This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Improve code generated by cmpstrsi patterns
- To: Jan Hubicka <jh at suse dot cz>
- Subject: Re: Improve code generated by cmpstrsi patterns
- From: Jakub Jelinek <jakub at redhat dot com>
- Date: Sun, 16 Jul 2000 01:28:16 +0200
- Cc: zack at wolery dot cumb dot org, gcc-patches at gcc dot gnu dot org
- References: <20000716011745.A23991@atrey.karlin.mff.cuni.cz>
- Reply-To: Jakub Jelinek <jakub at redhat dot com>
On Sun, Jul 16, 2000 at 01:17:45AM +0200, Jan Hubicka wrote:
> Zack,
> Nice trick :)
> Everything looks perfect for me, except for the
> (set (reg:CC flags) (reg:CC flags))
> It is not first time I see similar set getting out of combine - other common
> case is (set (pc) (pc)) - and thus combine is able to remove conditional only
> when it proves it to be taken, not conditionals that are not taken. This is
> closely related to one of older Jakub's testsuites. Combine then constructed
> instructions, that was obviously constant, reload created an constant
> operand, that attempted to reload again and because of VOIDmode in the
> match_operand it failed to do so creating move to VOIDmode.
>
> Instead of adding such a dummy patterns to each individual machine
> description, I think it would be great to teach combine to zap them.
>
> I was trying that in past, but didn't succeeded because of lack of time,
> so I would like to tell you about this second special case and motivate
> you to take a look at this issue. I will try to teach combine about killing
> such moves tomorrow in case I will dig out the Jakub's testcase.
testsuite/gcc.dg/20000614-1.c
Jakub