llama.cpp

ggml-cuda: gdn use shared mem for HIP

#20282

Closed