new FAILs on HEAD

[Roger Sayle]

Hi Richard,

On Thu, 23 Dec 2004, Richard Henderson wrote:
> On Thu, Dec 23, 2004 at 10:40:00AM +0100, Michael Ritzert wrote:
> > These new FAILs appeared since the last run:
> >
> > FAIL: gcc.dg/i386-387-2.c scan-assembler fsin
>
> Huh.  I must have only looked at the results of the new tests I added.
>
> This is testing exactly the opposite of what I would have expected --
> that explicit options take precidence over inferences from -march.
>
> Oh well.  Reverted.

Personally, I preferred your patch and think/agree that the testcases
gcc.dg/i386-387-2.c and gcc.dg/i386-387-6.c (as written) are counter
intuitive.

I think these tests are checking a side-effect of your original
x86_arch_always_fancy_math_387 change (which was to fix x87 emulator
issues on *BSD), rather than an intentional behaviour.

Given that you're the original author of those testcases, I'd support
a decision to remove the "-mno-fancy-math-387" command line option
from them (and perhaps also adding two new testcases that confirms
that -mno-fancy-math-387 overrides -march and *doesn't* generate x87
intrinsics).

What do you think?

Roger
--