On Sat, Jan 7, 2012 at 5:39 PM, Ronald S. Bultje <[email protected]> wrote: > Hi, > > On Thu, Jan 5, 2012 at 12:18 PM, Vitor Sessak <[email protected]> wrote: > [..] > > Does this make non-SSE decoding noticeably slower, i.e. should we make > the interleaving be decided upon by the windowing implementation, like > we do for transpose-coeffs-while-reading in IDCT (I think h264, > prores, mpeg1/2/4 etc. all do that)?
I tried to benchmark the impact of this patch and I didn't saw anything measurable. If one day we have several different implementations of this code requiring different shuffling, it would make sense to move this buffer to MPADSPContext, since it is only touched by imdct funtions. ITOH, to do it now would require moving the imdct12 code to DSP, which is not worth the trouble IMHO. -Vitor _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
