[Bug ipa/100801] New: Aggressive loop optimizations cause incorrect warning

[Bug ipa/100801] New: Aggressive loop optimizations cause incorrect warning

[jl_gccbugs at conductive dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100801

            Bug ID: 100801
           Summary: Aggressive loop optimizations cause incorrect warning
           Product: gcc
           Version: 11.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: ipa
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jl_gccbugs at conductive dot de
                CC: marxin at gcc dot gnu.org
  Target Milestone: ---

The following warning is triggered

> $ gcc-11 -c constproploopopt.c -O2 -Wall -mavx -g
> constproploopopt.c: In function ‘test’:
> constproploopopt.c:22:18: warning: iteration 4611686018427387903 invokes undefined behavior [-Waggressive-loop-optimizations]
>    22 |     dest[i] = src[i];
>       |                  ^
> constproploopopt.c:21:12: note: within this loop
>    21 |   for (; i < count; ++i) {  // handle residual elements
>       |          ~~^~~~~~~

by this (minimal) code:

> #include <stdint.h>
> #include <stdio.h>
> #if defined(_MSC_VER)
> #include <intrin.h>
> #else
> #include <x86intrin.h>
> #endif
> 
> void copy_32_unaligned(uint32_t* dest, const uint32_t* src, size_t count) {
>   // invariant/nop
>   __m128i shufmask =  _mm_set_epi8(15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
> 
>   size_t i;
>   for (i = 0; i + 4 <= count; i += 4) {
>     __m128i input = _mm_loadu_si128((const __m128i*)(&src[i]));
>     __m128i output = input;
>     // no warning without the shuffle:
>     output = _mm_shuffle_epi8(input, shufmask);
>     _mm_storeu_si128((__m128i*)(&dest[i]), output);
>   }
>   for (; i < count; ++i) {  // handle residual elements
>     dest[i] = src[i];
>   }
> }
> 
> void test(uint32_t* buf1, uint32_t* buf2) {
>     copy_32_unaligned(buf2, buf1,
>                       // multiples of 4 and greater or equal then 12 trigger it:
>                       12);
> }

>From objdump output I believe the generated code is correct though. The warning
seems to be incorrect in this context, especially since the "residual" loop
should be skipped for count=n*4 anyways.

[Bug tree-optimization/100801] Aggressive loop optimizations cause incorrect warning

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100801

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|ipa                         |tree-optimization
           Keywords|                            |diagnostic,
                   |                            |missed-optimization
             Status|UNCONFIRMED                 |ASSIGNED
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2021-05-28
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  We're warning on the obviously unreachable loop since:

<bb 4> [local count: 1073741824]:
_35 = 12;
if (_35 != 12)
  goto <bb 7>; [75.00%]
else
  goto <bb 6>; [25.00%]

$3 = <basic_block 0x7ffff434b680 (4)>

since we apply final value replacement to 'i' in sccp but do not propagate it
before the next number of iteration analysis in ivcanon:

          NEXT_PASS (pass_tree_loop_init);
          NEXT_PASS (pass_tree_unswitch);
          NEXT_PASS (pass_scev_cprop);          <<<< final value repl.
          NEXT_PASS (pass_loop_split);
          NEXT_PASS (pass_loop_versioning);
          NEXT_PASS (pass_loop_jam);
          /* All unswitching, final value replacement and splitting can expose
             empty loops.  Remove them now.  */
          NEXT_PASS (pass_cd_dce, false /* update_address_taken_p */);
          NEXT_PASS (pass_iv_canon);           <<<<< warning
          NEXT_PASS (pass_loop_distribution);
          NEXT_PASS (pass_linterchange);
          NEXT_PASS (pass_copy_prop);          <<<<<< propagation

IIRC final value replacement used to propagate and fold but I(?) removed this
at some point.

[Bug tree-optimization/100801] Aggressive loop optimizations cause incorrect warning

[jl_gccbugs at conductive dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100801

--- Comment #2 from Joel Linn <jl_gccbugs at conductive dot de> ---
Great. In the meantime I will use 
> if (count % 4 == 0) __builtin_unreachable();
at the start of the for loop to suppress the warning as suggested by Martin
Sebor https://gcc.gnu.org/pipermail/gcc-help/2021-May/140339.html

[Bug tree-optimization/100801] Aggressive loop optimizations cause incorrect warning

[matthijsvanduin at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100801

Matthijs van Duin <matthijsvanduin at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |matthijsvanduin at gmail dot com

--- Comment #3 from Matthijs van Duin <matthijsvanduin at gmail dot com> ---
Simpler testcase:

/* compiler flags required to trigger warning: -O1 -ftree-vrp  */

static void foo(int *x, long n)
{
        long i = 0;
        for (; i + 4 <= n; i += 4) {
        }
        for (; i < n; i++) {
                x[i]++;
        }
}

void bar(int *x)
{
        foo(x, 128);
}


A workaround with no runtime overhead is adding the following check between the
loops:

if (__builtin_constant_p(n % 4 == 0) && n % 4 == 0)
        return;

[Bug tree-optimization/100801] [9/10/11/12 Regression] Aggressive loop optimizations cause incorrect warning

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100801

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Aggressive loop             |[9/10/11/12 Regression]
                   |optimizations cause         |Aggressive loop
                   |incorrect warning           |optimizations cause
                   |                            |incorrect warning
      Known to fail|                            |10.1.0, 9.1.0
   Target Milestone|---                         |9.5
      Known to work|                            |8.5.0

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note the warning started in GCC 9 even. So it is a regression.

[Bug tree-optimization/100801] [9/10/11/12 Regression] Aggressive loop optimizations cause incorrect warning

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100801

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

[Bug tree-optimization/100801] [10/11/12/13 Regression] Aggressive loop optimizations cause incorrect warning

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100801

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|9.5                         |10.4

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 9 branch is being closed

[Bug tree-optimization/100801] [10/11/12/13 Regression] Aggressive loop optimizations cause incorrect warning

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100801

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|10.4                        |10.5

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 10.4 is being released, retargeting bugs to GCC 10.5.

[Bug tree-optimization/100801] [11/12/13/14 Regression] Aggressive loop optimizations cause incorrect warning

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100801

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|10.5                        |11.5

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 10 branch is being closed.

[Bug tree-optimization/100801] [11/12 Regression] Aggressive loop optimizations cause incorrect warning

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100801

Jeffrey A. Law <law at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at gcc dot gnu.org
            Summary|[11/12/13/14 Regression]    |[11/12 Regression]
                   |Aggressive loop             |Aggressive loop
                   |optimizations cause         |optimizations cause
                   |incorrect warning           |incorrect warning

--- Comment #8 from Jeffrey A. Law <law at gcc dot gnu.org> ---
Works with gcc-13 and the trunk.  Adjusting markers.