Hi, while the asm part is not very readable, this patch can be summarized as having 2 sets of bi-weight functions depending on whether weight fractions can be simplified. This trick was also used in the SSSE3 x86 implementation, but the check was performed every call, while it really needs to be done only once per frame.
I haven't modified the neon code (build would probably fail), although it should be very simple to modify. Christophe
0001-rv40dsp-implement-prescaled-versions-for-biweight.patch
Description: Binary data
_______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
