Updates:
Owner: [email protected]
Cc: [email protected] [email protected]
Comment #4 on issue 2458 by [email protected]: Very simple function of
float32array deopts on ia32
http://code.google.com/p/v8/issues/detail?id=2458
I think Slava's analysis is correct: If we used pcmpeqq instead of pcmpeqd,
the code would be correct, but pcmpeqq is an SSE 4.1 instruction, which we
can't use in our stubs. :-( So we are left with movmskps (which we have to
implement in our assembler) and an and/cmp combination, because AFAICT we
don't know the state of the upper half. Does anybody have a better and more
efficient solution?
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
