llama.cpp
e9641756 - cuda : keep columns in registers for cooperative softmax (#28982)

Commit
1 day ago
cuda : keep columns in registers for cooperative softmax (#28982) Template soft_max_f32_parallelize_cols_single_row to keep column elements in registers when they fit, reducing memory traffic from 5 global memory passes to 2 passes. Assisted-by: Gemini 3.8 Flash
Author
Parents
Loading