This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch] Fix excess-precision failures on darwin by passing -mfpmath=387


On Thu, Jun 11, 2009 at 11:26 AM, FX<fxcoudert@gmail.com> wrote:
> On Darwin, gcc.target/i386/excess-precision-{1,2,3}.c fail because the
> default is -mfpmap=sse, and the test expects i387 math. The attached patch
> fixes this by explicitly passing
> -mfpmath=387 as an option.
>
> Bootstrapped and regtested on i686-darwin (both -m32 and -m64) and
> x86_64-linux. OK to commit?

I think that -mfmpath=387 should be added to all excess-precision-X.c
tests. At least scan-assembler-not in e-p-6.c is fooled by SSE's
sqrtsd/sqrtsf.  OTOH, by using -mfpmath=387, we can remove requirement
for ilp32 effective target. This will enhance coverage for 64 bit
targers.

So, please add -mfpmath=387 and remove dg-require-effective-target
from all gcc.target/i386/excess-precision-X.c tests. The patch is
pre-approved if it works OK on 32 and 64bit targets.

Thanks,
Uros.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]