llama.cpp

ggml-virtgpu: make the code thread safe

#19204

Merged