[Bug tree-optimization/126789] Bogus bool vect_recog_mask_conversion_pattern with AVX512 style masks

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126789

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-*-* gcn

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
It seems to work fine with RISC-V on older trunk, possibly because of VnBImode
vs. integer mode masks.  I was looking at

static inline void
append_pattern_def_seq (vec_info *vinfo,
                        stmt_vec_info stmt_info, gimple *new_stmt,
                        tree vectype = NULL_TREE,
                        tree scalar_type_for_mask = NULL_TREE)
{
  gcc_assert (!scalar_type_for_mask
              == (!vectype || !VECTOR_BOOLEAN_TYPE_P (vectype)));
  if (vectype)
    {
      stmt_vec_info new_stmt_info = vinfo->add_stmt (new_stmt);
      STMT_VINFO_VECTYPE (new_stmt_info) = vectype;
      if (scalar_type_for_mask)
        new_stmt_info->mask_precision
          = GET_MODE_BITSIZE (SCALAR_TYPE_MODE (scalar_type_for_mask));

specifically which gets vector <bool:1> and bool:1 as types but then
mask_precision == 8.  Thus GCN should be similarly affected.