[Bug target/52568] New: suboptimal __builtin_shuffle on cycles with AVX

[Bug target/52568] New: suboptimal __builtin_shuffle on cycles with AVX

[marc.glisse at normalesup dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52568

             Bug #: 52568
           Summary: suboptimal __builtin_shuffle on cycles with AVX
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: marc.glisse@normalesup.org


Hello,
I compiled the following with -O3 (or -Os) and -mavx

#include <x86intrin.h>
__m256d left(__m256d x){
  __m256i mask={1,2,3,0};
  return __builtin_shuffle(x,mask);
}

(by the way, for some reason, gcc insists that 'mask' is set but not used with
-Wall)

and got:
    vunpckhpd    %xmm0, %xmm0, %xmm3
    vmovapd    %xmm0, %xmm1
    vextractf128    $0x1, %ymm0, %xmm0
    vmovaps    %xmm0, %xmm2
    vunpckhpd    %xmm0, %xmm0, %xmm0
    vunpcklpd    %xmm1, %xmm0, %xmm1
    vunpcklpd    %xmm2, %xmm3, %xmm0
    vinsertf128    $0x1, %xmm1, %ymm0, %ymm0
    ret

That doesn't really match the code I currently use to do this:
#ifdef __AVX2__
        __m256d d=_mm256_permute4x64_pd(x,1+2*4+3*16+0*64);
#else
        __m256d b=_mm256_shuffle_pd(x,x,5);
        __m256d c=_mm256_permute2f128_pd(b,b,1);
        __m256d d=_mm256_blend_pd(b,c,10);
#endif

Could something recognizing this permutation pattern (and the right cyclic
shift) be added? I know there are too many shuffles to hand-code them all, but
cycles seem like they shouldn't be too uncommon.

With -mavx2, I get a single vpermq, which is close enough to the expected
vpermpd.

[Bug target/52568] suboptimal __builtin_shuffle on cycles with AVX

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52568

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-03-13 08:36:29 UTC ---
-Wunused-but-set-* warning false positive tracked in PR52577.

[Bug target/52568] suboptimal __builtin_shuffle on cycles with AVX

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52568

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
             Target|                            |x86_64-*-*
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-03-13
     Ever Confirmed|0                           |1

--- Comment #2 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-03-13 08:44:06 UTC ---
Confirmed.

[Bug target/52568] suboptimal __builtin_shuffle on cycles with AVX

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52568

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-03-15 23:45:49 UTC ---
Author: jakub
Date: Thu Mar 15 23:45:45 2012
New Revision: 185446

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=185446
Log:
    PR target/52568
    * config/i386/sse.md (UNSPEC_VPERMDF): Remove.
    (avx2_permv4df): Remove.
    (avx2_permv4di): Macroize into...
    (avx2_perm<mode>): ... this using VI8F_256 iterator.
    (avx2_permv4di_1): Macroize into...
    (avx2_perm<mode>_1): ... this using VI8F_256 iterator.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/sse.md

[Bug target/52568] suboptimal __builtin_shuffle on cycles with AVX

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52568

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-03-15 23:46:37 UTC ---
Author: jakub
Date: Thu Mar 15 23:46:34 2012
New Revision: 185447

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=185447
Log:
    PR target/52568
    * config/i386/i386.c (expand_vec_perm_vperm2f128_vblend): New
    function.
    (ix86_expand_vec_perm_const_1): Use it.

    * gcc.dg/torture/vshuf-4.inc: Add two new tests.
    * gcc.dg/torture/vshuf-8.inc: Likewise.
    * gcc.dg/torture/vshuf-16.inc: Likewise.
    * gcc.dg/torture/vshuf-32.inc: Likewise.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/i386.c
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gcc.dg/torture/vshuf-16.inc
    trunk/gcc/testsuite/gcc.dg/torture/vshuf-32.inc
    trunk/gcc/testsuite/gcc.dg/torture/vshuf-4.inc
    trunk/gcc/testsuite/gcc.dg/torture/vshuf-8.inc

[Bug target/52568] suboptimal __builtin_shuffle on cycles with AVX

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52568

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-03-16 07:40:13 UTC ---
Should be fixed on the trunk.  For -mavx2 we now generate vpermpd, for -mavx
vpermilpd + vperm2f128 + vblendpd.