The following removes the scalar mask output from vect_check_scalar_mask
and deals with the fallout, eliminating uses of it. That's mostly
replacing checks on 'mask' by checks on 'mask_node' but also realizing
PR121349 and fixing that up a bit in check_load_store_for_partial_vectors.
PR tree-optimization/121349
* tree-vect-stmts.cc (check_load_store_for_partial_vectors):
Get full SLP mask, reduce to uniform scalar_mask for further
processing if possible.
(vect_check_scalar_mask): Remove scalar mask output, remove
code conditional on slp_mask.
(vectorizable_call): Adjust.
(check_scan_store): Get and check SLP mask.
(vectorizable_store): Eliminate scalar mask variable.
(vectorizable_load): Likewise.