[Bug middle-end/112324] New: phiopt fail to recog if (b < 0) max = MAX(-b, max); else max = MAX (b, max) into max = MAX (ABS(b), max)

[Bug middle-end/112324] New: phiopt fail to recog if (b < 0) max = MAX(-b, max); else max = MAX (b, max) into max = MAX (ABS(b), max)

[crazylht at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112324

            Bug ID: 112324
           Summary: phiopt fail to recog if (b < 0) max = MAX(-b, max);
                    else max = MAX (b, max) into max = MAX (ABS(b), max)
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: crazylht at gmail dot com
  Target Milestone: ---

#define MAX(a, b) ((a) > (b) ? (a) : (b)) 
int
foo (int n, int* a)
{
    int max = 0;
    for (int i = 0; i != n; i++)
    {
        int tmp = a[i];
        if (tmp < 0)
          max = MAX (-tmp, max);
        else
          max = MAX (tmp, max);
    }
    return max;
}

int
foo1 (int n, int* a)
{
    int max = 0;
    for (int i = 0; i != n; i++)
    {
        int tmp = a[i];
        max = MAX ((tmp < 0 ? -tmp : tmp), max);
    }
    return max;
}

foo should be same as foo1, but gcc failed to recognize ABS_EXPR in foo.
It's from pr110015(originally from source code in openjpeg).

[Bug tree-optimization/112324] phiopt fail to recog if (b < 0) max = MAX(-b, max); else max = MAX (b, max) into max = MAX (ABS(b), max)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112324

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu.org
     Ever confirmed|0                           |1
          Component|middle-end                  |tree-optimization
             Status|UNCONFIRMED                 |ASSIGNED
           Severity|normal                      |enhancement
           Assignee|unassigned at gcc dot gnu.org      |pinskia at gcc dot gnu.org
   Last reconfirmed|                            |2023-11-01

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Mine.

There is another bug which is very much similar to this but:
  max_19 = MAX_EXPR <_4, max_6>;
  goto <bb 6>; [INV]

  <bb 5> :
  _8 = MAX_EXPR <max_6, tmp_15>;

  <bb 6> :
  # max_5 = PHI <max_19(4), _8(5)>

Basically we need to pull MAX_EXPR out of the if and then the ABS will be
recognized and all.

[Bug tree-optimization/112324] phiopt fail to recog if (b < 0) max = MAX(-b, max); else max = MAX (b, max) into max = MAX (ABS(b), max)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112324

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Depends on|                            |64700

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
PR 64700 (and others).


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64700
[Bug 64700] Sink common code through PHI

[Bug tree-optimization/112324] phiopt fail to recog if (b < 0) max = MAX(-b, max); else max = MAX (b, max) into max = MAX (ABS(b), max)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112324

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note a simple patch to improve phiopt here does not improve the original code
in openjpeg.

The reason why is due to the code in phiopt here is only operates on the case
where the phi would be the only one. In that case we also have a vop phi which
changes which causes the code not to be done.

[Bug tree-optimization/112324] phiopt fail to recog if (b < 0) max = MAX(-b, max); else max = MAX (b, max) into max = MAX (ABS(b), max)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112324

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #3)
> Note a simple patch to improve phiopt here does not improve the original
> code in openjpeg.
> 
> The reason why is due to the code in phiopt here is only operates on the
> case where the phi would be the only one. In that case we also have a vop
> phi which changes which causes the code not to be done.

But maybe ifcvt could use the same function to do the transformation before it
does the conversion ...

[Bug tree-optimization/112324] phiopt fail to recog if (b < 0) max = MAX(-b, max); else max = MAX (b, max) into max = MAX (ABS(b), max)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112324

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #4)
> (In reply to Andrew Pinski from comment #3)
> > Note a simple patch to improve phiopt here does not improve the original
> > code in openjpeg.
> > 
> > The reason why is due to the code in phiopt here is only operates on the
> > case where the phi would be the only one. In that case we also have a vop
> > phi which changes which causes the code not to be done.
> 
> But maybe ifcvt could use the same function to do the transformation before
> it does the conversion ...

I will try to figure that out next but I want to get this patch though.

[Bug tree-optimization/112324] phiopt fail to recog if (b < 0) max = MAX(-b, max); else max = MAX (b, max) into max = MAX (ABS(b), max)

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112324

Sam James <sjames at gcc dot gnu.org> changed:

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

--- Comment #6 from Sam James <sjames at gcc dot gnu.org> ---
Someone filed a bug for LLVM for x <= abs(x) being always true
(https://github.com/llvm/llvm-project/issues/72653) and it looks like we're
missing that too. Not sure if that's a dupe though.

[Bug tree-optimization/112324] phiopt fail to recog if (b < 0) max = MAX(-b, max); else max = MAX (b, max) into max = MAX (ABS(b), max)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112324

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Sam James from comment #6)
> Someone filed a bug for LLVM for x <= abs(x) being always true
> (https://github.com/llvm/llvm-project/issues/72653) and it looks like we're
> missing that too. Not sure if that's a dupe though.

Yes and there is no dup filed so far. Let me file about them.

[Bug tree-optimization/112324] phiopt fail to recog if (b < 0) max = MAX(-b, max); else max = MAX (b, max) into max = MAX (ABS(b), max)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112324

--- Comment #8 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Sam James from comment #6)
> Someone filed a bug for LLVM for x <= abs(x) being always true
> (https://github.com/llvm/llvm-project/issues/72653) and it looks like we're
> missing that too. Not sure if that's a dupe though.

Filed as PR 112626.