llama.cpp

ggml-cuda : fix CDNA2 compute capability constant for gfx90a (MI210)

#21519

Merged