Hi, On Thu, Apr 5, 2012 at 1:07 PM, Christophe Gisquet <[email protected]> wrote: > 2012/4/5 Ronald S. Bultje <[email protected]>: >> That's ... quite a non-standard IDCT, generally you do a 1D IDCT, >> transpose, then another 1D IDCT, right? I don't quite see that here. >> Am I misunderstanding it? > > It's the case here: INIT has the row transform, then the STEP1 macro > performs the column transform without the help of a transpose. > > I went with a transpose at first, but as the 2nd stage goes from 16 to > to 32bits, that code ended better in my tests. Maybe now that I have > something working, I could try again, but I don't expect a lot from it > anyway.
I would expect it to be at least somewhat faster. Yeah, can you give it a try? Ronald _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
