llama.cpp

cuda : fix -Wtautological-compare in fattn vec dispatch

#30519

Merged