"Ronald S. Bultje" <[email protected]> writes: > Hi, > > On Wed, Dec 28, 2011 at 8:34 PM, Michael Kostylev > <[email protected]> wrote: >> >> On Wed Dec 28 23:13:50 2011 >> Måns Rullgård wrote: >> >>>>>>> Sometimes the vsynth1-dnxhd_720p_rd test fails on alpha, pa-risc >>>>>>> and sparc FATE machines. It explicitly uses '-threads 4' when >>>>>>> encoding, and for some reason vsynth1/dnxhd-720p-rd.dnxhd may be >>>>>>> spoiled (the first frame contains a small number of randomly placed >>>>>>> zero blocks). For comparison, the single-thread mode does not show >>>>>>> any issues. >>>>>> This sounds threading related. Feel like trying valgrind's Helgrind >>>>>> and/or >>>>>> DRD tools to examine this further? >>>>> >>>>> Thread debugging is another story. I meant that the hard-coded '-threads >>>>> 4' >>>>> option looks rather harmful, >>> >>> It is exposing a bug somewhere. That does not make it harmful in itself. >> >> It overrides the setting from the THREADS variable which was introduced >> for a purpose. OTOH there is --disable-pthreads.
None of that changes the fact that we're looking at a bug that should be fixed, not covered up by adjusting test parameters. > Well a problem is that -theads for encoding is used for slices also, > and thus changes output for some mpeg encodings. This should be split > but I haven't had time for that yet. So right now, -threads for > encoding should ideally not be customizable, only for decoding. This is true for MPEG1/2, not dnxhd. There are probably other good reasons for dropping that flag from the test, but those are outside to the scope of this thread. We know there is a bug somewhere; we should find it and fix it. -- Måns Rullgård [email protected] _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
