[FFmpeg-cvslog] r25196 - in trunk: configure doc/filters.texi libavfilter/Makefile libavfilter/allfilters.c libavfilter/vf_yadif.c libavfilter/x86 libavfilter/x86/yadif.c libavfilter/yadif.h

Ronald S. Bultje rsbultje
Wed Sep 29 16:07:15 CEST 2010


Hi,

On Wed, Sep 29, 2010 at 9:00 AM, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Wed, Sep 29, 2010 at 08:34:40AM -0400, Ronald S. Bultje wrote:
>> On Tue, Sep 28, 2010 at 7:56 AM, Michael Niedermayer <michaelni at gmx.at> wrote:
>> > On Mon, Sep 27, 2010 at 08:31:22PM -0400, Alexander Strange wrote:
>> >> On Sep 27, 2010, at 1:46 PM, Ronald S. Bultje wrote:
>> >> > On Sat, Sep 25, 2010 at 12:43 PM, michael <subversion at mplayerhq.hu> wrote:
>> >> >> Author: michael
>> >> >> Date: Sat Sep 25 18:43:42 2010
>> >> >> New Revision: 25196
>> >> >>
>> >> >> Log:
>> >> >> yadif filter, based on stefanos port of my yadif from mplayer.
>> >> >> Compared to stefanos, 2 frame output works with ffplay.
>> >> >
>> >> > http://fate.ffmpeg.org/x86_32-macosx-gcc-4.2.1/20100927154653
>> >> >
>> >> > Broke OSX build. Haven't had time to look at it yet.
>> >>
>> >> > ? ? ? ? ? ? :[prev] "r"(prev),\
>> >> > ? ? ? ? ? ? ?[cur] ?"r"(cur),\
>> >> > ? ? ? ? ? ? ?[next] "r"(next),\
>> >> > ? ? ? ? ? ? ?[prefs]"r"((x86_reg)refs),\
>> >> > ? ? ? ? ? ? ?[mrefs]"r"((x86_reg)-refs),\
>> >> > ? ? ? ? ? ? ?[pw1] ?"m"(pw_1),\
>> >> > ? ? ? ? ? ? ?[pb1] ?"m"(pb_1),\
>> >>
>> >>
>> >> 7 registers with PIC.
>> >> Moving pw/pb_1 to MANGLE() fixes clang -m32 -fomit-frame-pointer.
>> >> 5 registers should work, but removing -fomit-frame-pointer too still makes it fail.
>> >>
>> >
>> >> ?yadif.c | ? 13 ++++++-------
>> >> ?1 file changed, 6 insertions(+), 7 deletions(-)
>> >> 8feb4f50ce2609042645bc6cc3354e37247d6a45 ?yadif_mangle.diff
>> >
>> > looks ok
>>
>> Applied.
>>
>> do we want to share pw_1 and pb_1 between libavfilter and libavcodec
>> or would that slow it down?
>
> with real existing tools i think it will break right and left.
> in theory copy relocations could allow PIC and such shared constants to
> coexists without speedloss

I'll leave it for now then, it's only 16 bytes... If it becomes a
bigger issue I'll look into it.

Ronald



More information about the ffmpeg-cvslog mailing list