Recent JM h264 Reference S/W Binary Req.

[Pookie]

Might some kind soul compile an exe of Ldecode from the latest
source? 13-Mar-2007

http://iphome.hhi.de/suehring/tml/download/jm12.2.zip

The docs claim it is faster...(we can only hope)

Thanks for your consideration.

[bob0r]

$ unixprep.sh
/bin/sed
Creating obj directories...
Removing DOS LF chars...
Done.



$ make

checking dependencies

compiling object file "obj/annexb.o" ...
In file included from inc/global.h:39,
from src/annexb.c:18:
inc/ifunctions.h:21: error: syntax error before "int"
inc/ifunctions.h:26: error: syntax error before "int"
inc/ifunctions.h:31: error: syntax error before "double"
inc/ifunctions.h:36: error: syntax error before "double"
inc/ifunctions.h:41: error: syntax error before "int64"
inc/ifunctions.h:46: error: syntax error before "int64"
inc/ifunctions.h:51: error: syntax error before "int"
inc/ifunctions.h:56: error: syntax error before "double"
inc/ifunctions.h:61: error: syntax error before "int64"
inc/ifunctions.h:66: error: syntax error before "double"
inc/ifunctions.h:71: error: syntax error before "int"
inc/ifunctions.h:75: error: syntax error before "int64"
inc/ifunctions.h:80: error: syntax error before "int"
inc/ifunctions.h:85: error: syntax error before "int"
inc/ifunctions.h:90: error: syntax error before "int"
inc/ifunctions.h:95: error: syntax error before "unsigned"
inc/ifunctions.h:100: error: syntax error before "int"
inc/ifunctions.h:105: error: syntax error before "unsigned"
inc/ifunctions.h:110: error: syntax error before "int"
inc/ifunctions.h:118: error: syntax error before "int"
inc/ifunctions.h:126: error: syntax error before "double"
inc/ifunctions.h:134: error: syntax error before "int"
inc/ifunctions.h:139: error: syntax error before "int"
make: *** [obj/annexb.o] Error 1


Advice needed!

[akupenguin]

@Pookie
Yep, now only 8x slower than lavc, as opposed to JM12.1 which was 12x slower.

@bob0r
Works For Me.

[Pookie]

akupenguin - Agreed, but until PAFF support happens, the options are limited...

[celtic_druid]

http://mirror.brokenlogic.net/jm12.2.7z

[Pookie]

Big thanks, fellas. I truly appreciate the effort. I'll report back on the speed difference.


EDIT:

v10.2 - Decoded in 128.986 seconds

v12.2 - Decoded in 50.874 seconds:D

[Guest]

What is your figure for 12.1?

[Pookie]

If you can point me to a 12.1 .exe, I'd be happy to post that result.

[drmpeg]

If you can point me to a 12.1 .exe, I'd be happy to post that result.
12.1 compiled with MS Visual Studio .NET 2003

http://www.w6rz.net/jmfrextdecoder.zip

Ron

[Pookie]

Thanks, Ron.

Interesting results - not quite what I expected in 12.1

----------------------------- JM 10.2 (FRExt) -----------------------------
Execution time: 65.227 s
----------------------------- JM 12.1 (FRExt) -----------------------------
Execution time: 340.201 s
----------------------------- JM 12.2 (FRExt) -----------------------------
Execution time: 22.215 s

[Guest]

Maybe he gave you a debug build!

What was the real time length of the clip that yielded those last timings, please?

[Pookie]

45 frames , 29.97, 1440x1080 AVCHD source, 2.4GHZ P4 CPU

PM me with an FTP password and I'll send you the whole testing kit...

Actually, no need. Here it is. 2.4MB

http://www.momoshare.com/file.php?file=a19092e9db40d7a60e6094fbdd031429

Performance results piped out to output.log

[drmpeg]

Maybe he gave you a debug build!

What was the real time length of the clip that yielded those last timings, please?
Doh! It was a debug build. I guess I learned something today. I didn't realize debug had such a performance impact. I mistakenly assumed debug just added a symbol table. I'll have to update xport also (although it's probably I/O bound).

Here's the release build for 12.1:

http://www.w6rz.net/ldecod.zip

Since other sites have linked to the old url, I've updated that with a release build of 12.2:

http://www.w6rz.net/frextdecoder.zip

Ron

[Guest]

I mistakenly assumed debug just added a symbol table. You know it now but just for other readers I'll point out that the debug build disables all code optimizations, so that tracing through the source code is coherent. Suppose some code got optimized away while the source code file in your debugger window shows that code -- it would make debugging very confusing and difficult.

@Pookie

If you could give us the figure for 12.1 with the optimized build that would be useful. Thank you.

[Pookie]

Will do...

Edit - Okey dokey, same test, using the 12.1 release build.

=== 10.2.exe -i bits0001.mpv -o output.yuv ===

Execution time: 63.650 s

=== 12.1.exe -i bits0001.mpv -o output.yuv ===

Execution time: 74.700 s

=== 12.2.exe -i bits0001.mpv -o output.yuv ===

Execution time: 22.986 s


Significant speedups in 12.2... Maybe it won't matter after the Summer of Code work (you still sure you want to pass on that project, Don ? :D )

[Guest]

Significant speedups in 12.2... Maybe it won't matter after the Summer of Code work (you still sure you want to pass on that project, Don? I can't commit to a fixed schedule although I still plan to tackle PAFF. If someone else gets there first, that's fine. Do you know if anyone has signed on to this project?

[akupenguin]

PAFF is not on the final list of SoC projects (http://code.google.com/soc/ffmpeg/about.html).

[Pookie]

Bummer, I was glad to see this when it came out:

http://wiki.multimedia.cx/index.php?title=Summer_Of_Code_2007

[celtic_druid]

http://mirror.brokenlogic.net/jm12.1.7z

You might want to compare the above build since it is ICL compiled like my 12.2 build.

[Pookie]

=== 10.2.exe -i bits0001.mpv -o output.yuv ===

Execution time: 59.581 s

=== 12.1.ms.exe (MS Build) -i bits0001.mpv -o output.yuv ===

Execution time: 74.502 s

=== 12.1.exe (Intel Build) -i bits0001.mpv -o output.yuv ===

Execution time: 35.903 s

=== 12.2.exe (Intel Build) -i bits0001.mpv -o output.yuv ===

Execution time: 22.586 s

I guess we've learned 2 things - Intel's compiler is more optimized for this particular app than Microsoft's, and some major optimizations did occur to the code between 12.1 and 12.2.

[akupenguin]

My guess as to why there's so much difference: icc does autovectorization and msvc doesn't. JM doesn't have any asm optimizations, so converting a few simple loops to sse makes a big difference.

[Warren]

PAFF is not on the final list of SoC projects (http://code.google.com/soc/ffmpeg/about.html).

I think that list is still the 2006 projects list.

[akupenguin]

No, I'm one of the mentors, and I can confirm that those projects are for this year.

[Warren]

Oh weird, I just saw that the same guy is working on E-AC3 this year an figured it was the old list.

[drmpeg]

JM reference decoder version 12.3 now available:

http://iphome.hhi.de/suehring/tml/download/jm12.3.zip

Has new 4:4:4 independent mode profile.

Ron

[bob0r]

Nice, celtic_druid might a compile again? :)

[celtic_druid]

Compiled it the other day
http://mirror.brokenlogic.net/jm12.3.7z

[MothaFUNKa]

Might some kind soul (celtic_druid ;-) ) please compile an exe of Lencode / ldecode from the latest source available since the beginning of last week?

http://iphome.hhi.de/suehring/tml/download/jm13.0.zip

Thx in advance!

[Terranigma]

Might some kind soul (celtic_druid ;-) ) please compile an exe of Lencode / ldecode from the latest source available since the beginning of last week?

Requested (http://www.zshare.net/download/47356681140d58/)

[Sharktooth]

Requested (http://www.zshare.net/download/47356681140d58/)
what compiler you used?

[Terranigma]

what compiler you used?

vs.net 2005

[MothaFUNKa]

So I did it myself.
Have fun, lay back and enjoy ;)

http://rapidshare.com/files/68311808/bin_jm13.0.zip

[titof]

Hello everybody
Thanks celtic druid for the compiled decoder (12.3)

Do you know where I can find compiled last version of the decoder, the coder and if it's possible the rtpdump.

I tried but it doesn't work
Thanks

:helpful:

[LessThanJake]

Hello,

I tried to compile the JM H264 reference encoder / decoder (JM 15.0) (http://iphome.hhi.de/suehring/tml/download/), with enabled trace option in the "defines.h" for the decoder, with msys and gcc. But it fails.

I did this:

Zero@ECKE /c/dev/sources/jm/ldecod
$ cd /c/dev/sources/jm/

Zero@ECKE /c/dev/sources/jm
$ ls
CHANGES.TXT bin jm_vc8.sln lencod_vc7.vcproj
Changes_detail.txt copyright.txt ldecod lencod_vc8.vcproj
FREXT_changes.txt disclaimer.txt ldecod_vc7.vcproj rtp_loss
Makefile doc ldecod_vc8.vcproj rtpdump
Readme.txt jm_vc7.sln lencod unixprep.sh

Zero@ECKE /c/dev/sources/jm
$ ./unixprep.sh
/bin/sed
Removing DOS LF chars...
Done.

Zero@ECKE /c/dev/sources/jm
$ cd /c/dev/sources/jm/ldecod

Zero@ECKE /c/dev/sources/jm/ldecod
$ make
Creating obj ...

checking dependencies

compiling object file "obj/annexb.o" ...
In file included from inc/global.h:38,
from src/annexb.c:15:
inc/ifunctions.h:26: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:31: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:36: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'double'
inc/ifunctions.h:41: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'double'
inc/ifunctions.h:46: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int64'
inc/ifunctions.h:51: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int64'
inc/ifunctions.h:56: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:61: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:66: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'double'
inc/ifunctions.h:71: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int64'
inc/ifunctions.h:76: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'double'
inc/ifunctions.h:81: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:86: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int64'
inc/ifunctions.h:91: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:96: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:101: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:106: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:111: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'unsigned'
inc/ifunctions.h:116: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:121: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'unsigned'
inc/ifunctions.h:126: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:134: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:142: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'double'
inc/ifunctions.h:150: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:155: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:160: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:165: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
inc/ifunctions.h:176: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'float'
src/annexb.c:35: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
src/annexb.c:56: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'byte'
src/annexb.c:84: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'int'
src/annexb.c: In function 'GetAnnexbNALU':
src/annexb.c:152: warning: implicit declaration of function 'getfbyte'
src/annexb.c:230: warning: implicit declaration of function 'FindStartCode'
src/annexb.c: In function 'OpenAnnexBFile':
src/annexb.c:310: warning: implicit declaration of function 'getChunk'
make: *** [obj/annexb.o] Error 1


Can someone explain the errors, or compile some binaries with enabled trace-option for me?

Thank you.
greets
LTJ

Edit //
Done.

[k-c-ksum]

sorry to dig up an old thread but does anyone have the latest version compiled

thanks