ggml-cuda : reduce redundant expf calls in GDN KDA kernel#20295

simonxluo wants to merge 5 commits intoggml-org:masterggml-org/llama.cpp:masterfrom

simonxluo:ggml-cuda-gdn-expf-optsimonxluo/llama.cpp:ggml-cuda-gdn-expf-optCopy head branch name to clipboard