This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug target/52908] xop-mul-1:f9 miscompiled on bulldozer (-mxop)
- From: "ubizjak at gmail dot com" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: Fri, 04 May 2012 12:51:49 +0000
- Subject: [Bug target/52908] xop-mul-1:f9 miscompiled on bulldozer (-mxop)
- Auto-submitted: auto-generated
- References: <bug-52908-4@http.gcc.gnu.org/bugzilla/>
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52908
--- Comment #5 from Uros Bizjak <ubizjak at gmail dot com> 2012-05-04 12:51:49 UTC ---
(In reply to comment #4)
> A Quick make check on i386.exp result is shown below:
>
> Tests that now fail, but worked before:
>
> gcc.target/i386/xop-imul32widen-vector.c scan-assembler vpmacsdql
This is expected, there is no need to emit vpmacsdql, IMO vpmuldq works as good
(if not even better, since we don't have to preload accumulator with zero). The
test should be either changed to really emit vpmacsdql, or asm scan should be
adjusted.
> Tests that now work, but didn't before:
>
> gcc.target/i386/xop-mul-1.c execution test
Yes this one should be fixed with the patch.
If the change vmpacsdql -> vpmuldq is OK with you, I can commit the patch, but
I'd ask you for full bootstrap/regression test.