[Bug target/115418] New: [14 regression] Extra movapd emitted for MAX implementation

[Bug target/115418] New: [14 regression] Extra movapd emitted for MAX implementation

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115418

            Bug ID: 115418
           Summary: [14 regression] Extra movapd emitted for MAX
                    implementation
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: sjames at gcc dot gnu.org
  Target Milestone: ---

```
#define MAX(a, b) ((a) < (b) ? (b) : (a))
#define MAX3(a, b, c) MAX((a), MAX((b), (c)))

double f(double a, double b, double c)
{
return MAX3(a,b,c);
}
```

```
$ gcc-14 /tmp/foo200.c -S -o - -O2
.file   "foo200.c"
.text
.p2align 4
.globl  f
.type   f, @function
f:
.LFB0:
.cfi_startproc
movapd  %xmm2, %xmm3
maxsd   %xmm1, %xmm3
maxsd   %xmm0, %xmm3
movapd  %xmm3, %xmm0
ret
.cfi_endproc
.LFE0:
.size   f, .-f
.ident  "GCC: (Gentoo Hardened 14.1.1_p20240608 p2) 14.1.1 20240608"
.section        .note.GNU-stack,"",@progbits
```

I only see the extra movapd with 14.

[Bug target/115418] [14 regression] Extra movapd emitted for MAX implementation

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115418

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
             Target|                            |x86_64-linux-gnu
   Last reconfirmed|                            |2024-06-10

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The difference between GCC 14 and 15 is:

GCC 14:
  _1 = a_4(D) < _5;
  _7 = _1 ? _5 : a_4(D);

GCC 15:
  _8 = _4 > a_5(D);
  _7 = _8 ? _4 : a_5(D);


GCC 15's IR gets recognized for UNSPEC_IEEE_MAX while GCC 14's is not.

If anything the target should recognize both cases ... Since both are
canonical.

[Bug target/115418] Extra movapd emitted for MAX implementation

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115418

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[14 regression] Extra       |Extra movapd emitted for
                   |movapd emitted for MAX      |MAX implementation
                   |implementation              |

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note the issue is ix86_expand_sse_fp_minmax only handles LT/UNGE but it should
handle GT/UNLT with both parts swapped (comparison and true/false).

Note this was just by accident that GCC 14 does not work.

[Bug target/115418] Extra movapd emitted for MAX implementation

[liuhongt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115418

Hongtao Liu <liuhongt at gcc dot gnu.org> changed:

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

--- Comment #3 from Hongtao Liu <liuhongt at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #2)
> Note the issue is ix86_expand_sse_fp_minmax only handles LT/UNGE but it
> should handle GT/UNLT with both parts swapped (comparison and true/false).
> 
GT/UNLT is "canonicalized" to GT/UNGT in ix86_prepare_sse_fp_compare_args
 4410    case GE:
 4411    case GT:
 4412    case UNLE:
 4413    case UNLT:
 4414      /* These are not supported directly before AVX, and furthermore
 4415         ix86_expand_sse_fp_minmax only optimizes LT/UNGE.  Swap the
 4416         comparison operands to transform into something that is
 4417         supported.  */
 4418      std::swap (*pop0, *pop1);
 4419      code = swap_condition (code);