llama.cpp

a44d7712 - vulkan: Fix GGML_VULKAN_CHECK_RESULTS to better handle fusion (#16919)