[Bug tree-optimization/122736] loop masking with SIMD clones ICEs in vect_verify_full_masking

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122736

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
So we are doing

                vect_record_loop_mask (loop_vinfo,
                                       &LOOP_VINFO_MASKS (loop_vinfo),
                                       ncopies, vectype, op);

with ncopies == 1, vectype V4DF from

(gdb) p debug (slp_node)
t.c:10:1: note: node 0x512d2c0 (max_nunits=8, refcnt=1) vector(4) double
t.c:10:1: note: op template: _2 = .MASK_CALL (baz, d_1, _32);
t.c:10:1: note:         stmt 0 _2 = .MASK_CALL (baz, d_1, _32);
t.c:10:1: note:         children 0x512d020 0x512d368

but

(gdb) p loop_vinfo->vectorization_factor
$28 = {coeffs = {8}}
(gdb) p ncopies
$29 = 1
(gdb) p nunits
$30 = {coeffs = {8}}
(gdb) p group_size
$31 = 1

as the simdclone for 'baz' choosen wants a V8DFmode argument.  The VF
is because of some mask operation.

So either the 'vectype' or the 'ncopies' argument is wrong above.