Re: [PATCH, i386]: Macroize float operators

["Andrew Pinski" <pinskia at gmail dot com>]

On Sat, May 31, 2008 at 2:47 AM, Uros Bizjak <ubizjak@gmail.com> wrote:
> Hello!
>
> Attached patch macroizes various float operators. Patch was bootstrapped and
> regression tested on i686-pc-linux-gnu and x86_64-pc-linux-gnu {,-m32}.
> Patch is committed to mainline.

This causes a regression on i386-darwin on gcc.target/i386/pr26778.c .
Compiling it with -O2 -msse -mno-sse2 -mfpmath=sse on i686-linux-gnu
should produce the ICE.
The ICE we get is:
(insn 9 8 10 3 gcc/gcc/testsuite/gcc.target/i386/pr26778.c:12 (set (reg:DF 64)
        (minus:DF (reg/v:DF 58 [ th ])
            (reg:DF 61 [ D.1397 ]))) -1 (nil))

As we have DF mode and we turned off sse2, there is no pattern to
match this instruction or any DF mode as they are all turned off via
TARGET_SSE_MATH option.

Thanks,
Andrew Pinski