llama.cpp

ggml-webgpu: support non-square subgroup matrix configs for Intel GPUs

#21669

Merged