This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: AIX regression due to DFA scheduler merge
- From: David Edelsohn <dje at watson dot ibm dot com>
- To: Vladimir Makarov <vmakarov at redhat dot com>
- Cc: Jeffrey Law <law at redhat dot com>, gcc-patches at gcc dot gnu dot org, Mark Mitchell <mark at codesourcery dot com>
- Date: Fri, 31 May 2002 14:08:20 -0400
- Subject: Re: AIX regression due to DFA scheduler merge
>>>>> Vladimir Makarov writes:
Vlad> I see the following solutions of the problem:
Vlad> o as REG_NO_CONFLICT is inside libcalls we could schedule libcall as
Vlad> group.
Vlad> o take REG_NO_CONFLICT into account during building dependencies.
Vlad> The 1st solution is easier but it is wrong because it means a
Vlad> degradation and the advantages of gcc libcalls are not used in
Vlad> optimization.
Scheduling libcalls as groups sounds like the patch to treat
libcalls as SCHED_GROUP that Jeff and I have been discussing for a while.
The problem is that the current version of the patch causes regressions
due to other latent bugs which I have not been able to track down yet.
The libcall patch needs to go in eventually because of increased
register pressure caused by the CLOBBERs. Do you want to work on the
REG_NO_CONFLICT patch or do you want to help me track down the remaining
latent bug holding up the libcall patch?
I think that we really need the libcall patch finished because we
are finding more and more failures due to libcalls. The libcall patch
only limits scheduling in pass 1. The libcalls have been removed by the
second scheduling pass, so those instructions can be shifted slightly.
http://gcc.gnu.org/ml/gcc-patches/2002-05/msg02307.html
Thanks, David