[Bug tree-optimization/110991] New: [14 Regression] Dead Code Elimination Regression at -O2 since r14-1135-gc53f51005de

[Bug tree-optimization/110991] New: [14 Regression] Dead Code Elimination Regression at -O2 since r14-1135-gc53f51005de

[scherrer.sv at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110991

            Bug ID: 110991
           Summary: [14 Regression] Dead Code Elimination Regression at
                    -O2 since r14-1135-gc53f51005de
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: scherrer.sv at gmail dot com
  Target Milestone: ---

static unsigned char a;
static char b;
void foo(void);
int main() {
  a = 25;
  for (; a > 13; --a)
    b = a > 127 ?: a << 3;
  if (!b)
    foo();
}

gcc-3a13884b23a (trunk) -O2 cannot eliminate the call to foo but
gcc-releases/gcc-13.1.0 -O2 can.
-----------------------------------------------------------------------
gcc-3a13884b23ae32b43d56d68a9c6bd4ce53d60017 -O2 case.c -S -o case.s
--------- OUTPUT ---------
main:
.LFB0:
        .cfi_startproc
        movd    .LC0(%rip), %xmm0
        movd    .LC1(%rip), %xmm1
        xorl    %eax, %eax
.L2:
        addl    $1, %eax
        movdqa  %xmm0, %xmm2
        paddb   %xmm1, %xmm0
        cmpl    $3, %eax
        jne     .L2
        movdqa  %xmm2, %xmm0
        movb    $13, a(%rip)
        paddb   %xmm2, %xmm0
        paddb   %xmm0, %xmm0
        movdqa  %xmm0, %xmm1
        paddb   %xmm0, %xmm1
        pxor    %xmm0, %xmm0
        pcmpgtb %xmm2, %xmm0
        movd    .LC2(%rip), %xmm2
        pand    %xmm0, %xmm2
        pandn   %xmm1, %xmm0
        por     %xmm2, %xmm0
        movd    %xmm0, %eax
        sarl    $24, %eax
        movb    %al, b(%rip)
        testb   %al, %al
        je      .L10
        xorl    %eax, %eax
        ret
.L10:
        pushq   %rax
        .cfi_def_cfa_offset 16
        call    foo
        xorl    %eax, %eax
        popq    %rdx
        .cfi_def_cfa_offset 8
        ret
---------- END OUTPUT ---------

-----------------------------------------------------------------------
gcc-2b98cc24d6af0432a74f6dad1c722ce21c1f7458 -O2 case.c -S -o case.s
--------- OUTPUT ---------
main:
.LFB0:
        .cfi_startproc
        movb    $112, b(%rip)
        xorl    %eax, %eax
        movb    $13, a(%rip)
        ret
---------- END OUTPUT ---------

-----------------------------------------------------------------------
Bisects to r14-1135-gc53f51005de

[Bug tree-optimization/110991] [14 Regression] Dead Code Elimination Regression at -O2 since r14-1135-gc53f51005de

[ubizjak at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110991

Uroš Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2023-08-11

--- Comment #1 from Uroš Bizjak <ubizjak at gmail dot com> ---
For gcc-13, fre4 pass is able to simplify the scalar code, but nothing
simplifies vectorized code in gcc-14.

[Bug tree-optimization/110991] [14 Regression] Dead Code Elimination Regression at -O2 since r14-1135-gc53f51005de

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110991

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
What is interesting is -O3 unrolls the loop in cunroll and the loop becomes a
nothing as everything can be almost constant folded away ... Maybe that is
something which can be tuned for -O2 and unrolling ...

[Bug tree-optimization/110991] [14 Regression] Dead Code Elimination Regression at -O2 since r14-1135-gc53f51005de

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110991

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
             Target|                            |x86_64-*-*
   Target Milestone|---                         |14.0
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
So the difference is that GCC 14 vectorizes the loop and that vectorized loop
is not completely unrolled because

Loop 1 likely iterates at most 2 times.
Estimating sizes for loop 1
 BB: 3, after_exit: 0
  size:   1 _34 = vect_vec_iv_.15_33 + { 252, 252, 252, 252 };
  size:   0 vect_a.16_35 = VIEW_CONVERT_EXPR<vector(4) signed
char>(vect_vec_iv_.15_33);
  size:   1 vect_iftmp.17_36 = vect_a.16_35 << 3;
  size:   1 mask__23.18_38 = vect_a.16_35 < { 0, 0, 0, 0 };
  size:   1 vect_iftmp.19_40 = VEC_COND_EXPR <mask__23.18_38, { 1, 1, 1, 1 },
vect_iftmp.17_36>;
  size:   1 ivtmp_44 = ivtmp_43 + 1;
   Induction variable computation will be folded away.
  size:   2 if (ivtmp_44 < 3)
   Exit condition will be eliminated in peeled copies.
   Exit condition will be eliminated in last copy.
   Constant conditional.
 BB: 9, after_exit: 1
size: 7-3, last_iteration: 7-3
  Loop size: 7
  Estimated size after unrolling: 8
Not unrolling loop 1: size would grow.

when we still have a loop there's nothing that can fully elide things.
Without vectorization we have

Loop 2 likely iterates at most 11 times.
Estimating sizes for loop 2
 BB: 10, after_exit: 0
  size:   0 a.2_13 = (signed char) a.6_22;
   Induction variable computation will be folded away.
  size:   2 if (a.2_13 < 0)
   Constant conditional.
 BB: 13, after_exit: 1
 BB: 12, after_exit: 0
  size:   1 _26 = a.6_22 + 255;
   Induction variable computation will be folded away.
  size:   1 ivtmp_27 = ivtmp_4 - 1;
   Induction variable computation will be folded away.
  size:   2 if (ivtmp_27 != 0)
   Exit condition will be eliminated in peeled copies.
   Exit condition will be eliminated in last copy.
   Constant conditional. 
 BB: 11, after_exit: 0
  size:   1 iftmp.0_12 = a.2_13 << 3;
   Induction variable computation will be folded away.
size: 7-7, last_iteration: 7-7
  Loop size: 7
  Estimated size after unrolling: 1

unrolling relies on constant_after_peeling which relies on SCEV which
doesn't handle vector IVs.

I have a patch improving it to

size: 7-4, last_iteration: 7-4
  Loop size: 7
  Estimated size after unrolling: 6

IIRC I also had a patch more appropriately "propagating" constness at some
point.

[Bug tree-optimization/110991] [14 Regression] Dead Code Elimination Regression at -O2 since r14-1135-gc53f51005de

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110991

--- Comment #4 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:bcdbedb3e6083ad01d844ed97cf19645c1ef6568

commit r14-3216-gbcdbedb3e6083ad01d844ed97cf19645c1ef6568
Author: Richard Biener <rguenther@suse.de>
Date:   Mon Aug 14 09:31:18 2023 +0200

    tree-optimization/110991 - unroll size estimate after vectorization

    The following testcase shows that we are bad at identifying inductions
    that will be optimized away after vectorizing them because SCEV doesn't
    handle vectorized defs.  The following rolls a simpler identification
    of SSA cycles covering a PHI and an assignment with a binary operator
    with a constant second operand.

            PR tree-optimization/110991
            * tree-ssa-loop-ivcanon.cc (constant_after_peeling): Handle
            VIEW_CONVERT_EXPR <op>, handle more simple IV-like SSA cycles
            that will end up constant.

            * gcc.dg/tree-ssa/cunroll-16.c: New testcase.

[Bug tree-optimization/110991] [14 Regression] Dead Code Elimination Regression at -O2 since r14-1135-gc53f51005de

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110991

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=91975
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.