CUDA: fix shared-memory WAR hazard in flash_attn_ext_vec - #30573

CUDA: fix shared-memory WAR hazard in flash_attn_ext_vec#30573

am17an merged 2 commits into

ggml-org:masterggml-org/llama.cpp:masterfrom

invaliddayta:cuda-fattn-vec-raceinvaliddayta/llama.cpp:cuda-fattn-vec-raceCopy head branch name to clipboard