This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug target/55939] [4.6/4.7/4.8 regression] gcc miscompiles gmp-5.0.5 on m68k-linux
- From: "joseph at codesourcery dot com" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: Wed, 30 Jan 2013 21:11:31 +0000
- Subject: [Bug target/55939] [4.6/4.7/4.8 regression] gcc miscompiles gmp-5.0.5 on m68k-linux
- Auto-submitted: auto-generated
- References: <bug-55939-4@http.gcc.gnu.org/bugzilla/>
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55939
--- Comment #20 from joseph at codesourcery dot com <joseph at codesourcery dot com> 2013-01-30 21:11:31 UTC ---
On Tue, 29 Jan 2013, jakub at gcc dot gnu.org wrote:
> Perhaps -fexcess-precision=standard might fix this too (and be less
> expensive).
Note that -fexcess-precision=standard is not fully implemented for m68k:
the back-end patterns claiming to carry out SFmode/DFmode operations, that
actually do XFmode arithmetic, haven't been disabled for
-fexcess-precision=standard as I did for x86 (along with some related
fixes for conversions to SFmode/DFmode). While the disabling of SFmode /
DFmode operations is just for safety - if the architecture-independent
parts of GCC are working correctly, it shouldn't matter whether they are
enabled or not because there will be no arithmetic in those modes in the
GIMPLE anyway - I don't know if m68k needs similar fixes to conversion
operations.