This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Patch to fix poor Altivec code gen in combine phase.
- From: Andrew Pinski <pinskia at physics dot uc dot edu>
- To: Fariborz Jahanian <fjahanian at apple dot com>
- Cc: "gcc-patches at gcc dot gnu dot org Patches" <gcc-patches at gcc dot gnu dot org>
- Date: Mon, 20 Jun 2005 13:41:56 -0400
- Subject: Re: Patch to fix poor Altivec code gen in combine phase.
- References: <49DFDB12-604E-420D-B116-03FC35E5269F@apple.com>
On Jun 20, 2005, at 1:32 PM, Fariborz Jahanian wrote:
This patch, suggested and approved by Roger Sayle, addresses an
inefficiency in altivec compare code gen on powerpc. This is because
combine phase fails to combine several rtl insns for comparing,
setting and generating jump_insn into a single jump_insn. Patch fixes
this by simplifying setting of a CC register by removing the redundant
compare with 0 on its rhs (old pattern does not recognize on powerpc).
Patch has been bootstrapped and dejagnu tested on ppc-darwin and
x86-darwin.
This is also PR 22077 and a regression from a 3.4.x. I had filed the
bug after
seeing it on the performance list and after Sanjay Patel said he filed
it (and
yes I replied saying I filed a FSF GCC PR for this bug).
Thanks,
Andrew Pinski