CUDA: fastdiv, launch bounds for mmvq + q8_1 quant#15802

JohannesGaessler merged 2 commits intoggml-org:masterggml-org/llama.cpp:masterfrom

JohannesGaessler:cuda-mmvq-fastdiv-3JohannesGaessler/llama.cpp:cuda-mmvq-fastdiv-3Copy head branch name to clipboard