llama.cpp

sync : ggml (im2col, GPU conv, 32-bit arm compat)

#4060

Merged