On Thu, Aug 1, 2019 at 3:56 AM David Bryant <[email protected]> wrote:
> > On 7/30/19 1:43 AM, Carl Eugen Hoyos wrote: > > Am So., 28. Juli 2019 um 23:35 Uhr schrieb Paul B Mahol < > [email protected]>: > >> Hi, > >> > >> patch attached. > > As just posted on irc: > > On an 8-core Intel cpu, this makes decoding slower by a factor two, > > heating the cpu > > ten times as much, on a multi-core powerpc, decoding is nearly two > > times slower, and > > heats nearly eight times as much. > > I can verify the greatly increased CPU load on a dual core system also. I > also verified that the outputted audio data is correct > (identical to unmodified code) and that there are no more samples being > processed, indicating it's probably just threading overhead. > > The dsd2pcm conversion is pretty optimized and I suspect that's just way > too fast to be worth multi-threading. > It appears it is not, see new patches. As suspected using uint8s and unaligned bufer for stuff leads to big slowdown when multi-threading. > > As I suggested earlier, it's the DST decompression that is worth looking > into for threading. > > -David > > > > > > Carl Eugen > > _______________________________________________ > > ffmpeg-devel mailing list > > [email protected] > > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > > > To unsubscribe, visit link above, or email > > [email protected] with subject "unsubscribe". > > > _______________________________________________ > ffmpeg-devel mailing list > [email protected] > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email > [email protected] with subject "unsubscribe". _______________________________________________ ffmpeg-devel mailing list [email protected] https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email [email protected] with subject "unsubscribe".
