[FFmpeg-devel] [PATCH 1/2] avcodec/options_table: disable CRC checking by default

[Reimar Döffinger]

Michael Niedermayer <michaelni at gmx.at> wrote:
>On Sun, Oct 27, 2013 at 08:20:47PM +0000, Paul B Mahol wrote:
>> On 10/27/13, Michael Niedermayer <michaelni at gmx.at> wrote:
>> > On Sun, Oct 27, 2013 at 07:53:27PM +0000, Paul B Mahol wrote:
>> >> On 10/27/13, Michael Niedermayer <michaelni at gmx.at> wrote:
>> >> > On Sun, Oct 27, 2013 at 06:55:06PM +0000, Paul B Mahol wrote:
>> >> >> On 10/27/13, Michael Niedermayer <michaelni at gmx.at> wrote:
>> >> >> > On Sun, Oct 27, 2013 at 05:11:18PM +0000, Paul B Mahol wrote:
>> >> >> >> On 10/27/13, Michael Niedermayer <michaelni at gmx.at> wrote:
>> >> >> >> > From: Anton Khirnov <anton at khirnov.net>
>> >> >> >> >
>> >> >> >> > Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
>> >> >> >> > ---
>> >> >> >> >  libavcodec/options_table.h |    2 +-
>> >> >> >> >  1 file changed, 1 insertion(+), 1 deletion(-)
>> >> >> >> >
>> >> >> >>
>> >> >> >> useless pointless bad
>> >> >> >
>> >> >> > its reqired for hevc, otherwise some hevc will be kind of
>slow.
>> >> >>
>> >> >> numbers please!
>> >> >
>> >> > its my pleassure
>> >> >
>> >> > 3 runs each of fate-suite/hevc-conformance/ENTP_C_LG_3.bit
>> >> >
>> >> > with crc flag
>> >> > 1841819858 decicycles in frame decode, 32 runs, 0 skips
>> >> > 1833937180 decicycles in frame decode, 32 runs, 0 skips
>> >> > 1843283884 decicycles in frame decode, 32 runs, 0 skips
>> >> >
>> >> > without crc flag
>> >> > 1703897301 decicycles in frame decode, 32 runs, 0 skips
>> >> > 1695031879 decicycles in frame decode, 32 runs, 0 skips
>> >> > 1697524383 decicycles in frame decode, 32 runs, 0 skips
>> >>
>> >> Not much.
>> >
>> > for a useless computation (from the end users point of view)
>> > in a major codec, its enough.
>> >
>> > applied
>> >
>> > thanks
>> 
>> Good, now i will ignore your comments too.
>
>iam not ignoring your comments
>you asked for numbers, i provided some
>they show that the enabled crc checks slow hevc down by 7% for a
>randomly picked file. Thats quite a huge amount for a important
>codec.
>your next comment was that its "not much", i have a different
>oppinon but i do not ignore yours. It just felt to me you where
>joking and trolling
>
>Also id like to make the ffmpeg 2.1 release today or tomorrow and
>a 7% slowdown for a check that is primarely usefull for debuging
>didnt seem to me to be something that would be controversal if its
>disabled by default

I agree, 7% is huge IMO, we should not just waste that kind of performance for no real reason.