Eventually got access to some new power hardware (and I'd love to access some more) and I started to work on fixing and refreshing it the arch specific support for it.
VSX does provide nice intrinsics to load and store unaligned data and takes care also of the endianess issues between power8(little endian) and the rest. Sadly tricks like using vec_mergeh to unpack integers need to have some abstractions (see Patch 10 for an example, I have more coming) so I'm making alternative versions VSX-only hopefully working correctly on LE and BE. Altivec optimizations were my first contribution to the project (about 9 years ago) so feels sort of going back to the memory late to me =) _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
