[clang] [llvm] [LTO] Reduce memory usage for import lists (PR #106772)

[Kazu Hirata via cfe-commits]

================
@@ -174,51 +174,33 @@ std::string llvm::computeLTOCacheKey(
   for (auto GUID : ExportsGUID)
     Hasher.update(ArrayRef<uint8_t>((uint8_t *)&GUID, sizeof(GUID)));
 
-  // Include the hash for every module we import functions from. The set of
-  // imported symbols for each module may affect code generation and is
-  // sensitive to link order, so include that as well.
-  using ImportMapIteratorTy =
-      FunctionImporter::ImportMapTy::ImportMapTyImpl::const_iterator;
-  struct ImportModule {
-    ImportMapIteratorTy ModIt;
-    const ModuleSummaryIndex::ModuleInfo *ModInfo;
-
-    StringRef getIdentifier() const { return ModIt->getFirst(); }
-    const FunctionImporter::FunctionsToImportTy &getFunctions() const {
-      return ModIt->second;
-    }
-
-    const ModuleHash &getHash() const { return ModInfo->second; }
-  };
-
-  std::vector<ImportModule> ImportModulesVector;
-  ImportModulesVector.reserve(ImportList.getImportMap().size());
-
-  for (ImportMapIteratorTy It = ImportList.getImportMap().begin();
-       It != ImportList.getImportMap().end(); ++It) {
-    ImportModulesVector.push_back({It, Index.getModule(It->getFirst())});
-  }
   // Order using module hash, to be both independent of module name and
   // module order.
-  llvm::sort(ImportModulesVector,
-             [](const ImportModule &Lhs, const ImportModule &Rhs) -> bool {
-               return Lhs.getHash() < Rhs.getHash();
-             });
-  std::vector<std::pair<uint64_t, uint8_t>> ImportedGUIDs;
-  for (const ImportModule &Entry : ImportModulesVector) {
-    auto ModHash = Entry.getHash();
-    Hasher.update(ArrayRef<uint8_t>((uint8_t *)&ModHash[0], sizeof(ModHash)));
-
-    AddUint64(Entry.getFunctions().size());
-
-    ImportedGUIDs.clear();
-    for (auto &[Fn, ImportType] : Entry.getFunctions())
-      ImportedGUIDs.push_back(std::make_pair(Fn, ImportType));
-    llvm::sort(ImportedGUIDs);
-    for (auto &[GUID, Type] : ImportedGUIDs) {
-      AddUint64(GUID);
-      AddUint8(Type);
+  auto Comp = [&](const std::pair<StringRef, GlobalValue::GUID> &L,
+                  const std::pair<StringRef, GlobalValue::GUID> &R) {
+    return std::make_pair(Index.getModule(L.first)->second, L.second) <
----------------
kazutakahirata wrote:

Yes, that's true unfortunately.  If the performance ever becomes a problem here, we can do all the hash lookups before sorting.

https://github.com/llvm/llvm-project/pull/106772