llama.cpp

10b4f82d - Added comments explaining thread block size selection logic based on row count and column size, derived from historical commit context (#18212)