Updates:
Labels: HW-All
Comment #3 on issue 2248 by [email protected]: Wrong type feedback for
STRICT_EQ
http://code.google.com/p/v8/issues/detail?id=2248
Just a small update on this issue: It seems to be the case that GVN is
doing something strange here, not the type feedback for STRICT_EQ. With
--nouse-gvn, there are no deopts and the generated hydrogen code looks OK:
The loop variable i is represented as an unboxed integer, the constant 10
is tagged, and the initial === in eq is done on tagged values (tagging i),
which is OK, because we have seen Smis and Objects at that call site.
Investigating further...
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
