I'm not working on superoptimization, but yes it does seem like eq(x,max_unsigned(x,y)) is an "obvious" cmpge_unsigned if one just browses all the intrinsics available here: https://www.intel.com/content/www/us/en/docs/intrinsics-guid...

I would like to read more about superoptimizers targeting AVX2, since previously I've complained that gcc will not make transformations between various equivalent instruction sequences for moving bytes around within and between registers, and some of these are a lot slower than others.