This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Gcc 3.1 performance regressions with respect to 2.95.3
- From: David Edelsohn <dje at watson dot ibm dot com>
- To: law at redhat dot com
- Cc: Michael Matz <matzmich at cs dot tu-berlin dot de>, gcc-patches at gcc dot gnu dot org
- Date: Tue, 26 Mar 2002 23:34:21 -0500
- Subject: Re: Gcc 3.1 performance regressions with respect to 2.95.3
>>>>> law writes:
Jeff> Yeah, I missed that step. Ugh. If we continue down this path we'd have to
Jeff> stall issuing the clobber until its dependent instructions are ready to
Jeff> fire, except for the clobber. What a mess!
Yes, I was thinking about issuing the CLOBBER in the same cycle as
the first SET because it really is just a placeholder, but I do not know
how to enforce the dependency order within the cycle. I think that
dependencies must issue in the previous cycle.
Jeff> If y'all can find a way to make the SCHED_GROUP_P stuff work it'll probably
Jeff> be cleaner. You'd still want code to not decrement can_issue_more when you
Jeff> issue the clobber though since you don't want the clobber to use up one of
Jeff> your issue slots.
Yes, I completely agree that Dale's patch also is good. I'm just
not sure if SCHED_GROUP_P will provide enough feedback to limit register
pressure. The proposed patch only addresses LIBCALL blocks, not all
CLOBBERs.
Thanks, David