[Bug target/126959] -march=znver4 prefers load from memory for mask register

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126959

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
It's a bit odd that adjusting store cost affects this testcase.  That said, I
cannot easily see how the cost of the kxnorw is computed and why that would
be prefered if the mask store cost is lower?!