[LTO][ELF][lld] Use unique string saver in ELF bitcode symbol parsing (#106670)

lld ELF
[BitcodeFile](https://github.com/llvm/llvm-project/blob/a527248a3c2d638b0c92a06992f3f1c1f80842ad/lld/ELF/InputFiles.h#L328)
uses [string
saver](https://github.com/llvm/llvm-project/blob/a527248a3c2d638b0c92a06992f3f1c1f80842ad/lld/include/lld/Common/CommonLinkerContext.h#L57)
to keep copies of bitcode symbols. Symbol duplication is very common
when compiling application binaries.

This change proposes to introduce a UniqueStringSaver in lld context and
use it for bitcode symbol parsing. The implementation covers ELF only.
Similar opportunities should exist on other (COFF, MachO, wasm) formats.

For an internal production binary where lto indexing takes ~10GiB
originally, this changes optimizes away ~800MiB (~7.8%), measured by
https://github.com/google/pprof. Flame graph breaks down memory by usage
call stacks and agrees with this measurement.

Author: mingmingl-llvm

lld/ELF/InputFiles.cpp

@@ -1744,8 +1744,10 @@ createBitcodeSymbol(Symbol *&sym, const std::vector<bool> &keptComdats,
   uint8_t type = objSym.isTLS() ? STT_TLS : STT_NOTYPE;
   uint8_t visibility = mapVisibility(objSym.getVisibility());
 
+  // Symbols can be duplicated in bitcode files because of '#include' and
+  // linkonce_odr. Use unique_saver to save symbol names for de-duplication.
   if (!sym)
-    sym = symtab.insert(saver().save(objSym.getName()));
+    sym = symtab.insert(unique_saver().save(objSym.getName()));
 
   int c = objSym.getComdatIndex();
   if (objSym.isUndefined() || (c != -1 && !keptComdats[c])) {
@@ -1797,7 +1799,9 @@ void BitcodeFile::parseLazy() {
   symbols = std::make_unique<Symbol *[]>(numSymbols);
   for (auto [i, irSym] : llvm::enumerate(obj->symbols()))
     if (!irSym.isUndefined()) {
-      auto *sym = symtab.insert(saver().save(irSym.getName()));
+      // Symbols can be duplicated in bitcode files because of '#include' and
+      // linkonce_odr. Use unique_saver to save symbol names for de-duplication.
+      auto *sym = symtab.insert(unique_saver().save(irSym.getName()));
       sym->resolve(LazySymbol{*this});
       symbols[i] = sym;
     }

lld/include/lld/Common/CommonLinkerContext.h

@@ -38,6 +38,7 @@ class CommonLinkerContext {
 
   llvm::BumpPtrAllocator bAlloc;
   llvm::StringSaver saver{bAlloc};
+  llvm::UniqueStringSaver unique_saver{bAlloc};
   llvm::DenseMap<void *, SpecificAllocBase *> instances;
 
   ErrorHandler e;
@@ -54,8 +55,13 @@ template <typename T = CommonLinkerContext> T &context() {
 
 bool hasContext();
 
-inline llvm::StringSaver &saver() { return context().saver; }
 inline llvm::BumpPtrAllocator &bAlloc() { return context().bAlloc; }
+inline llvm::StringSaver &saver() { return context().saver; }
+inline llvm::UniqueStringSaver &unique_saver() {
+  // FIXME: Look into other places where duplications are common in saved
+  // strings and unique saver make sense.
+  return context().unique_saver;
+}
 } // namespace lld
 
 #endif