AstValueFactory: add a cache for one-character strings.

src/ast/ast-value-factory.cc

Index: src/ast/ast-value-factory.cc
diff --git a/src/ast/ast-value-factory.cc b/src/ast/ast-value-factory.cc
index ed2976f52a98b15ef6872e9363bd272c6dd7e9e9..6097ff714598e40bba679a48beb401c3194e3cd6 100644
--- a/src/ast/ast-value-factory.cc
+++ b/src/ast/ast-value-factory.cc
@@ -222,6 +222,15 @@ void AstValue::Internalize(Isolate* isolate) {
 
 AstRawString* AstValueFactory::GetOneByteStringInternal(
     Vector<const uint8_t> literal) {
+  if (literal.length() == 1 && literal[0] >= 'a' && literal[0] <= 'z') {

[Toon Verwaest, 2016/12/05 15:30:40]

IsInRange(literal[0], 'a', 'z') is faster ;)

[marja, 2016/12/08 15:09:59]

Done.

+    int key = literal[0] - 'a';
+    if (one_character_strings_[key] == nullptr) {
+      uint32_t hash = StringHasher::HashSequentialString<uint8_t>(

[Toon Verwaest, 2016/12/05 15:30:40]

We could just have a-z heapstrings preallocated in a similar table and just
manually fill in without internalization lookup in the symbol table. Or in a
follow-up CL :)

[marja, 2016/12/08 15:09:59]

Hmm yeah, sounds like a good idea to keep the CLs separate. (I'd also like to
see the potential perf gain from this CL alone.)

+          literal.start(), literal.length(), hash_seed_);
+      one_character_strings_[key] = GetString(hash, true, literal);
+    }
+    return one_character_strings_[key];
+  }
   uint32_t hash = StringHasher::HashSequentialString<uint8_t>(
       literal.start(), literal.length(), hash_seed_);
   return GetString(hash, true, literal);