[Bug c/107493] New: Wrong code at -Os on x86_64-linux-gnu

[Bug c/107493] New: Wrong code at -Os on x86_64-linux-gnu

[shaohua.li at inf dot ethz.ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107493

            Bug ID: 107493
           Summary: Wrong code at -Os on x86_64-linux-gnu
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: shaohua.li at inf dot ethz.ch
  Target Milestone: ---

% gcc-tk -v
Using built-in specs.
COLLECT_GCC=gcc-tk
COLLECT_LTO_WRAPPER=/zdata/shaoli/compilers/ccbuilder-compilers/gcc-db55f1dda2692c3f778ae783bc7121891c79aec4/libexec/gcc/x86_64-pc-linux-gnu/13.0.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: ../configure --disable-multilib --disable-bootstrap
--enable-languages=c,c++
--prefix=/zdata/shaoli/compilers/ccbuilder-compilers/gcc-db55f1dda2692c3f778ae783bc7121891c79aec4
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 13.0.0 20221031 (experimental) (GCC) 
% 
% gcc-tk -O0 a.c && ./a.out
65525
% gcc-tk -Os a.c && ./a.out
65515
%
% cat a.c
int printf(const char *format, ...);
int a;
int b(int c) { 
    return c; 
}
int main() {
  a = -21;
  for (; a <= 0; a = (unsigned short)(b(a + 2) + 8))
    ;
  printf("%d\n", a);
}
%

Compiler explorer: https://godbolt.org/z/hhqrv7hWh

[Bug tree-optimization/107493] [13 Regression] Wrong code at -Os on x86_64-linux-gnu

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107493

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |12.1.0
   Last reconfirmed|                            |2022-11-01
      Known to fail|                            |13.0
     Ever confirmed|0                           |1
            Summary|Wrong code at -Os on        |[13 Regression] Wrong code
                   |x86_64-linux-gnu            |at -Os on x86_64-linux-gnu
   Target Milestone|---                         |13.0
             Status|UNCONFIRMED                 |NEW

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
sccp is broken:

Estimating # of iterations of loop 1
Statement (exit)if (a_lsm.10_13 <= 0)
 is executed at most 0 (bounded by 0) + 1 times in loop 1.
Induction variable (int) 65517 + 10 * iteration does not wrap in statement _1 =
a_lsm.10_13 + 2;
 in loop 1.
Statement _1 = a_lsm.10_13 + 2;
 is executed at most 429490178 (bounded by 429490178) + 1 times in loop 1.
Induction variable (int) 65525 + 10 * iteration does not wrap in statement _4 =
(int) _3;
 in loop 1.
Statement _4 = (int) _3;
 is executed at most 1 (bounded by 1) + 1 times in loop 1.

final value replacement:
  a_lsm.10_14 = PHI <a_lsm.10_13(4)>
 with expr: 65515
 final stmt:
  a_lsm.10_14 = 65515;

Confirmed.

[Bug tree-optimization/107493] [13 Regression] Wrong code at -Os on x86_64-linux-gnu

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107493

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |needs-bisection

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
self contained testcase:
int a;
int b(int c) { 
    return c; 
}
int main() {
  a = -21;
  for (; a <= 0; a = (unsigned short)(b(a + 2) + 8))
    ;
  if (a != 65525)
    __builtin_abort();
 // __builtin_printf("%d\n", a);
}

[Bug tree-optimization/107493] [13 Regression] Wrong code at -Os on x86_64-linux-gnu

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107493

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

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

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
I will have a look, possibly fallout of the PR107176 fix.

[Bug tree-optimization/107493] [13 Regression] Wrong code at -Os on x86_64-linux-gnu

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107493

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
a_lsm.9_12 = PHI <-21(2), _4(3)>

(set_scalar_evolution
  instantiated_below = 2
  (scalar = a_lsm.9_12)
  (scalar_evolution = (int) {65515, +, 10}_1))

is definitely wrong since the initial value is off.  The evolution after
that looks OK though, so not sure why SCCP gets the wrong final value here,
we compute the number of iterations to zero (also wrong).

It goes wrong when we get to the truncation, we then do

        *evolution_of_loop = chrec_convert (type, *evolution_of_loop, at_stmt);

with {-21, + 2}_1 converting that to unsigned short.  This ends up in
convert_affine_scev but that assumes it converts the cycle and not that
the conversion is inside the cycle.

The old sanity checking of the initial value would have catched this but the
actual error is elsewhere.  The question is what's valid here.  I'm going to
try restricting this all further.

[Bug tree-optimization/107493] [13 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3486-g4c5b116077638277

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107493

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[13 Regression] Wrong code  |[13 Regression] Wrong code
                   |at -Os on x86_64-linux-gnu  |at -Os on x86_64-linux-gnu
                   |                            |since
                   |                            |r13-3486-g4c5b116077638277
                 CC|                            |marxin at gcc dot gnu.org
           Keywords|needs-bisection             |

--- Comment #5 from Martin Liška <marxin at gcc dot gnu.org> ---
Started with r13-3486-g4c5b116077638277.

[Bug tree-optimization/107493] [13 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3486-g4c5b116077638277

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107493

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Blocks|                            |107176
           Priority|P3                          |P1


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107176
[Bug 107176] [10/11/12 Regression] Wrong code at -Os on x86_64-pc-linux-gnu
since r7-2012-g43aabfcfd4139e4c

[Bug tree-optimization/107493] [13 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3486-g4c5b116077638277

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107493

--- Comment #6 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:2b293a949c0fbe21e47c4bc99f807dc941c02bb6

commit r13-4357-g2b293a949c0fbe21e47c4bc99f807dc941c02bb6
Author: Richard Biener <rguenther@suse.de>
Date:   Mon Nov 28 10:25:44 2022 +0100

    tree-optimization/107493 - SCEV analysis with conversions

    This shows another case where trying to validate conversions during
    the CHREC SCC analysis fails because said analysis assumes we are
    converting a complete SCC.  Like the constraint on the initial
    conversion seen restrict all conversions handled to sign-changes.

            PR tree-optimization/107493
            * tree-scalar-evolution.cc (scev_dfs::follow_ssa_edge_expr):
            Only handle no-op and sign-changing conversions.

            * gcc.dg/torture/pr107493.c: New testcase.

[Bug tree-optimization/107493] [13 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3486-g4c5b116077638277

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107493

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
             Status|ASSIGNED                    |RESOLVED

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.