https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121075
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2025-07-15
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed with -O3 -march=rv64gcv -fwrapv -fwhole-program -fsigned-char
(gdb) p *rgc
$9 = {max_nscalars_per_iter = 0, factor = 0, type = <tree 0x0>, compare_type =
<tree 0x0>, controls = {m_vec = 0x6279750 = {0x7ffff6c46dc8}},
bias_adjusted_ctrl = <tree 0x0>}
(gdb) p debug_tree (0x7ffff6c46dc8)
<ssa_name 0x7ffff6c46dc8
type <integer_type 0x7ffff6e4a1f8 public unsigned DI
size <integer_cst 0x7ffff6e391c8 constant 64>
unit-size <integer_cst 0x7ffff6e391e0 constant 8>
align:64 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
0x7ffff6e4a1f8 precision:64 min <integer_cst 0x7ffff6e395d0 0> max <integer_cst
0x7ffff6e03740 18446744073709551615>>
def_stmt GIMPLE_NOP
version:102>
so something goes wrong with the rgroup_controls. The 2nd one would be
(gdb) p (*(*controls).m_vec)[1]
$15 = (rgroup_controls &) @0x6268f60: {max_nscalars_per_iter = 1, factor = 1,
type = <vector_type 0x7ffff6fd72a0>, compare_type = <tree 0x0>, controls = {
m_vec = 0x627acd0 = {0x7ffff6c46948, 0x7ffff6c46990}}, bias_adjusted_ctrl =
<tree 0x0>}
which at least has a type. I guess vect_verify_loop_lens should have rejected
this, but that doesn't look at the controls at all.
