https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127475
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|UNCONFIRMED |ASSIGNED
Ever confirmed|0 |1
Last reconfirmed| |2026-09-18
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
gdb) p debug_tree (elt)
<integer_cst 0x7ffff7408f60 type <boolean_type 0x7ffff77c9348 bool8> constant
public overflow 16>
{ 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF),
16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF) }
<bb 5> [local count: 744103079]:
# vect_vec_iv_.8_37 = PHI <{ 0(OVF), 1(OVF), 2(OVF), 3, 4, 5, 6, 7, 8, 9, 10,
11, 12, 13, 14, 15 }(4), _40(5)>
# ivtmp.25_6 = PHI <ivtmp.25_27(4), ivtmp.25_4(5)>
vect__7.9_41 = VIEW_CONVERT_EXPR<vector(16) unsigned
char>(vect_vec_iv_.8_37);
_26 = (void *) ivtmp.25_6;
MEM <vector(16) unsigned char> [(unsigned char *)_26] = vect__7.9_41;
_38 = (vector(16) <signed-boolean:8>) vect_vec_iv_.8_37;
_39 = _38 + { 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF),
16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF), 16(OVF)
};
_40 = (vector(16) signed char) _39;
ivtmp.25_4 = ivtmp.25_6 + 16;
if (ivtmp.25_4 != _48)
the thing is that I'd consider the input broken:
__BB(3):
i_10 = __PHI (__BB2: 0, __BB4: i_11);
j_20 = __PHI (__BB2: _Literal (bool8) 0, __BB4: j_21);
...
j_21 = j_20 + _Literal (bool8) 1;
unless we iterate exactly once.
But the issue is of course that we mistake the signed BOOLEAN_TYPE : 8 as
mask when it is actually data.
Not sure where we get the (OVF)s from.
So the issue seems that get_same_sized_vectype, for a scalar boolean
type always produces a truth type. And it's vectorizable_induction
using that. I'm a bit confused as to why we have both a vector type
for the step and one for the actual induction. They should be always
the same ...
