--- Comment #10 from hjl dot tools at gmail dot com 2008-09-04 20:30
---
Created an attachment (id=16224)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16224&action=view)
A patch
I am testing this patch
2008-09-04 H.J. Lu <[EMAIL PROTECTED]>
PR target/37364
*
--- Comment #9 from hjl dot tools at gmail dot com 2008-09-04 20:17 ---
I am concerned about those "*Yi"/"*Ym" and "r" pairs:
[EMAIL PROTECTED] i386]$ grep "\*Yi" *.md
i386.md:"=r,m ,*y,*y,?rm,?*y,*x,*x,?r ,m ,?*Yi,*x")
i386.md:"g ,ri,C ,*
--- Comment #8 from rguenth at gcc dot gnu dot org 2008-09-04 20:12 ---
If we decide the test is bogus we should remove it, not xfail it.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37364
--- Comment #7 from hjl dot tools at gmail dot com 2008-09-04 19:45 ---
(In reply to comment #6)
> First of all, I've check the generated code on Core2 and I found it is not
> slower than using movd.
I think MMX may be slow anyway.
> The reason for this is in insn
>
> (insn:HI 14
--- Comment #6 from vmakarov at redhat dot com 2008-09-04 19:25 ---
First of all, I've check the generated code on Core2 and I found it is not
slower than using movd.
IRA assigns hard registers calculating their costs. It the memory is
cheaper, it assigns memory. The first decisio
--- Comment #5 from hjl dot tools at gmail dot com 2008-09-04 18:39 ---
Just for the record, move between MMX and GPR isn't a major issue
since we prefer SSE anyway. If it is the only inter class register
move problem for IRA, I am OK to close it as WONTFIX.
--
http://gcc.gnu.org/bu
