[Bug tree-optimization/127190] New: [17 Regression] ICE in vect_create_constant_vectors at tree-vect-slp.cc:10912 since r17-3793

[dhruvc at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127190

            Bug ID: 127190
           Summary: [17 Regression] ICE in vect_create_constant_vectors at
                    tree-vect-slp.cc:10912 since r17-3793
           Product: gcc
           Version: 17.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: dhruvc at gcc dot gnu.org
                CC: rguenth at gcc dot gnu.org
  Target Milestone: ---
            Target: aarch64

Seen in SPEC2026 palm and llvm benchmarks. Looks like the commit's testcase is
itself failing on AArch64:

===
unsigned a, c;
int b, d;
signed char e;

void __attribute__((noipa))
f (void)
{
  for (; b; ++b)
    {
      c = 0;
      for (; c < 10; c++)
        {
          d = 0;
          for (; d < c; d++)
            a += e;
        }
    }
}
===

Flags: -O3 -march=armv8-a+sve

Compiler explorer: https://godbolt.org/z/nesnGaea6

===

LLM-generated explanation (if useful): 

  Commit adds an extra requirement to vect_get_num_copies_for_invariant
(gcc/tree-vect-slp.cc:8946):

  bool res = (can_div_away_from_zero_p (vf, TYPE_VECTOR_SUBPARTS (vectype),
nvectors)
              && ((TYPE_VECTOR_SUBPARTS (vectype) * *nvectors - vf).is_constant
(&rem)));

  For SVE the vectype is variable-length, so subparts * nvectors - vf is not
constant → returns false. Both callers assume it cannot fail:
  - vect_create_constant_vectors    gcc_assert (res)                           
  — tree-vect-slp.cc:10912
  - vectorizable_lane_reducing      gcc_assert (res && ncopies_for_cost >= 1)  
  — tree-vect-loop.cc:6861