I try to merge VP3's mmx decoder into theora. http://kyoto.cool.ne.jp/vp3/developers/theora-alpha3-MMXd-src.zip You can see the change by searching the keyword "_UsingMMX_" in all lib folder's file. From VP3???YO vp3@go8.enjoy.ne.jp
<20040803231955.GA32632@ebed.etf.cuni.cz> Message-ID: <4110662B.1040501@go8.enjoy.ne.jp>> Any chance this will work on linux sometimes?I'm sorry that I check it with VC++6.0 only now, so other compilers can't compile it except one which can analyze VC's inline assembler. If you want to use it on linux, You have to try to modify it to compile it with gcc or any other compilers you have. From VP3???YO
now someone has to convert this to gcc inline assambler. :-/ and its decoding only for now. after i found out that, i did recompile libtheora with CFLAGS="-march=pentium3 -O3" and could see a difference encoding ~ 48 seconds of video (352x288) on my P3 1,2GHz using ffmpeg2theora. libtheora without extra flags real 2m0.800s user 1m48.569s sys 0m0.195s libtheora compiled with CFLAGS="-march=pentium3 -O3" real 1m42.825s user 1m33.208s sys 0m0.175s j
> now someone has to convert this to gcc inline assambler. :-/ > and its decoding only for now. after i found out that, > i did recompile libtheora with CFLAGS="-march=pentium3 -O3" > and could see a difference encoding ~ 48 seconds of > video (352x288) on my P3 1,2GHz using ffmpeg2theora. > > libtheora without extra flags > real 2m0.800s > user 1m48.569s > sys 0m0.195s > > libtheora compiled with CFLAGS="-march=pentium3 -O3" > real 1m42.825s > user 1m33.208s > sys 0m0.175sA long time ago someone posted here mmx code for both encoder and decoder. Does anyone know if this source can run in the new theora? Or has the author ported it to the new source of theora? Jens
<20040803153916.4c2f5261.Jens.Arm@gmx.de> Message-ID: <Pine.GSO.4.58.0408030836070.19285@shell> On Tue, 3 Aug 2004, Jens Arm wrote:> A long time ago someone posted here mmx code for both encoder and decoder. > > Does anyone know if this source can run in the new theora? > Or has the author ported it to the new source of theora?I am not sure which source is being discussed here. Some time ago, however, I took the MMX- and SSE2-optimized IDCT functions from the original VP3 source tree and converted them to be used in the ffmpeg source tree, compilable with gcc. Is that of any use here? -- -Mike Melanson
<20040803153916.4c2f5261.Jens.Arm@gmx.de> <Pine.GSO.4.58.0408030836070.19285@shell> Message-ID: <20040803164045.3c46c430.Jens.Arm@gmx.de>> > A long time ago someone posted here mmx code for both encoder and decoder. > > > > Does anyone know if this source can run in the new theora? > > Or has the author ported it to the new source of theora? > > I am not sure which source is being discussed here. Some time ago, > however, I took the MMX- and SSE2-optimized IDCT functions from the > original VP3 source tree and converted them to be used in the ffmpeg > source tree, compilable with gcc. Is that of any use here?No Mike, the source I mean is more than one year old, perhaps some more. BTW - which theora-version is decodable by ffmpeg? No version I tried is working. Jens
On Tue, Aug 03, 2004 at 01:41:01PM +0900, VP3?$B%[%9%#YO wrote:> I try to merge VP3's mmx decoder into theora. > http://kyoto.cool.ne.jp/vp3/developers/theora-alpha3-MMXd-src.zip > > You can see the change by searching the keyword "_UsingMMX_" in all lib > folder's file. >Any chance this will work on linux sometimes? My notebook is too slow to _play_ theora files, but it's almost there (it's P3 @ 667MHz). So for me the question is whether to encode stuff into Theora or use DIVx. Thus I would be very interested in helping testing optimized theora decoders. Unfortunately this source You provided doesn't even run autogen.sh (beside the fact, that the mmx code wouldn't probably compile anyway). Is there something I'm missing on my system? I have: autoconf-2.59-3 automake-1.8.3-1 Thanks! Petr Tomasek -- Petr Tomasek, http://www.etf.cuni.cz/~tomasek/