[Bug target/126770] inefficient reduction of comparison sum

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126770

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
        vmovd   %xmm0, %eax
        vpextrd $1, %xmm0, %edx
        addl    %edx, %eax

could be addressed on the target side by providing reduc_plus_scal_v2si