On ARM the code looks like this:
if (result_type_ == TRBinaryOpIC::UNINITIALIZED ||
result_type_ == TRBinaryOpIC::SMI) {
// Only allow smi results.
GenerateSmiCode(masm, NULL, NO_HEAPNUMBER_RESULTS);
} else {
// Allow heap number result and don't make a transition if a heap number
// cannot be allocated.
GenerateSmiCode(masm, &call_runtime, ALLOW_HEAPNUMBER_RESULTS);
}
// Code falls through if the result is not returned as either a smi or
heap
// number.
GenerateTypeTransition(masm);
__ bind(&call_runtime);
GenerateCallRuntime(masm);
It is basically the same, but maybe we should make them be the same. Making
a
runtime call instead of a type transition when allocation fails is the right
thing to do, but in practice it will very rarely make a difference.
http://codereview.chromium.org/6803010/
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
