[Bug tree-optimization/94834] New: Failure to optimize loop bswap pattern

[Bug tree-optimization/94834] New: Failure to optimize loop bswap pattern

[gabravier at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94834

            Bug ID: 94834
           Summary: Failure to optimize loop bswap pattern
           Product: gcc
           Version: 10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gabravier at gmail dot com
  Target Milestone: ---

uint32_t load(const uint8_t* data)
{
    uint32_t val = 0;
    for (int i = 0; i < sizeof(val) * CHAR_BIT; i += CHAR_BIT)
    {
        val |= *data++ << i;
    }
    return val;
}

This can be optimized to a single 32-bit load. LLVM does this transformation,
gcc just unrolls the loop and misses the transformation.

LLVM gives :

load(unsigned char const*): # @load(unsigned char const*)
  mov eax, dword ptr [rdi]
  ret

GCC gives :

load(unsigned char const*):
  movzx edx, BYTE PTR [rdi+1]
  movzx eax, BYTE PTR [rdi]
  sal edx, 8
  or edx, eax
  movzx eax, BYTE PTR [rdi+2]
  sal eax, 16
  or edx, eax
  movzx eax, BYTE PTR [rdi+3]
  sal eax, 24
  or eax, edx
  ret

See also https://godbolt.org/z/kmYTLZ

[Bug tree-optimization/94834] Failure to optimize loop bswap pattern

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94834

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
           Keywords|                            |missed-optimization
   Last reconfirmed|                            |2020-04-29
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  We have a pass that would grok the unrolled form but unrolling
happens too late (if at all) for it to trigger.  Not sure if we should
try to pattern-match this (or other reductions) in loop distribution.

[Bug tree-optimization/94834] Failure to optimize loop bswap pattern

[gabravier at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94834

--- Comment #2 from Gabriel Ravier <gabravier at gmail dot com> ---
Now I wonder why the unrolling happens too late since there was 1 ecp check
that should happen after the unrolling, from my understanding. Are the multiple
ecp passes detecting different things?

[Bug tree-optimization/94834] Failure to optimize loop bswap pattern

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94834

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

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

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
This is a dup of bug 89811.

*** This bug has been marked as a duplicate of bug 89811 ***