Skip to content

Improve wiener filter C implementation using loop interchange

Michail Alvanos requested to merge malvanos/dav1d:wiener_c into master

Improve wiener filter C implementation using loop interchange. This will improve the performance in simple architectures.

Power 9 Before: wiener_chroma_8bpc_c: 57830.7 wiener_luma_8bpc_c: 75434.7 Power 9 After: wiener_chroma_8bpc_c: 48185.3 wiener_luma_8bpc_c: 47948.1

In modern x86 (Ryzen 1600) the difference is not noticeable.

Edited by Michail Alvanos

Merge request reports