Re: GCC 3.1.1 Status

[Bernd Schmidt <bernds at redhat dot com>]

On Wed, 19 Jun 2002, Jan Hubicka wrote:

> > On Tue, 18 Jun 2002, Gerald Pfeifer wrote:
> > 
> > > On Tue, 18 Jun 2002, Mark Mitchell wrote:
> > > > There are about ten high-priority bugs in GNATS that we should try to
> > > > fix before the release.
> > > 
> > > One is http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view&pr=6957
> > > 
> > >   could not find a spill register
> > >   (insn 96 94 97 (set (subreg:SF (reg:QI 75) 0)
> > >         (plus:SF (reg:SF 8 st(0) [76])
> > >             (reg:SF 9 st(1) [80]))) 525 {*fop_sf_comm_nosse}
> > >   (insn_list 87 (nil))
> > >     (expr_list:REG_DEAD (reg:SF 8 st(0) [76])
> > >         (nil)))
> > >   ../../../../../extras/Mesa/src/trans_tmp.h:124: Internal compiler error in
> > >   failed_reload, at reload1.c:5050
> > > 
> > > which really caused quite some troubles in FreeBSD land (and which I
> > > was able to reproduce with the current GCC 3.1 branch).
> > 
> > I'm currently looking into this one.
> 
> I think it is equivalent to 6845 that has small testcase.  THere is
> also 7034.  I think both should be high prioerity as they are
> regressions, but I am not sure whether I can change the prioerity right
> now.
> 
> I was tracking it down a bit and the problem appears to be in register
> used in very contradictionary ways sometimes as integer, sometimes as
> float using subreg.
> 
> The register costs are high for fp<->integer conversions and somehow the
> cheapest comes out MMX_CLASS. Register allocation decides to put
> register there and later we die.
> 
> I am not sure how to fix that properly...  Hope that helps :)

Well, I'm relatively sure how to fix it properly - disallow paradoxical
FLOAT_MODE subregs like the one we have up there.  This kind of rtl seems
a bit dodgy at best.  But before I commit the patch I have I'll need to
figure out why exactly reload blows up; that may or may not be a bug.


Bernd