[PATCH] D60297: [COFF] Pack Name in Symbol as is done in ELF

[Reid Kleckner via Phabricator via llvm-commits]

rnk created this revision.
rnk added reviewers: ruiu, inglorion, amccarth, aganea.
Herald added a project: LLVM.

This assumes all symbols are <4GB long, so we can store them as a 32-bit
integer. This reorders the fields so the length appears first, packing
with the other bitfield data in the base Symbol object.

This saved 70MB / 3.60% of heap allocations when linking
browser_tests.exe with no PDB. It's not much as a percentage, but worth
doing. I didn't do performance measurements, I don't think it will be
measurable in time.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D60297

Files:
  lld/COFF/Symbols.cpp
  lld/COFF/Symbols.h


Index: lld/COFF/Symbols.h
===================================================================
--- lld/COFF/Symbols.h
+++ lld/COFF/Symbols.h
@@ -79,7 +79,8 @@
   explicit Symbol(Kind K, StringRef N = "")
       : SymbolKind(K), IsExternal(true), IsCOMDAT(false),
         WrittenToSymtab(false), PendingArchiveLoad(false), IsGCRoot(false),
-        IsRuntimePseudoReloc(false), Name(N) {}
+        IsRuntimePseudoReloc(false), NameLen(N.size()),
+        NameBase(N.empty() ? nullptr : N.data()) {}
 
   const unsigned SymbolKind : 8;
   unsigned IsExternal : 1;
@@ -106,7 +107,10 @@
   unsigned IsRuntimePseudoReloc : 1;
 
 protected:
-  StringRef Name;
+  // Symbol name length. Assume symbol lengths fit in a 32-bit integer.
+  unsigned NameLen;
+
+  const char *NameBase;
 };
 
 // The base class for any defined symbols, including absolute symbols,
@@ -129,7 +133,7 @@
 // Symbols defined via a COFF object file or bitcode file.  For COFF files, this
 // stores a coff_symbol_generic*, and names of internal symbols are lazily
 // loaded through that. For bitcode files, Sym is nullptr and the name is stored
-// as a StringRef.
+// as a decomposed StringRef.
 class DefinedCOFF : public Defined {
   friend Symbol;
 
Index: lld/COFF/Symbols.cpp
===================================================================
--- lld/COFF/Symbols.cpp
+++ lld/COFF/Symbols.cpp
@@ -20,6 +20,9 @@
 
 using namespace lld::coff;
 
+static_assert(sizeof(SymbolUnion) <= 48,
+              "symbols should be optimized for memory usage");
+
 // Returns a symbol name for an error message.
 std::string lld::toString(coff::Symbol &B) {
   if (Config->Demangle)
@@ -39,11 +42,15 @@
   // name. Object files contain lots of non-external symbols, and creating
   // StringRefs for them (which involves lots of strlen() on the string table)
   // is a waste of time.
-  if (Name.empty()) {
+  if (NameBase == nullptr) {
     auto *D = cast<DefinedCOFF>(this);
-    cast<ObjFile>(D->File)->getCOFFObj()->getSymbolName(D->Sym, Name);
+    StringRef NameStr;
+    cast<ObjFile>(D->File)->getCOFFObj()->getSymbolName(D->Sym, NameStr);
+    NameBase = NameStr.data();
+    NameLen = NameStr.size();
+    assert(NameLen == NameStr.size() && "name length truncated");
   }
-  return Name;
+  return StringRef(NameBase, NameLen);
 }
 
 InputFile *Symbol::getFile() {
@@ -67,9 +74,10 @@
 
 // MinGW specific.
 void Symbol::replaceKeepingName(Symbol *Other, size_t Size) {
-  StringRef OrigName = Name;
+  StringRef OrigName = getName();
   memcpy(this, Other, Size);
-  Name = OrigName;
+  NameBase = OrigName.data();
+  NameLen = OrigName.size();
 }
 
 COFFSymbolRef DefinedCOFF::getCOFFSymbol() {


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D60297.193813.patch
Type: text/x-patch
Size: 2670 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190405/dbd771e2/attachment.bin>