[Bug middle-end/28367] accessing via union on a vector does not cause vec_extract to be used

[Bug middle-end/28367] accessing via union on a vector does not cause vec_extract to be used

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28367

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |rguenth at gcc dot gnu.org
                   |gnu.org                     |

--- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-03-16 11:41:03 UTC ---
comment #1 is fixed, the original problem persists.  optimized dump:

;; Function f (f)

f (vector(4) float t)
{
  union
  {
    vector(4) float t1;
    float t2[4];
  } t5;
  float D.2687;

<bb 2>:
  t5.t1 = t_1(D);
  D.2687_2 = t5.t2[0];
  return D.2687_2;

}

FRE does not know how to handle different sizes in its support for
type-punning stores/loads (visit_reference_op_load).

Mine.

[Bug middle-end/28367] accessing via union on a vector does not cause vec_extract to be used

[glisse at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=28367

Marc Glisse <glisse at gcc dot gnu.org> changed:

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

--- Comment #6 from Marc Glisse <glisse at gcc dot gnu.org> ---
Created attachment 33674
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33674&action=edit
rough patch

It needs more work (seems more complicated than it should be), but at least it
managed to produce what I expected for:

#include <x86intrin.h>
union u { __v4sf v; float a[4]; };

float f(__v4sf x){
  u t;
  t.v=x;
  return t.a[2];
}

[Bug middle-end/28367] accessing via union on a vector does not cause vec_extract to be used

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=28367

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
Yeah, something along that line would be needed.  The issue is that
the partial value is not "available" (as in: it doesn't have a SSA name)
and thus there is nothing to CSE the load to.

I suppose that we should generalize this support for real-/imag-part of
complex values as well.  Not sure if it is worth using BIT_FIELD_REF
for getting at pieces of scalar registers (maybe it would pay off for
targets that can efficiently compute lowpart subregs).

[Bug middle-end/28367] New: accessing via union on a vector does not cause vec_extract to be used

[pinskia at gcc dot gnu dot org]

Take the following code and compile with -O2 -m64 -maltivec -mabi=altivec:
#define vector __attribute__((vector_size(16)))
float f(vector float t)
{
  union {
    vector float t1;
    float t2[4];
  }t5;
  t5.t1 = t;
  return t5.t2[0];
}

Currently we get:
        addi 12,1,-32
        stvx 2,0,12
        ld 11,0(12)
        srdi 9,11,32
        stw 9,-16(1)
        lfs 1,-16(1)
        blr

Which is bad for at least the Cell as we now hit two store load hazards.  We
can remove one if the compile uses the vec_extract patterns causing us to use
stvewx and lfs.

This does not effect x86 as the union is recorded as BLKmode but x86_64 has the
same issue, maybe even worse as there is no need to go through memory there.
X86_64 produces:
f:
.LFB2:
        movaps  %xmm0, -24(%rsp)
        movss   -24(%rsp), %xmm0
        ret
When really it should produce just a ret.


-- 
           Summary: accessing via union on a vector does not cause
                    vec_extract to be used
           Product: gcc
           Version: 4.2.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: enhancement
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: pinskia at gcc dot gnu dot org
GCC target triplet: powerpc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28367

[Bug middle-end/28367] accessing via union on a vector does not cause vec_extract to be used

[pinskia at gcc dot gnu dot org]

------- Comment #1 from pinskia at gcc dot gnu dot org  2006-07-13 05:37 -------
Note a related testcase is:
#define vector __attribute__((vector_size(16)))
float f(vector float t)
{
  return *(float*)(&t);
}

Which is only mentioned here in get_alias_set in the GCC sources as being not
violating aliasing rules (see also
http://gcc.gnu.org/ml/gcc/2006-07/msg00250.html).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28367

[Bug middle-end/28367] accessing via union on a vector does not cause vec_extract to be used

[pinskia at gcc dot gnu dot org]

------- Comment #2 from pinskia at gcc dot gnu dot org  2006-07-14 07:58 -------
(In reply to comment #1)
> Note a related testcase is:

I have a patch to fix that related testcase and also fix some other issues
related to BIT_FIELD_EXPR expansion with vec_extract (and why vec_set is still
unused after my patch too).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28367

[Bug middle-end/28367] accessing via union on a vector does not cause vec_extract to be used

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2006-07-19 14:46 -------
(In reply to comment #2)
> I have a patch to fix that related testcase and also fix some other issues
> related to BIT_FIELD_EXPR expansion with vec_extract (and why vec_set is still
> unused after my patch too).

And I filed it as PR 28436.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28367

[Bug middle-end/28367] accessing via union on a vector does not cause vec_extract to be used

[pinskia at gcc dot gnu dot org]

------- Comment #4 from pinskia at gcc dot gnu dot org  2006-08-26 04:48 -------
This is not target specific.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
 GCC target triplet|powerpc-linux-gnu           |
   Last reconfirmed|0000-00-00 00:00:00         |2006-08-26 04:48:03
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28367