[Bug tree-optimization/103222] New: [12 Regression] wrong code at -O2 and above on x86_64-linux-gnu

[Bug tree-optimization/103222] New: [12 Regression] wrong code at -O2 and above on x86_64-linux-gnu

[haoxintu at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103222

            Bug ID: 103222
           Summary: [12 Regression] wrong code at -O2 and above on
                    x86_64-linux-gnu
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: haoxintu at gmail dot com
  Target Milestone: ---

Hi all.

$cat small.c
#include <stdint.h>
#include <stdio.h>
int16_t a;
static uint32_t *b ;
static uint8_t func_2();
static int32_t func_1() {
  int16_t a = 1;
  func_2(0, a, a);
  return 0;
}
uint8_t func_2(uint32_t p1, uint32_t p2, uint32_t p3) {
  int p = 0;
  for (15;; a++) {
    for (0;;) {
      if (p2)
        break;
      b = &p2;
      return p2;
    }
     p3 = (p2 = p3, p);
  }
  return 0;
}

int main() {
  func_1();
  printf("%d\n", a);
  return 0;
}

$gcc -O1 small.c ; ./a.out
2

$gcc -O2 small.c ; ./a.out
1

$gcc -O3 small.c ; ./a.out
1

$gcc -Os small.c ; ./a.out
1

$gcc -v
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/home/haoxin/haoxin-data/compilers/gcc/build/libexec/gcc/x86_64-pc-linux-gnu/12.0.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with:../configure
--prefix=/home/haoxin/haoxin-data/compilers/gcc/build/ --enable-bootstrap
--enable-checking=release --enable-languages=c,c++ --enable-multilib :
(reconfigured) ../configure
--prefix=/home/haoxin/haoxin-data/compilers/gcc/build/ --enable-bootstrap
--enable-checking=release --enable-multilib --enable-languages=c,c++,lto
--no-create --no-recursion
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 12.0.0 20211113 (experimental) (GCC)


Note that this issue only happens in the current trunk version (12.0) of GCC.

Can also check in Godbolt: https://godbolt.org/z/rz7b9WxvT


Thanks,
Haoxin

[Bug tree-optimization/103222] [12 Regression] wrong code at -O2 and above on x86_64-linux-gnu

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103222

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
                 CC|                            |aldyh at gcc dot gnu.org,
                   |                            |hubicka at gcc dot gnu.org
   Last reconfirmed|                            |2021-11-13

--- Comment #1 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
So with -O1 we get:
int main ()
{
  int16_t a_lsm.11;
  int _2;
  unsigned short a.5_4;
  unsigned short _11;
  short int _12;

  <bb 2> [local count: 1073741835]:
  a_lsm.11_16 = a;
  a.5_4 = (unsigned short) a_lsm.11_16;
  _11 = a.5_4 + 2;
  _12 = (short int) _11;
  a = _12;
  _2 = (int) _12;
  printf ("%d\n", _2);
  return 0;

}

while with -O2 we get:
int main ()
{
  int _2;
  short int a.3_8;
  unsigned short a.5_9;
  unsigned short _10;
  short int _11;

  <bb 2> [local count: 1073741824]:
  a.3_8 = a;
  a.5_9 = (unsigned short) a.3_8;
  _10 = a.5_9 + 1;
  _11 = (short int) _10;
  a = _11;
  _2 = (int) _11;
  printf ("%d\n", _2);
  return 0;

}

Adding always_inline to func_2 makes -O1 and -O2 to agree up to ccp2
110t.mergephi2 and then 111t.threadfull makes a difference:

+  FAIL: path through PHI in bb3 (incoming bb:2) crosses loop
+path: 2->3->xx REJECTED
+Checking profitability of path (backwards):  bb:3 (3 insns) bb:5 (latch)
+  Control statement insns: 2
+  Overall: 1 insns
+Checking profitability of path (backwards):  bb:3 (3 insns) bb:5 (latch)
+  Control statement insns: 2
+  Overall: 1 insns
+  [1] Registering jump thread: (5, 3) incoming edge;  (3, 4) nocopy; 
+path: 5->3->4 SUCCESS
+Jump threading proved probability of edge 3->4 too small (it is 11.0%
(guessed) should be always (guessed))

 int main ()
 {
-  uint32_t p3;
   uint32_t p2;
+  uint32_t p3;
   short int a.0_1;
   int _2;
   short int a.3_8;
@@ -13,28 +72,15 @@
   short int _11;

   <bb 2> [local count: 1073741824]:
-
-  <bb 3> [local count: 9761289362]:
-  # p3_7 = PHI <1(2), 0(5)>
-  # p2_17 = PHI <1(2), p3_7(5)>
-  if (p2_17 != 0)
-    goto <bb 5>; [89.00%]
-  else
-    goto <bb 4>; [11.00%]
-
-  <bb 4> [local count: 1073741824]:
-  a.0_1 = a;
-  _2 = (int) a.0_1;
-  printf ("%d\n", _2);
-  return 0;
-
-  <bb 5> [local count: 8687547547]:
   a.3_8 = a;
   a.5_9 = (unsigned short) a.3_8;
   _10 = a.5_9 + 1;
   _11 = (short int) _10;
   a = _11;
-  goto <bb 3>; [100.00%]
+  a.0_1 = a;
+  _2 = (int) a.0_1;
+  printf ("%d\n", _2);
+  return 0;

 }

and --disable-tree-threadfull1 fixes the difference.

[Bug tree-optimization/103222] [12 Regression] wrong code at -O2 and above on x86_64-linux-gnu

[aldyh at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103222

Aldy Hernandez <aldyh at gcc dot gnu.org> changed:

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

--- Comment #2 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
Oh my.  So it does happen in real life.

This is exactly what we've been discussing here:

https://gcc.gnu.org/pipermail/gcc-patches/2021-November/584336.html

Thanks for providing a much needed reduced testcase ;-).

[Bug tree-optimization/103222] [12 Regression] wrong code at -O2 and above on x86_64-linux-gnu

[aldyh at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103222

--- Comment #3 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
Created attachment 51783
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51783&action=edit
patch in testing

[Bug tree-optimization/103222] [12 Regression] wrong code at -O2 and above on x86_64-linux-gnu

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103222

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

https://gcc.gnu.org/g:b7a23949b0dcc4205fcc2be6b84b91441faa384d

commit r12-5228-gb7a23949b0dcc4205fcc2be6b84b91441faa384d
Author: Aldy Hernandez <aldyh@redhat.com>
Date:   Sat Nov 13 12:37:25 2021 +0100

    path solver: Compute all PHI ranges simultaneously.

    PHIs must be resolved simulatenously, otherwise we may not pick up the
    ranges incoming to the block.

    For example.  If we put p3_7 in the cache before all PHIs have been
    computed, we will pick up the wrong p3_7 value for p2_17:

        # p3_7 = PHI <1(2), 0(5)>
        # p2_17 = PHI <1(2), p3_7(5)>

    This patch delays updating the cache until all PHIs have been
    analyzed.

    gcc/ChangeLog:

            PR tree-optimization/103222
            * gimple-range-path.cc (path_range_query::compute_ranges_in_phis):
            New.
            (path_range_query::compute_ranges_in_block): Call
            compute_ranges_in_phis.
            * gimple-range-path.h (path_range_query::compute_ranges_in_phis):
            New.

    gcc/testsuite/ChangeLog:

            * gcc.dg/pr103222.c: New test.

[Bug tree-optimization/103222] [12 Regression] wrong code at -O2 and above on x86_64-linux-gnu

[aldyh at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103222

Aldy Hernandez <aldyh at gcc dot gnu.org> changed:

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

--- Comment #5 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
fixed

[Bug tree-optimization/103222] [12 Regression] wrong code at -O2 and above on x86_64-linux-gnu

[haoxintu at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103222

--- Comment #6 from Haoxin Tu <haoxintu at gmail dot com> ---
(In reply to Aldy Hernandez from comment #5)
> fixed

Thank you for the quick fixing!

[Bug tree-optimization/103222] [12 Regression] wrong code at -O2 and above on x86_64-linux-gnu

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103222

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |12.0