/ELF/LTO.cpp
@@ -312,7 +312,9 @@ static void thinLTOCreateEmptyIndexFiles(Ctx &ctx) {
// Merge all the bitcode files we have seen, codegen the result
// and return the resulting ObjectFile(s).
SmallVector<std::unique_ptr<InputFile>, 0>
-BitcodeCompiler::compile(const SmallVector<const char *> &bitcodeLibfuncs) {
+BitcodeCompiler::compile(const SmallVector<const char *> &bitcodeLibFuncs) {
+ ltoObj->setBitcodeLibFuncs(bitcodeLibFuncs);
+
unsigned maxTasks = ltoObj->getMaxTasks();
buf.resize(maxTasks);
files.resize(maxTasks);
diff --git a/lld/ELF/LTO.h b/lld/ELF/LTO.h
index 6f568d13a3182..99a61f8279c6a 100644
--- a/lld/ELF/LTO.h
+++ b/lld/ELF/LTO.h
@@ -43,7 +43,7 @@ class BitcodeCompiler {
void add(BitcodeFile &f);
SmallVector<std::unique_ptr<InputFile>, 0>
- compile(const SmallVector<const char *> &bitcodeLibfuncs);
+ compile(const SmallVector<const char *> &bitcodeLibFuncs);
private:
Ctx &ctx;
diff --git a/llvm/include/llvm/LTO/LTO.h b/llvm/include/llvm/LTO/LTO.h
index 2e98b2a304174..346ff9817737f 100644
--- a/llvm/include/llvm/LTO/LTO.h
+++ b/llvm/include/llvm/LTO/LTO.h
@@ -264,7 +264,8 @@ class ThinBackendProc {
using ThinBackendFunction = std::function<std::unique_ptr<ThinBackendProc>(
const Config &C, ModuleSummaryIndex &CombinedIndex,
const DenseMap<StringRef, GVSummaryMapTy> &ModuleToDefinedGVSummaries,
- AddStreamFn AddStream, FileCache Cache)>;
+ AddStreamFn AddStream, FileCache Cache,
+ const SmallVector<const char *> &BitcodeLibFuncs)>;
/// This type defines the behavior following the thin-link phase during ThinLTO.
/// It encapsulates a backend function and a strategy for thread pool
@@ -279,10 +280,11 @@ struct ThinBackend {
std::unique_ptr<ThinBackendProc> operator()(
const Config &Conf, ModuleSummaryIndex &CombinedIndex,
const DenseMap<StringRef, GVSummaryMapTy> &ModuleToDefinedGVSummaries,
- AddStreamFn AddStream, FileCache Cache) {
+ AddStreamFn AddStream, FileCache Cache,
+ const SmallVector<const char *> &BitcodeLibFuncs) {
assert(isValid() && "Invalid backend function");
return Func(Conf, CombinedIndex, ModuleToDefinedGVSummaries,
- std::move(AddStream), std::move(Cache));
+ std::move(AddStream), std::move(Cache), BitcodeLibFuncs);
}
ThreadPoolStrategy getParallelism() const { return Parallelism; }
bool isValid() const { return static_cast<bool>(Func); }
@@ -400,6 +402,8 @@ class LTO {
LLVM_ABI Error add(std::unique_ptr<InputFile> Obj,
ArrayRef<SymbolResolution> Res);
+ LLVM_ABI void setBitcodeLibFuncs(const SmallVector<const char *> &BitcodeLibFuncs);
+
/// Returns an upper bound on the number of tasks that the client may expect.
/// This may only be called after all IR object files have been added. For a
/// full description of tasks see LTOBackend.h.
@@ -599,6 +603,8 @@ class LTO {
// Diagnostic optimization remarks file
LLVMRemarkFileHandle DiagnosticOutputFile;
+
+ const SmallVector<const char *> *BitcodeLibFuncs;
};
/// The resolution for a symbol. The linker must provide a SymbolResolution for
diff --git a/llvm/include/llvm/LTO/LTOBackend.h b/llvm/include/llvm/LTO/LTOBackend.h
index 48ad5aa64f61f..26bff22819d91 100644
--- a/llvm/include/llvm/LTO/LTOBackend.h
+++ b/llvm/include/llvm/LTO/LTOBackend.h
@@ -39,13 +39,15 @@ LLVM_ABI bool opt(const Config &Conf, TargetMachine *TM, unsigned Task,
Module &Mod, bool IsThinLTO,
ModuleSummaryIndex *ExportSummary,
const ModuleSummaryIndex *ImportSummary,
- const std::vector<uint8_t> &CmdArgs);
+ const std::vector<uint8_t> &CmdArgs,
+ const SmallVector<const char*>& BitcodeLibFuncs);
/// Runs a regular LTO backend. The regular LTO backend can also act as the
/// regular LTO phase of ThinLTO, which may need to access the combined index.
LLVM_ABI Error backend(const Config &C, AddStreamFn AddStream,
unsigned ParallelCodeGenParallelismLevel, Module &M,
- ModuleSummaryIndex &CombinedIndex);
+ ModuleSummaryIndex &CombinedIndex,
+ const SmallVector<const char *> &BitcodeLibFuncs);
/// Runs a ThinLTO backend.
/// If \p ModuleMap is not nullptr, all the module files to be imported have
@@ -62,6 +64,7 @@ thinBackend(const Config &C, unsigned Task, AddStreamFn AddStream, Module &M,
const FunctionImporter::ImportMapTy &ImportList,
const GVSummaryMapTy &DefinedGlobals,
MapVector<StringRef, BitcodeModule> *ModuleMap, bool CodeGenOnly,
+ const SmallVector<const char *> &BitcodeLibFuncs,
AddStreamFn IRAddStream = nullptr,
const std::vector<uint8_t> &CmdArgs = std::vector<uint8_t>());
diff --git a/llvm/lib/LTO/LTO.cpp b/llvm/lib/LTO/LTO.cpp
index 1d38120c6e642..95f06103689c1 100644
--- a/llvm/lib/LTO/LTO.cpp
+++ b/llvm/lib/LTO/LTO.cpp
@@ -762,6 +762,10 @@ Error LTO::add(std::unique_ptr<InputFile> Input,
return Error::success();
}
+void LTO::setBitcodeLibFuncs(const SmallVector<const char *> &BitcodeLibFuncs) {
+ this->BitcodeLibFuncs = &BitcodeLibFuncs;
+}
+
Expected<ArrayRef<SymbolResolution>>
LTO::addModule(InputFile &Input, ArrayRef<SymbolResolution> InputRes,
unsigned ModI, ArrayRef<SymbolResolution> Res) {
@@ -1386,7 +1390,8 @@ Error LTO::runRegularLTO(AddStreamFn AddStream) {
if (!RegularLTO.EmptyCombinedModule || Conf.AlwaysEmitRegularLTOObj) {
if (Error Err =
backend(Conf, AddStream, RegularLTO.ParallelCodeGenParallelismLevel,
- *RegularLTO.CombinedModule, ThinLTO.CombinedIndex))
+ *RegularLTO.CombinedModule, ThinLTO.CombinedIndex,
+ *BitcodeLibFuncs))
return Err;
}
@@ -1498,6 +1503,7 @@ class CGThinBackend : public ThinBackendProc {
class InProcessThinBackend : public CGThinBackend {
protected:
FileCache Cache;
+ const SmallVector<const char*> &BitcodeLibFuncs;
public:
InProcessThinBackend(
@@ -1505,11 +1511,12 @@ class InProcessThinBackend : public CGThinBackend {
ThreadPoolStrategy ThinLTOParallelism,
const DenseMap<StringRef, GVSummaryMapTy> &ModuleToDefinedGVSummaries,
AddStreamFn AddStream, FileCache Cache, lto::IndexWriteCallback OnWrite,
- bool ShouldEmitIndexFiles, bool ShouldEmitImportsFiles)
+ bool ShouldEmitIndexFiles, bool ShouldEmitImportsFiles,
+ const SmallVector<const char *> &BitcodeLibFuncs)
: CGThinBackend(Conf, CombinedIndex, ModuleToDefinedGVSummaries,
AddStream, OnWrite, ShouldEmitIndexFiles,
ShouldEmitImportsFiles, ThinLTOParallelism),
- Cache(std::move(Cache)) {}
+ Cache(std::move(Cache)), BitcodeLibFuncs(BitcodeLibFuncs) {}
virtual Error runThinLTOBackendThread(
AddStreamFn AddStream, FileCache Cache, unsigned Task, BitcodeModule BM,
@@ -1530,7 +1537,7 @@ class InProcessThinBackend : public CGThinBackend {
return thinBackend(Conf, Task, AddStream, **MOrErr, CombinedIndex,
ImportList, DefinedGlobals, &ModuleMap,
- Conf.CodeGenOnly);
+ Conf.CodeGenOnly, BitcodeLibFuncs);
};
if (ShouldEmitIndexFiles) {
if (auto E = emitFiles(ImportList, ModuleID, ModuleID.str()))
@@ -1615,13 +1622,14 @@ class FirstRoundThinBackend : public InProcessThinBackend {
const Config &Conf, ModuleSummaryIndex &CombinedIndex,
ThreadPoolStrategy ThinLTOParallelism,
const DenseMap<StringRef, GVSummaryMapTy> &ModuleToDefinedGVSummaries,
- AddStreamFn CGAddStream, FileCache CGCache, AddStreamFn IRAddStream,
+ AddStreamFn CGAddStream, FileCache CGCache,
+ const SmallVector<const char *> &BitcodeLibFuncs, AddStreamFn IRAddStream,
FileCache IRCache)
: InProcessThinBackend(Conf, CombinedIndex, ThinLTOParallelism,
ModuleToDefinedGVSummaries, std::move(CGAddStream),
std::move(CGCache), /*OnWrite=*/nullptr,
/*ShouldEmitIndexFiles=*/false,
- /*ShouldEmitImportsFiles=*/false),
+ /*ShouldEmitImportsFiles=*/false, BitcodeLibFuncs),
IRAddStream(std::move(IRAddStream)), IRCache(std::move(IRCache)) {}
Error runThinLTOBackendThread(
@@ -1644,7 +1652,7 @@ class FirstRoundThinBackend : public InProcessThinBackend {
return thinBackend(Conf, Task, CGAddStream, **MOrErr, CombinedIndex,
ImportList, DefinedGlobals, &ModuleMap,
- Conf.CodeGenOnly, IRAddStream);
+ Conf.CodeGenOnly, BitcodeLibFuncs, IRAddStream);
};
// Like InProcessThinBackend, we produce index files as needed for
// FirstRoundThinBackend. However, these files are not generated for
@@ -1711,6 +1719,7 @@ class SecondRoundThinBackend : public InProcessThinBackend {
ThreadPoolStrategy ThinLTOParallelism,
const DenseMap<StringRef, GVSummaryMapTy> &ModuleToDefinedGVSummaries,
AddStreamFn AddStream, FileCache Cache,
+ const SmallVector<const char *> &BitcodeLibFuncs,
std::unique_ptr<SmallVector<StringRef>> IRFiles,
stable_hash CombinedCGDataHash)
: InProcessThinBackend(Conf, CombinedIndex, ThinLTOParallelism,
@@ -1718,7 +1727,7 @@ class SecondRoundThinBackend : public InProcessThinBackend {
std::move(Cache),
/*OnWrite=*/nullptr,
/*ShouldEmitIndexFiles=*/false,
- /*ShouldEmitImportsFiles=*/false),
+ /*ShouldEmitImportsFiles=*/false, BitcodeLibFuncs),
IRFiles(std::move(IRFiles)), CombinedCGDataHash(CombinedCGDataHash) {}
Error runThinLTOBackendThread(
@@ -1739,7 +1748,7 @@ class SecondRoundThinBackend : public InProcessThinBackend {
return thinBackend(Conf, Task, AddStream, *LoadedModule, CombinedIndex,
ImportList, DefinedGlobals, &ModuleMap,
- /*CodeGenOnly=*/true);
+ /*CodeGenOnly=*/true, BitcodeLibFuncs);
};
if (!Cache.isValid() || !CombinedIndex.modulePaths().count(ModuleID) ||
all_of(CombinedIndex.getModuleHash(ModuleID),
@@ -1778,11 +1787,12 @@ ThinBackend lto::createInProcessThinBackend(ThreadPoolStrategy Parallelism,
auto Func =
[=](const Config &Conf, ModuleSummaryIndex &CombinedIndex,
const DenseMap<StringRef, GVSummaryMapTy> &ModuleToDefinedGVSummaries,
- AddStreamFn AddStream, FileCache Cache) {
+ AddStreamFn AddStream, FileCache Cache,
+ const SmallVector<const char *> &BitcodeLibFuncs) {
return std::make_unique<InProcessThinBackend>(
Conf, CombinedIndex, Parallelism, ModuleToDefinedGVSummaries,
AddStream, Cache, OnWrite, ShouldEmitIndexFiles,
- ShouldEmitImportsFiles);
+ ShouldEmitImportsFiles, BitcodeLibFuncs);
};
return ThinBackend(Func, Parallelism);
}
@@ -1899,7 +1909,8 @@ ThinBackend lto::createWriteIndexesThinBackend(
auto Func =
[=](const Config &Conf, ModuleSummaryIndex &CombinedIndex,
const DenseMap<StringRef, GVSummaryMapTy> &ModuleToDefinedGVSummaries,
- AddStreamFn AddStream, FileCache Cache) {
+ AddStreamFn AddStream, FileCache Cache,
+ const SmallVector<const char *> &BitcodeLibFuncs) {
return std::make_unique<WriteIndexesThinBackend>(
Conf, CombinedIndex, Parallelism, ModuleToDefinedGVSummaries,
OldPrefix, NewPrefix, NativeObjectPrefix, ShouldEmitImportsFiles,
@@ -2117,7 +2128,7 @@ Error LTO::runThinLTO(AddStreamFn AddStream, FileCache Cache,
if (!CodeGenDataThinLTOTwoRounds) {
std::unique_ptr<ThinBackendProc> BackendProc =
ThinLTO.Backend(Conf, ThinLTO.CombinedIndex, ModuleToDefinedGVSummaries,
- AddStream, Cache);
+ AddStream, Cache, *BitcodeLibFuncs);
return RunBackends(BackendProc.get());
}
@@ -2140,7 +2151,7 @@ Error LTO::runThinLTO(AddStreamFn AddStream, FileCache Cache,
LLVM_DEBUG(dbgs() << "[TwoRounds] Running the first round of codegen\n");
auto FirstRoundLTO = std::make_unique<FirstRoundThinBackend>(
Conf, ThinLTO.CombinedIndex, Parallelism, ModuleToDefinedGVSummaries,
- CG.AddStream, CG.Cache, IR.AddStream, IR.Cache);
+ CG.AddStream, CG.Cache, *BitcodeLibFuncs, IR.AddStream, IR.Cache);
if (Error E = RunBackends(FirstRoundLTO.get()))
return E;
@@ -2156,7 +2167,7 @@ Error LTO::runThinLTO(AddStreamFn AddStream, FileCache Cache,
LLVM_DEBUG(dbgs() << "[TwoRounds] Running the second round of codegen\n");
auto SecondRoundLTO = std::make_unique<SecondRoundThinBackend>(
Conf, ThinLTO.CombinedIndex, Parallelism, ModuleToDefinedGVSummaries,
- AddStream, Cache, IR.getResult(), CombinedHash);
+ AddStream, Cache, *BitcodeLibFuncs, IR.getResult(), CombinedHash);
return RunBackends(SecondRoundLTO.get());
}
@@ -2540,7 +2551,8 @@ ThinBackend lto::createOutOfProcessThinBackend(
auto Func =
[=](const Config &Conf, ModuleSummaryIndex &CombinedIndex,
const DenseMap<StringRef, GVSummaryMapTy> &ModuleToDefinedGVSummaries,
- AddStreamFn AddStream, FileCache /*Cache*/) {
+ AddStreamFn AddStream, FileCache /*Cache*/,
+ const SmallVector<const char *> &BitcodeLibFuncs) {
return std::make_unique<OutOfProcessThinBackend>(
Conf, CombinedIndex, Parallelism, ModuleToDefinedGVSummaries,
AddStream, OnWrite, ShouldEmitIndexFiles, ShouldEmitImportsFiles,
diff --git a/llvm/lib/LTO/LTOBackend.cpp b/llvm/lib/LTO/LTOBackend.cpp
index db7c449cca75c..38486d755f906 100644
--- a/llvm/lib/LTO/LTOBackend.cpp
+++ b/llvm/lib/LTO/LTOBackend.cpp
@@ -376,7 +376,8 @@ static bool isEmptyModule(const Module &Mod) {
bool lto::opt(const Config &Conf, TargetMachine *TM, unsigned Task, Module &Mod,
bool IsThinLTO, ModuleSummaryIndex *ExportSummary,
const ModuleSummaryIndex *ImportSummary,
- const std::vector<uint8_t> &CmdArgs) {
+ const std::vector<uint8_t> &CmdArgs,
+ const SmallVector<const char *> &BitcodeLibFuncs) {
llvm::TimeTraceScope timeScope("opt");
if (EmbedBitcode == LTOBitcodeEmbedding::EmbedPostMergePreOptimized) {
// FIXME: the motivation for capturing post-merge bitcode and command line
@@ -564,7 +565,8 @@ Error lto::finalizeOptimizationRemarks(LLVMRemarkFileHandle DiagOutputFile) {
Error lto::backend(const Config &C, AddStreamFn AddStream,
unsigned ParallelCodeGenParallelismLevel, Module &Mod,
- ModuleSummaryIndex &CombinedIndex) {
+ ModuleSummaryIndex &CombinedIndex,
+ const SmallVector<const char *> &BitcodeLibFuncs) {
llvm::TimeTraceScope timeScope("LTO backend");
Expected<const Target *> TOrErr = initAndLookupTarget(C, Mod);
if (!TOrErr)
@@ -576,7 +578,7 @@ Error lto::backend(const Config &C, AddStreamFn AddStream,
if (!C.CodeGenOnly) {
if (!opt(C, TM.get(), 0, Mod, /*IsThinLTO=*/false,
/*ExportSummary=*/&CombinedIndex, /*ImportSummary=*/nullptr,
- /*CmdArgs*/ std::vector<uint8_t>()))
+ /*CmdArgs*/ std::vector<uint8_t>(), BitcodeLibFuncs))
return Error::success();
}
@@ -616,7 +618,9 @@ Error lto::thinBackend(const Config &Conf, unsigned Task, AddStreamFn AddStream,
const FunctionImporter::ImportMapTy &ImportList,
const GVSummaryMapTy &DefinedGlobals,
MapVector<StringRef, BitcodeModule> *ModuleMap,
- bool CodeGenOnly, AddStreamFn IRAddStream,
+ bool CodeGenOnly,
+ const SmallVector<const char *> &BitcodeLibFuncs,
+ AddStreamFn IRAddStream,
const std::vector<uint8_t> &CmdArgs) {
llvm::TimeTraceScope timeScope("Thin backend", Mod.getModuleIdentifier());
Expected<const Target *> TOrErr = initAndLookupTarget(Conf, Mod);
@@ -655,7 +659,7 @@ Error lto::thinBackend(const Config &Conf, unsigned Task, AddStreamFn AddStream,
// Perform optimization and code generation for ThinLTO.
if (!opt(Conf, TM, Task, Mod, /*IsThinLTO=*/true,
/*ExportSummary=*/nullptr, /*ImportSummary=*/&CombinedIndex,
- CmdArgs))
+ CmdArgs, BitcodeLibFuncs))
return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
// Save the current module before the first codegen round.
diff --git a/llvm/lib/LTO/LTOCodeGenerator.cpp b/llvm/lib/LTO/LTOCodeGenerator.cpp
index 8aa404da15286..599c5c2eb5f84 100644
--- a/llvm/lib/LTO/LTOCodeGenerator.cpp
+++ b/llvm/lib/LTO/LTOCodeGenerator.cpp
@@ -614,7 +614,7 @@ bool LTOCodeGenerator::optimize() {
TargetMach = createTargetMachine();
if (!opt(Config, TargetMach.get(), 0, *MergedModule, /*IsThinLTO=*/false,
/*ExportSummary=*/&CombinedIndex, /*ImportSummary=*/nullptr,
- /*CmdArgs*/ std::vector<uint8_t>())) {
+ /*CmdArgs*/ std::vector<uint8_t>(), {})) {
emitError("LTO middle-end optimizations failed");
return false;
}
@@ -639,7 +639,7 @@ bool LTOCodeGenerator::compileOptimized(AddStreamFn AddStream,
Config.CodeGenOnly = true;
Error Err = backend(Config, AddStream, ParallelismLevel, *MergedModule,
- CombinedIndex);
+ CombinedIndex, {});
assert(!Err && "unexpected code-generation failure");
(void)Err;
>From 33aa80b6751187fbaedc83a7bf3ef22322c5a295 Mon Sep 17 00:00:00 2001
From: Daniel Thornburgh <mysterymath at gmail.com>
Date: Sat, 25 Oct 2025 15:48:00 -0700
Subject: [PATCH 6/7] Actually use the passed in bitcode libfuncs
---
llvm/lib/LTO/LTOBackend.cpp | 27 ++++++++++++++++++++-------
1 file changed, 20 insertions(+), 7 deletions(-)
diff --git a/llvm/lib/LTO/LTOBackend.cpp b/llvm/lib/LTO/LTOBackend.cpp
index 38486d755f906..0f4151dda3a91 100644
--- a/llvm/lib/LTO/LTOBackend.cpp
+++ b/llvm/lib/LTO/LTOBackend.cpp
@@ -239,7 +239,8 @@ createTargetMachine(const Config &Conf, const Target *TheTarget, Module &M) {
static void runNewPMPasses(const Config &Conf, Module &Mod, TargetMachine *TM,
unsigned OptLevel, bool IsThinLTO,
ModuleSummaryIndex *ExportSummary,
- const ModuleSummaryIndex *ImportSummary) {
+ const ModuleSummaryIndex *ImportSummary,
+ const DenseSet<StringRef> &BitcodeLibFuncs) {
std::optional<PGOOptions> PGOOpt;
if (!Conf.SampleProfile.empty())
PGOOpt = PGOOptions(Conf.SampleProfile, "", Conf.ProfileRemapping,
@@ -283,14 +284,23 @@ static void runNewPMPasses(const Config &Conf, Module &Mod, TargetMachine *TM,
TLII->disableAllFunctions();
TargetLibraryInfo TLI(*TLII);
- for (unsigned I = 0, E = static_cast<unsigned>(LibFunc::NumLibFuncs);
- I != E; ++I) {
+ for (unsigned I = 0, E = static_cast<unsigned>(LibFunc::NumLibFuncs); I != E;
+ ++I) {
LibFunc F = static_cast<LibFunc>(I);
- GlobalValue *Val = Mod.getNamedValue(TLI.getName(F));
+ StringRef Name = TLI.getName(F);
+ GlobalValue *Val = Mod.getNamedValue(Name);
+
+ // LibFuncs present in the current TU can always be referenced.
+ if (Val && !Val->isDeclaration())
+ continue;
+
+ // LibFuncs not implemented in bitcode can always be referenced.
+ if (!BitcodeLibFuncs.contains(Name))
+ continue;
+
// TODO: Non-lto functions should not be disabled, nor should functions that
// are somewhere in a ThinLTO link (just not imported in this module).
- if (!Val || Val->isDeclaration())
- TLII->setUnavailable(F);
+ TLII->setUnavailable(F);
}
FAM.registerPass([&] { return TargetLibraryAnalysis(*TLII); });
@@ -402,9 +412,12 @@ bool lto::opt(const Config &Conf, TargetMachine *TM, unsigned Task, Module &Mod,
// analysis in the case of a ThinLTO build where this might be an empty
// regular LTO combined module, with a large combined index from ThinLTO.
if (!isEmptyModule(Mod)) {
+ DenseSet<StringRef> BitcodeLibFuncsSet;
+ for (const char *F : BitcodeLibFuncs)
+ BitcodeLibFuncsSet.insert(F);
// FIXME: Plumb the combined index into the new pass manager.
runNewPMPasses(Conf, Mod, TM, Conf.OptLevel, IsThinLTO, ExportSummary,
- ImportSummary);
+ ImportSummary, BitcodeLibFuncsSet);
}
return !Conf.PostOptModuleHook || Conf.PostOptModuleHook(Task, Mod);
}
>From 4388d283cb2628e94bf87f169962bec94ee2bee8 Mon Sep 17 00:00:00 2001
From: Daniel Thornburgh <mysterymath at gmail.com>
Date: Thu, 13 Nov 2025 15:43:57 -0800
Subject: [PATCH 7/7] Fix wild pointer issue
---
llvm/include/llvm/LTO/LTO.h | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/llvm/include/llvm/LTO/LTO.h b/llvm/include/llvm/LTO/LTO.h
index 346ff9817737f..d0a447df69bf3 100644
--- a/llvm/include/llvm/LTO/LTO.h
+++ b/llvm/include/llvm/LTO/LTO.h
@@ -604,7 +604,8 @@ class LTO {
// Diagnostic optimization remarks file
LLVMRemarkFileHandle DiagnosticOutputFile;
- const SmallVector<const char *> *BitcodeLibFuncs;
+ const SmallVector<const char *> *BitcodeLibFuncs = &EmptyBitcodeLibFuncs;
+ SmallVector<const char *> EmptyBitcodeLibFuncs;
};
/// The resolution for a symbol. The linker must provide a SymbolResolution for
More information about the llvm-commits
mailing list