This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: 3.3.4-prerelease fails to build on alphaev6-dec-osf5.1
- From: Roger Sayle <roger at eyesopen dot com>
- To: Gabriel Dos Reis <gdr at integrable-solutions dot net>, Joern Rennecke <joern dot rennecke at superh dot com>, Eric Botcazou <ebotcazou at libertysurf dot fr>
- Cc: Richard Henderson <rth at redhat dot com>, Joe Buck <Joe dot Buck at synopsys dot COM>, <gcc at gcc dot gnu dot org>
- Date: Sun, 23 May 2004 22:48:16 -0600 (MDT)
- Subject: Re: 3.3.4-prerelease fails to build on alphaev6-dec-osf5.1
On Sun, 23 May 2004, Richard Henderson wrote:
> Note that it doesn't happen during stage1, but during stage2;
> genattrtab is being miscompiled. It's not immediately obvious
> what's wrong. Perhaps someone can do a search?
The regression hunt on alphaev67-dec-osf5.1 converges on the following:
2004-03-01 Gabriel Dos Reis <gdr@integrable-solutions.net>
Backport from 3.4.0:
2004-01-16 J"orn Rennecke <joern.rennecke@superh.com>
PR optimization/11864
* reload1.c (reload_cse_simplify_operands): Don't remove
implicit extension from LOAD_EXTEND_OP.
And indeed reverting the change between 1.366.2.7 and 1.366.2.8 of
reload1.c against current gcc-3_3-branch permits this target to
proceed past this failure.
Joern's fix to PR optimization/11864 was known to cause problems for
some 64-bit targets, including Sparc64. Indeed this fix wasn't backported
to the 3.3 branch until after Eric Botcazou's fix for the fallout.
2004-02-29 Gabriel Dos Reis <gdr@integrable-solutions.net>
Backport from mainline:
2004-01-27 Eric Botcazou <ebotcazou@libertysurf.fr>
PR target/10904
PR target/13058
* config/sparc/sparc.h (CANNOT_CHANGE_MODE_CLASS): New.
Forbid mode changes from SImode for lower FP regs if ARCH64.
Almost prophetically Eric's comment #22 on bugzilla PR 11864 reads
"Beware, however, that other 64-bit ports might be affected by the
same problem." It's getting late, so I haven't yet checked if there's
a CANNOT_CHANGE_MODE_CLASS patch for alpha on mainline, which might
resolve this failure if backported.
I hope this helps.
Roger
--