This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: -mfpmath=sse -mno-80387 troubles
- From: tbp <tbptbp at gmail dot com>
- To: Uros Bizjak <uros at kss-loka dot si>
- Cc: gcc at gcc dot gnu dot org
- Date: Fri, 28 Jan 2005 16:54:51 +0100
- Subject: Re: -mfpmath=sse -mno-80387 troubles
- References: <41F8E9DA.9050504@kss-loka.si>
- Reply-to: tbp <tbptbp at gmail dot com>
On Thu, 27 Jan 2005 14:17:14 +0100, Uros Bizjak <uros@kss-loka.si> wrote:
> In latest CVS, these problems should be fixed. Please try to compile
> your code with the latest CVS gcc, and if it still crashes, please file
> a bugreport. If you spot an x87 constant, please add a testcase to
> PR19653. However ...
I've given the latest cvs (as of today, right now) a try.
No more inapropriate 387 use in the middle of SSE stuff and
performance are back within ICC range (give or take 5%) without
-mno-80387.
With -mno-80387 there's no ICE anymore, that's the good news, but the
binary spits some frames as expected and then only black ones; it's
not a straight dead lock or crash. I seriously doubt that switch
uncovers a bug in my code and i'll put the blame on the code gen.
My first pass to isolate the problem failed, i'll try harder to see
what went wrong.