Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Video Encoding > MPEG-4 AVC / H.264

Reply
 
Thread Tools Search this Thread Display Modes
Old 29th March 2008, 05:27   #181  |  Link
MythCreator
Registered User
 
Join Date: Dec 2007
Location: Beijing,China
Posts: 77
Quote:
Originally Posted by burfadel View Post
Well, the modified.final build of Mythcreator's is a fraction fractionally (...?!) slower than that of Bobor's modified 798. On a test I just did its was 99.998 percent (rounded up!) the performance of Bobor's modified build.

I'm just curious, if Mythcreator is willing to give it a try, purely as an experimental build whether he could try making a build with GCC 4.4.0. It is very much in testing stage, but snapshots can be downloaded by mirror sites. They're updated weekly according to the gcc site. One such site is:

http://gcc-ca.internet.bs/snapshots/

(of course then select the latest 4.4 folder at the bottom of the list)

Course could always build your own latest 4.4.0 snapshot from the GCC SVN
svn://gcc.gnu.org/svn/gcc/trunk
http://gcc.gnu.org/svn/gcc/trunk
^^ view contents

I realise you probably know all this already, I listed it in case anyone was wondering

GCC 4.4.0 supposedly has many optimisations that may finally allow it to be faster than GCC 3.4.6. Would be interesting to see anyway!


I know it,but 4.4.0 is still in develope, maybe have some serious bug in it
MythCreator is offline   Reply With Quote
Old 29th March 2008, 07:52   #182  |  Link
burfadel
Registered User
 
Join Date: Aug 2006
Posts: 2,235
I was actually taking that in to consideration! Like I said above, just as a purely experimental build just to see whether gcc 4.4.0 performs as well or outperforms the older 3.4.6. It may even be the case it, in terms of comparing it to 4.3.0 or 4.2.2, more stable for the use of x264 since due to the crashes and slowdowns somethings not quite right with those two revisions!

Just thought it would be interesting to see, and compare to your 4.3.0 build. Its ok if its too much trouble, just thought it may be interesting!
burfadel is offline   Reply With Quote
Old 29th March 2008, 13:58   #183  |  Link
morph166955
Registered User
 
Join Date: Mar 2006
Posts: 443
Quote:
Originally Posted by MythCreator View Post
I mean, is the Thread Pool.diff still useful at this time?
For those of us who have very fast octa-core machines it is. For those who don't have them, results have varied between minimal speed boost to nothing at all to even a slight decrease in speed. Unfortunately I haven't been able to find a diff for thread pool that will cleanly apply in a few months. The whole reason that it does still work for the higher speed machines is that our threads are completing and destroying themeslves faster then x264 is expecting them to so there is a lag time created. Its actually taking longer to create/destroy a thread then it is for the thread to do its job in some cases. Having the thread pool there makes it so that we don't have to wait. In tests I ran a while back, when I had the thread pool patch installed I was getting a pretty significant speed boost on SD content (HD content is large enough that it maxes the CPU out anyway).

If we could get a thread pool patch that applied cleanly to the current version that would be awesome. What would be even more awesome is just putting it into the git version in a way where under normal running it would do its normal process but where we could put like --threadpool on the cli and it would run in that mode. No idea how hard that would be to do but I think that would be optimal that way one can choose the threading method they want on the fly.
morph166955 is offline   Reply With Quote
Old 29th March 2008, 18:50   #184  |  Link
MythCreator
Registered User
 
Join Date: Dec 2007
Location: Beijing,China
Posts: 77
x264.798.modified.experimental.exe

General thread:
http://forum.doom9.org/showthread.php?t=130364

x264_aq_var.48.786.fixed.diff
http://forum.doom9.org/showthread.php?t=132760
x264.gaussian.cplxblur.01.diff
Dark Shikari: - gaussian cplxblur: gives a tiny improvement in 2pass ratecontrol
x264_me-prepass_DeathTheSheep.01.diff
http://forum.doom9.org/showthread.php?p=1093523
x264_2pass_vbv.7.diff
http://thread.gmane.org/gmane.comp.v...093/focus=3748
x264_hrd_pulldown.04_interlace.diff
- HRD and pulldown for HD compatibility, updated patch for interlacing
http://forum.doom9.org/showthread.ph...19#post1047919

Link to x264 patches collected: http://files.x264.nl/x264_patches/



make frofiled in GCC 4.4.0 20080328 experimental,totally for experiment & test



To test the speed change, Please use this build and my beta2.fixed , and bob0r's build

Last edited by MythCreator; 29th March 2008 at 18:54.
MythCreator is offline   Reply With Quote
Old 29th March 2008, 19:54   #185  |  Link
burfadel
Registered User
 
Join Date: Aug 2006
Posts: 2,235
The experimental version worked fine, no crashes or anything unexpected... with good results!
On the test clip I used, with 1000 frames (not much I know but good enough for this purpose), I did several runs of each and used the average of all the runs. I used each version after the other one, not do all the runs at the same time to ensure accurate results.

With 798.modified.final (GCC 4.3.0):
Speed: 37.11 fps

With Bobor's 798 modified (GCC 3.4.6):
Speed: 37.25 fps

With 798.experimental (GCC 4.4.0):
speed: 37.37 fps

Of course on each run the speeds were slightly different, but in each run the slowest of Bobor's build run was still faster than the fastest for 798.modified and the slowest run of 798.experimental was still faster than the fastest for Bobor's build.

GCC 4.4.0 looks promising to be faster and regain the speed lost so far with GCC 4.x.x! The final may even be slightly better again (although an assumption one would presume it will be optimised further).

Thanks for the trial, it looks good! Maybe others with different CPU's could also test, the results may be different with AMD's for example?...
burfadel is offline   Reply With Quote
Old 29th March 2008, 21:45   #186  |  Link
Dethis
Registered User
 
Join Date: Jan 2006
Posts: 57
Burfadel, thanks for the test

But, as MythCreator suggested, you should use the "beta2.fixed" instead of the ".final". The ".final" one contains the bssd patch which probably inserts some more computation load which is absend for the "bobor" and "experimental" versions.
Dethis is offline   Reply With Quote
Old 29th March 2008, 23:02   #187  |  Link
MasterNobody
Registered User
 
Join Date: Jul 2007
Posts: 511
Quote:
Originally Posted by morph166955 View Post
For those of us who have very fast octa-core machines it is. For those who don't have them, results have varied between minimal speed boost to nothing at all to even a slight decrease in speed. Unfortunately I haven't been able to find a diff for thread pool that will cleanly apply in a few months. The whole reason that it does still work for the higher speed machines is that our threads are completing and destroying themeslves faster then x264 is expecting them to so there is a lag time created. Its actually taking longer to create/destroy a thread then it is for the thread to do its job in some cases. Having the thread pool there makes it so that we don't have to wait. In tests I ran a while back, when I had the thread pool patch installed I was getting a pretty significant speed boost on SD content (HD content is large enough that it maxes the CPU out anyway).

If we could get a thread pool patch that applied cleanly to the current version that would be awesome. What would be even more awesome is just putting it into the git version in a way where under normal running it would do its normal process but where we could put like --threadpool on the cli and it would run in that mode. No idea how hard that would be to do but I think that would be optimal that way one can choose the threading method they want on the fly.
Here is my variant of thread-pool patch which was made relatively to current git version: http://stashbox.org/96770/x264_thread_pool.r798.diff
As I know Dark Shikari's AQ patch slightly conflicts with thread-pool patch (in ratecontrol.c) so one of them need some modifications for compatibility.
MasterNobody is offline   Reply With Quote
Old 30th March 2008, 02:06   #188  |  Link
morph166955
Registered User
 
Join Date: Mar 2006
Posts: 443
Quote:
Originally Posted by MasterNobody View Post
Here is my variant of thread-pool patch which was made relatively to current git version: http://stashbox.org/96770/x264_thread_pool.r798.diff
As I know Dark Shikari's AQ patch slightly conflicts with thread-pool patch (in ratecontrol.c) so one of them need some modifications for compatibility.
Sweet! Thanks! Can't wait to give it a shot.
morph166955 is offline   Reply With Quote
Old 31st March 2008, 05:55   #189  |  Link
MythCreator
Registered User
 
Join Date: Dec 2007
Location: Beijing,China
Posts: 77
x264.805.modified.experimental.exe

General thread:
http://forum.doom9.org/showthread.php?t=130364

x264.gaussian.cplxblur.01.diff
Dark Shikari: - gaussian cplxblur: gives a tiny improvement in 2pass ratecontrol
x264_me-prepass_DeathTheSheep.01.diff
http://forum.doom9.org/showthread.php?p=1093523
x264_2pass_vbv.7.diff
http://thread.gmane.org/gmane.comp.v...093/focus=3748
x264_hrd_pulldown.04_interlace.diff
- HRD and pulldown for HD compatibility, updated patch for interlacing
http://forum.doom9.org/showthread.ph...19#post1047919

Link to x264 patches collected: http://files.x264.nl/x264_patches/



make frofiled in GCC 4.4.0 20080328 experimental,totally for experiment & test
MythCreator is offline   Reply With Quote
Old 31st March 2008, 06:40   #190  |  Link
bob0r
Pain and suffering
 
bob0r's Avatar
 
Join Date: Jul 2002
Posts: 1,337
x264.805.modified.exe

General thread:
http://forum.doom9.org/showthread.php?t=130364

x264.gaussian.cplxblur.01.diff
Dark Shikari: - gaussian cplxblur: gives a tiny improvement in 2pass ratecontrol
x264_me-prepass_DeathTheSheep.01.diff
http://forum.doom9.org/showthread.php?p=1093523
x264_2pass_vbv.7.diff
http://thread.gmane.org/gmane.comp.v...093/focus=3748
x264_hrd_pulldown.04_interlace.diff
- HRD and pulldown for HD compatibility, updated patch for interlacing
http://forum.doom9.org/showthread.ph...19#post1047919

Link to x264 patches collected: http://files.x264.nl/x264_patches/

Last edited by bob0r; 1st April 2008 at 06:33. Reason: x264_2pass_vbv.6.diff > x264_2pass_vbv.7.diff
bob0r is offline   Reply With Quote
Old 31st March 2008, 14:35   #191  |  Link
burfadel
Registered User
 
Join Date: Aug 2006
Posts: 2,235
Another test with revision 805, the GCC 4.4.0 version was faster every time!

Bobor's normal and modified builds with GCC 3.4.6 for this test averaged 29.84 fps (they both averaged very close so I grouped them together, I believe the additional patches don't affect the settings I was using).

The GCC 4.4.0 build averaged 30.32fps, and again on each test the slowest run was still faster than the fastest run with the GCC 3.4.6 builds. I used the exact same settings for both, run from the command line.

It looks like GCC 4.4.0 will be a good option once finalised for Bobor's website, since its a significant improvement over 4.3.0 and lower!
burfadel is offline   Reply With Quote
Old 31st March 2008, 19:08   #192  |  Link
survivant001
Registered User
 
Join Date: Nov 2007
Posts: 449
Quote:
Originally Posted by bob0r View Post
x264.805.modified.exe

General thread:
http://forum.doom9.org/showthread.php?t=130364

x264.gaussian.cplxblur.01.diff
Dark Shikari: - gaussian cplxblur: gives a tiny improvement in 2pass ratecontrol
x264_me-prepass_DeathTheSheep.01.diff
http://forum.doom9.org/showthread.php?p=1093523
x264_2pass_vbv.6.diff
http://thread.gmane.org/gmane.comp.v...093/focus=3550
x264_hrd_pulldown.04_interlace.diff
- HRD and pulldown for HD compatibility, updated patch for interlacing
http://forum.doom9.org/showthread.ph...19#post1047919

Link to x264 patches collected: http://files.x264.nl/x264_patches/

why did you include vbv patch 7 ?
survivant001 is offline   Reply With Quote
Old 31st March 2008, 19:22   #193  |  Link
Wishbringer
Silent Reader
 
Wishbringer's Avatar
 
Join Date: Dec 2003
Location: Germany
Posts: 281
Tested x264 bob0r's gcc 3.4.6 build 805 vs. MythCreator's gcc 4.4.0 build 805:
System: EVGA 780i, QX6700 (at 3.2GHz - Multi=12), 8GB DDR2-800 RAM, Vista64 Ultimate

bob0r
Quote:
-[Information] Log for job1 (video, Test.avs -> Test_video_bob0r.mp4)
--[Information] [31.03.2008 19:47:05] Started handling job
--[Information] [31.03.2008 19:47:05] Preprocessing
--[NoImage] Job commandline: "C:\Program Files (x86)\megui\tools\x264\x264.805.modified.exe" --crf 18 --level 4.1 --ref 8 --mixed-refs --no-fast-pskip

--bframes 16 --b-pyramid --b-rdo --bime --weightb --direct auto --subme 7 --trellis 2 --analyse p8x8,b8x8,i4x4,i8x8 --8x8dct --vbv-bufsize 9000 --vbv-

maxrate 24000 --me umh --threads auto --thread-input --sar 4993:5000 --progress --no-dct-decimate --output "D:\Filme\Test_video_bob0r.mp4"

"D:\Filme\Test.avs" --aud --nal-hrd --me-prepass
--[Information] [31.03.2008 19:47:06] Encoding started
--[NoImage] Standard output stream:
--[NoImage] Standard error stream
---[NoImage] avis [info]: 1920x816 @ 25.00 fps (2977 frames)
---[NoImage] x264 [info]: using SAR=4993/5000
---[NoImage] x264 [warning]: DPB size (18800640) > level limit (12582912)
---[NoImage] x264 [info]: using cpu capabilities: MMX MMXEXT SSE SSE2 SSE3 SSSE3 Cache64
---[NoImage] mp4 [info]: initial delay 2 (scale 25)
---[NoImage] x264 [info]: slice I:44 Avg QP:17.87 size:181942 PSNR Mean Y:45.82 U:50.39 V:51.29 Avg:46.92 Global:44.63
---[NoImage] x264 [info]: slice P:1291 Avg QP:19.59 size:104266 PSNR Mean Y:43.02 U:48.19 V:49.09 Avg:44.21 Global:43.92
---[NoImage] x264 [info]: slice B:1642 Avg QP:21.21 size: 37879 PSNR Mean Y:42.19 U:48.12 V:48.98 Avg:43.46 Global:43.20
---[NoImage] x264 [info]: mb I I16..4: 16.9% 51.8% 31.3%
---[NoImage] x264 [info]: mb P I16..4: 3.3% 11.2% 5.1% P16..4: 49.2% 23.4% 6.2% 0.0% 0.0% skip: 1.6%
---[NoImage] x264 [info]: mb B I16..4: 0.7% 2.0% 1.0% B16..8: 44.8% 3.5% 5.0% direct: 9.8% skip:33.3%
---[NoImage] x264 [info]: 8x8 transform intra:56.1% inter:39.6%
---[NoImage] x264 [info]: direct mvs spatial:99.9% temporal:0.1%
---[NoImage] x264 [info]: ref P 80.0% 13.0% 2.8% 1.4% 0.8% 0.8% 0.6% 0.5%
---[NoImage] x264 [info]: ref B 83.7% 13.1% 1.5% 0.7% 0.4% 0.3% 0.2%
---[NoImage] x264 [info]: SSIM Mean Y:0.9805205
---[NoImage] x264 [info]: PSNR Mean Y:42.605 U:48.185 V:49.062 Avg:43.840 Global:43.514 kb/s:13759.53
---[NoImage] encoded 2977 frames, 2.41 fps, 13763.75 kb/s
--[Information] Final statistics
---[NoImage] Desired video bitrate: 18 kbit/s
---[NoImage] Obtained video bitrate (approximate: 13764 kbit/s
--[Information] [31.03.2008 20:07:45] Job completed
--[Information] [31.03.2008 20:07:45] Postprocessing
---[Information] Deleting intermediate files

MythCreator
Quote:
-[Information] Log for job1 (video, Test.avs -> Test_video_MythCreator.mp4)
--[Information] [31.03.2008 19:15:23] Started handling job
--[Information] [31.03.2008 19:15:23] Preprocessing
--[NoImage] Job commandline: "C:\Program Files (x86)\megui\tools\x264\x264.805.modified.experimental.exe" --crf 18 --level 4.1 --ref 8 --mixed-refs --no-fast-pskip --bframes 16

--b-pyramid --b-rdo --bime --weightb --direct auto --subme 7 --trellis 2 --analyse p8x8,b8x8,i4x4,i8x8 --8x8dct --vbv-bufsize 9000 --vbv-maxrate 24000

--me umh --threads auto --thread-input --sar 4993:5000 --progress --no-dct-decimate --output "D:\Filme\Test_video_MythCreator.mp4" "D:\Filme\Test.avs"

--aud --nal-hrd --me-prepass
--[Information] [31.03.2008 19:15:24] Encoding started
--[NoImage] Standard output stream:
--[NoImage] Standard error stream
---[NoImage] avis [info]: 1920x816 @ 25.00 fps (2977 frames)
---[NoImage] x264 [info]: using SAR=4993/5000
---[NoImage] x264 [warning]: DPB size (18800640) > level limit (12582912)
---[NoImage] x264 [info]: using cpu capabilities: MMX MMXEXT SSE SSE2 SSE3 SSSE3 Cache64
---[NoImage] mp4 [info]: initial delay 2 (scale 25)
---[NoImage] x264 [info]: slice I:44 Avg QP:17.87 size:181942 PSNR Mean Y:45.82 U:50.39 V:51.29 Avg:46.92 Global:44.63
---[NoImage] x264 [info]: slice P:1291 Avg QP:19.58 size:104282 PSNR Mean Y:43.02 U:48.19 V:49.09 Avg:44.21 Global:43.92
---[NoImage] x264 [info]: slice B:1642 Avg QP:21.21 size: 37887 PSNR Mean Y:42.19 U:48.12 V:48.98 Avg:43.46 Global:43.20
---[NoImage] x264 [info]: mb I I16..4: 16.9% 51.8% 31.3%
---[NoImage] x264 [info]: mb P I16..4: 3.3% 11.2% 5.1% P16..4: 49.2% 23.4% 6.2% 0.0% 0.0% skip: 1.6%
---[NoImage] x264 [info]: mb B I16..4: 0.7% 2.0% 1.0% B16..8: 44.7% 3.5% 5.0% direct: 9.9% skip:33.3%
---[NoImage] x264 [info]: 8x8 transform intra:56.0% inter:39.6%
---[NoImage] x264 [info]: direct mvs spatial:99.9% temporal:0.1%
---[NoImage] x264 [info]: ref P 80.0% 13.0% 2.8% 1.4% 0.9% 0.8% 0.6% 0.5%
---[NoImage] x264 [info]: ref B 83.6% 13.1% 1.5% 0.7% 0.4% 0.3% 0.2%
---[NoImage] x264 [info]: SSIM Mean Y:0.9805220
---[NoImage] x264 [info]: PSNR Mean Y:42.605 U:48.184 V:49.062 Avg:43.840 Global:43.514 kb/s:13761.79
---[NoImage] encoded 2977 frames, 2.41 fps, 13766.00 kb/s
--[Information] Final statistics
---[NoImage] Desired video bitrate: 18 kbit/s
---[NoImage] Obtained video bitrate (approximate: 13767 kbit/s
--[Information] [31.03.2008 19:36:01] Job completed
--[Information] [31.03.2008 19:36:01] Postprocessing
---[Information] Deleting intermediate files
bob0r's build has a slightly lower bitrate: 13763.75 kb/s vs. 13766.00 kb/s
maybe because of "x264_2pass_vbv.6.diff" instead of "x264_2pass_vbv.7.diff"

both encoded at 2.41 fps: bob0r = 20min40sec; MythCreator = 20min38sec
2secs difference seems to be fault-tolerance

Last edited by Wishbringer; 31st March 2008 at 19:26. Reason: typos
Wishbringer is offline   Reply With Quote
Old 31st March 2008, 20:25   #194  |  Link
bob0r
Pain and suffering
 
bob0r's Avatar
 
Join Date: Jul 2002
Posts: 1,337
Quote:
Originally Posted by survivant001 View Post
why did you include vbv patch 7 ?
Fixed, just a typo, 7 was used.

Seems there is 8 now?
bob0r is offline   Reply With Quote
Old 31st March 2008, 22:19   #195  |  Link
tenkai
Registered User
 
Join Date: Feb 2008
Posts: 14
L 4.1 ref 8? and 16 bframes? Do you think that will work out? I´m just wondering if i missed anything.. wasn´t the max 5 ref on 1920x800 for example and 3 bframes on L.4.1 encoding?
tenkai is offline   Reply With Quote
Old 31st March 2008, 22:24   #196  |  Link
Wishbringer
Silent Reader
 
Wishbringer's Avatar
 
Join Date: Dec 2003
Location: Germany
Posts: 281
Quote:
Originally Posted by bob0r View Post
Fixed, just a typo, 7 was used.
Now I am a bit curios. I thought that same build with same patches should produce same bitcompatible encoded output of videostream, independend of used compiler...

See my previous post, where bob0r's build produced a slightly smaller encoded video.
Wishbringer is offline   Reply With Quote
Old 31st March 2008, 22:28   #197  |  Link
Wishbringer
Silent Reader
 
Wishbringer's Avatar
 
Join Date: Dec 2003
Location: Germany
Posts: 281
@tenkai:

I didn't said, that this is a useable clip on any standalone player.
I used my PS3-SD profile on a HD clip, because I wanted to test how encoding speed is in compare with nearly all options maxed out.
On the other hand, these settings work very well with my SD DVD conversions for my PS3.
Wishbringer is offline   Reply With Quote
Old 31st March 2008, 22:44   #198  |  Link
tenkai
Registered User
 
Join Date: Feb 2008
Posts: 14
oki.. SD with ref 8 etc shouldn´t be a prob, yea. so am i right? Its still the max for 1080p encoding to fit quality and full support on ps3 etc ref 4/5 but with maximal 3 bframes? I´m askin coz i want to have the maximum possible quality but having a compatible encode on the other hand.. and i have no idea if i can reach that with ref 5 and 3 bfr.. even with subme 7 etc
tenkai is offline   Reply With Quote
Old 31st March 2008, 22:46   #199  |  Link
tenkai
Registered User
 
Join Date: Feb 2008
Posts: 14
some ideas, tweaks are welcomed ofcoz
tenkai is offline   Reply With Quote
Old 2nd April 2008, 07:56   #200  |  Link
bob0r
Pain and suffering
 
bob0r's Avatar
 
Join Date: Jul 2002
Posts: 1,337
x264.808.modified.exe

General thread:
http://forum.doom9.org/showthread.php?t=130364

x264.gaussian.cplxblur.01.diff
Dark Shikari: - gaussian cplxblur: gives a tiny improvement in 2pass ratecontrol
x264_me-prepass_DeathTheSheep.01.diff
http://forum.doom9.org/showthread.php?p=1093523
x264_2pass_vbv.7.diff
http://thread.gmane.org/gmane.comp.v...093/focus=3748
x264_hrd_pulldown.04_interlace.diff
- HRD and pulldown for HD compatibility, updated patch for interlacing
http://forum.doom9.org/showthread.ph...19#post1047919

Link to x264 patches collected: http://files.x264.nl/x264_patches/
bob0r is offline   Reply With Quote
Reply

Tags
h.264, x264, x264 builds, x264 patches, x264 unofficial builds

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 16:29.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2018, vBulletin Solutions Inc.