llama.cpp

ggml: adds CONV_2D op and direct GEMM Vulkan implementation

#14316

Merged