llama.cpp

ggml : use ggml_row_size where possible

#4472

Merged