[Bug tree-optimization/114234] New: [14 Regression] verify_ssa failure with early-break vectorisation

[Bug tree-optimization/114234] New: [14 Regression] verify_ssa failure with early-break vectorisation

[rsandifo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114234

            Bug ID: 114234
           Summary: [14 Regression] verify_ssa failure with early-break
                    vectorisation
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rsandifo at gcc dot gnu.org
  Target Milestone: ---

The following test ICEs with -Ofast on aarch64:

void bar();
float
foo (float x)
{
  float a = 1;
  float b = x;
  long z = 200;
  for (;;)
    {
      float c = b - 1.0f;
      a *= c;
      z -= 1;
      if (z == 0)
        {
          bar ();
          break;
        }
      if (b <= 3.0f)
        break;
      b = c;
    }
  return a * b;
}

(reduced from wrf).  The ICE is:

foo.c:3:1: error: definition in block 15 does not dominate use in block 10
    3 | foo (float x)
      | ^~~
for SSA_NAME: stmp_a_9.10_103 in statement:
a_47 = PHI <stmp_a_9.10_103(10), stmp_a_9.10_103(15)>
PHI argument
stmp_a_9.10_103
for PHI node
a_47 = PHI <stmp_a_9.10_103(10), stmp_a_9.10_103(15)>
during GIMPLE pass: vect

[Bug tree-optimization/114234] [14 Regression] verify_ssa failure with early-break vectorisation

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114234

David Binderman <dcb314 at hotmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dcb314 at hotmail dot com

--- Comment #1 from David Binderman <dcb314 at hotmail dot com> ---

For this C code:

int ip4_getbit_a, ip4_getbit_pos, ip4_clrbit_pos;
void ip4_clrbit(int *a) { *a &= ip4_clrbit_pos; }
typedef struct {
  char pxlen;
  int prefix
} net_addr_ip4;
void fib_get_chain();
int trie_match_longest_ip4();
int trie_match_next_longest_ip4(net_addr_ip4 *n) {
  int __trans_tmp_1;
  while (n->pxlen) {
    n->pxlen--;
    ip4_clrbit(&n->prefix);
    __trans_tmp_1 = ip4_getbit_a >> ip4_getbit_pos;
    if (__trans_tmp_1)
      return 1;
  }
  return 0;
}
void net_roa_check_ip4_trie_tab() {
  net_addr_ip4 px0;
  for (int _n = trie_match_longest_ip4(&px0); _n;
       _n = trie_match_next_longest_ip4(&px0))
    fib_get_chain();
}

on x86_64, does this:

$ ../results/bin/gcc -c -O3 -march=znver3 -w ~/cvise/bug1021.c
/home/dcb38/cvise/bug1021.c: In function ‘net_roa_check_ip4_trie_tab’:
/home/dcb38/cvise/bug1021.c:20:6: error: definition in block 19 does not
dominate use in block 14
   20 | void net_roa_check_ip4_trie_tab() {
      |      ^~~~~~~~~~~~~~~~~~~~~~~~~~
for SSA_NAME: _117 in statement:
px0__lsm.22_47 = PHI <_117(14), _117(19)>
PHI argument
_117
for PHI node
px0__lsm.22_47 = PHI <_117(14), _117(19)>
during GIMPLE pass: vect
/home/dcb38/cvise/bug1021.c:20:6: internal compiler error: verify_ssa failed

The bug first seems to occur sometime between g:1e74ce8983fd4926
and g:71244316cf714725, which is 42 commits.

[Bug tree-optimization/114234] [14 Regression] verify_ssa failure with early-break vectorisation

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114234

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

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

--- Comment #2 from Sam James <sjames at gcc dot gnu.org> ---
That second one is probably worth its own PR.

[Bug tree-optimization/114234] [14 Regression] verify_ssa failure with early-break vectorisation

[tnfchris at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114234

Tamar Christina <tnfchris at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2024-03-05
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

--- Comment #3 from Tamar Christina <tnfchris at gcc dot gnu.org> ---
started with

commit g:324d2907c86f05e40dc52d226940308f53a956c2
Author: Richard Biener <rguenther@suse.de>
Date:   Mon Mar 4 09:46:13 2024 +0100

    tree-optimization/114192 - scalar reduction kept live with early break vect

    The following fixes a missing replacement of the reduction value
    used in the epilog, causing the scalar reduction to be kept live
    across the early break exit path.

            PR tree-optimization/114192
            * tree-vect-loop.cc (vect_create_epilog_for_reduction): Use the
            appropriate def for the live out stmt in case of an alternate
            exit.

the reduction seems to propagate out a single value for both exits:

  # a_39 = PHI <stmp_a_9.10_95(10), stmp_a_9.10_95(15)>

but since the PHI nodes carry their reduction values in the exits we have
different reduction statements, so using the same reduction value for all exit
is wrong.

should be

  # a_39 = PHI <stmp_a_9.10_89(10), stmp_a_9.10_95(15)>

[Bug tree-optimization/114234] [14 Regression] verify_ssa failure with early-break vectorisation

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114234

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |DUPLICATE
   Target Milestone|---                         |14.0

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
Verified the PR114239 fix fixes this as well, dup.

*** This bug has been marked as a duplicate of bug 114239 ***