This is the mail archive of the
egcs-patches@egcs.cygnus.com
mailing list for the EGCS project.
Re: PPC aborts
- To: David Edelsohn <dje@watson.ibm.com>
- Subject: Re: PPC aborts
- From: Jeffrey A Law <law@cygnus.com>
- Date: Tue, 13 Jul 1999 00:07:47 -0600
- cc: meissner@cygnus.com, egcs-patches@egcs.cygnus.com
- Reply-To: law@cygnus.com
In message <9907121737.AA31758@marc.watson.ibm.com>you write:
> I completely agree with the patch. There have been too many FP
> constants forced into memory and this clearly is the cause.
I installed the rs6000.md movsf expander fix a few minutes ago to both the
branch and the mainline sources.
I briefly played with tweaking PREFERRED_RELOAD_CLASS per my suggestion for
the mainline tree It caused some problems building lapack, but I did not
investigate them in any detail. It may simply be the case that I didn't tweak
PREFERRED_RELOAD_CLASS correctly. I'll file it away and take a look at it
in the future when we're not trying to get a release out the door :-)
I've also identified the code generation bug which is causing tests to fail.
We're using r0 as a base register in an indexed memory reference. I've got
a lapack test running right now (looks good so far). I'll send a more detailed
analysis and patch shortly.
jeff