This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Ensure vcond* expansion doesn't fail on x86 (PR target/50310)
On Wed, Sep 7, 2011 at 10:01 PM, Jakub Jelinek <jakub@redhat.com> wrote:
> On Wed, Sep 07, 2011 at 09:54:03PM +0200, Uros Bizjak wrote:
>> > 2011-09-07 ?Jakub Jelinek ?<jakub@redhat.com>
>> >
>> > ? ? ? ?PR target/50310
>> > ? ? ? ?* config/i386/i386.c (ix86_prepare_sse_fp_compare_args): For
>> > ? ? ? ?TARGET_AVX return code for LTGT and UNEQ.
>> > ? ? ? ?(ix86_expand_fp_vcond): Handle LTGT and UNEQ.
>> >
>> > ? ? ? ?* gcc.c-torture/execute/ieee/pr50310.c: New test.
>> > ? ? ? ?* gcc.dg/pr50310-2.c: New test.
>>
>> Please put early exit for TARGET_SSE at the beginning of
>> ix86_prepare_sse_fp_compare_args function. There is really no need to
>
> You mean for TARGET_AVX, right?
>
>> swap operands - and to help reload, since AVX instructions are
>> three-operand instructions.
>>
>> OK for mainline with this change.
>
> Here is the updated patch, I'll bootstrap/regtest it now.
>
> 2011-09-07 ?Jakub Jelinek ?<jakub@redhat.com>
>
> ? ? ? ?PR target/50310
> ? ? ? ?* config/i386/i386.c (ix86_prepare_sse_fp_compare_args): Return
> ? ? ? ?code early if TARGET_AVX.
> ? ? ? ?(ix86_expand_fp_vcond): Handle LTGT and UNEQ.
>
> ? ? ? ?* gcc.c-torture/execute/ieee/pr50310.c: New test.
> ? ? ? ?* gcc.dg/pr50310-2.c: New test.
>
> --- gcc/config/i386/i386.c.jj ? 2011-09-02 16:29:38.000000000 +0200
> +++ gcc/config/i386/i386.c ? ? ?2011-09-07 21:57:52.000000000 +0200
> @@ -18304,6 +18304,11 @@ ix86_prepare_sse_fp_compare_args (rtx de
> ?{
> ? rtx tmp;
>
> + ?/* AVX supports all the needed comparisons, no need to swap arguments
> + ? ? nor help reload. ?*/
> + ?if (TARGET_AVX)
> + ? ?return code;
> +
Unfortunately, this part prevents generation of vmin/vmax instructions
for TARGET_AVX. In ix86_expand_fp_movcc, we call
ix86_prepare_sse_fp_compare_args, where we previously converted GT
into LT. LT is recognized in ix86_expand_sse_fp_minmax as valid
operand for MIN/MAX, whereas GT is not.
I'm not sure if we can swap operands in ix86_expand_sse_fp_minmax,
there is a scary comment in front of ix86_expand_sse_fp_minmax w.r.t.
to IEEE safety.
The test is gcc.target/i386/ssefp-?.c, please compile it with "-O2 -mavx".
Uros.