[Interpreter] Enable allocation site mementos in CreateArrayLiterals.

src/interpreter/bytecode-generator.cc

Index: src/interpreter/bytecode-generator.cc
diff --git a/src/interpreter/bytecode-generator.cc b/src/interpreter/bytecode-generator.cc
index 0afa8e2c4e95c803b4074d786f58f73e98316142..f19977a1e670db684ad2906046387005c093ed49 100644
--- a/src/interpreter/bytecode-generator.cc
+++ b/src/interpreter/bytecode-generator.cc
@@ -1882,8 +1882,7 @@ void BytecodeGenerator::VisitObjectLiteral(ObjectLiteral* expr) {
 void BytecodeGenerator::VisitArrayLiteral(ArrayLiteral* expr) {
   // Deep-copy the literal boilerplate.
   builder()->CreateArrayLiteral(expr->constant_elements(),
-                                expr->literal_index(),
-                                expr->ComputeFlags(true));

[mythria, 2016/09/02 15:06:53]

Ross, I think this is the only change that is required for enabling mementos. We
call in to the runtime which already handles it. Was there any reason we
disabled it earlier? When Klass finishes porting the FastCloneShallowObjectStub
I think it should still work.

[rmcilroy, 2016/09/05 09:28:33]

AST Graph Builder doesn't create mementos (presumably because we don't need to
collect any more feedback once we are in optimized code). Maybe we need to
explicitly prevent mementos in BytecodeGraphBuilder to be consistent with
ASTGraphBuilder. 

Michi, what do you think?

[Michael Starzinger, 2016/09/05 12:23:04]

Yes, correct. The reasoning was that only the unoptimized code would collect
feedback and once we optimize the decision was "settled". We decided to go with
this approach until we'd see data to the contrary. Being consistent for now and
preventing creation of mementos in the BytecodeGraphBuilder sounds reasonable to
me (with a comment, explaining exactly this). I also CCed Michael Stanton about
this.

+                                expr->literal_index(), expr->ComputeFlags());
   Register index, literal;
 
   // Evaluate all the non-constant subexpressions and store them into the