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 > New and alternative video codecs

Reply
 
Thread Tools Search this Thread Display Modes
Old 4th February 2012, 01:29   #14641  |  Link
haruhiko_yamagata
Registered User
 
Join Date: Feb 2006
Location: Japan
Posts: 1,560
Quote:
Originally Posted by ipanema View Post
Given that all 3 of the files I selected at random show the problem within the first few frames, I would think that you would be able to reproduce this quite easily using one of your own files (all 3 were from different AVCHD camcorders).

However, if you still think you need an example file, let me know and I will upload one for you.
Finally I found one, but it was hard...
With one file, in combination with Haali's splitter.
__________________
[ Download ffdshow | Wiki ]
haruhiko_yamagata is offline   Reply With Quote
Old 4th February 2012, 08:41   #14642  |  Link
haruhiko_yamagata
Registered User
 
Join Date: Feb 2006
Location: Japan
Posts: 1,560
ipanema, please upload the file anyway. The file I mentioned above has the same timestamp as the older build.
Which source/splitter filter do you use?
__________________
[ Download ffdshow | Wiki ]
haruhiko_yamagata is offline   Reply With Quote
Old 4th February 2012, 11:18   #14643  |  Link
haruhiko_yamagata
Registered User
 
Join Date: Feb 2006
Location: Japan
Posts: 1,560
MSVC vc ICL x64

Madagascar.avi 1280x720
Average of about 2760 frames (QueryPerformanceCounter, which is high resolution time. Smaller number is faster.).
Image filter only, not overall frame rate.

RGB conversion (HQ, dithering enabled)
MSVC x86: 5,415
ICL x86: 5,488
MCVC x64: 4,856
ICL x64: 4,961

Denoise 3D HQ
MSVC x86: 36,047
ICL x86: 31,716
MSVC x64: 419,569
ICL x64: 32,220

Denoise 3D Normal
MSVC x86: 25,618
ICL x86: 18,753
MSVC x64: 16,917
ICL x64: 18,715

As you can see, ICL is faster only in special cases.
ICL is faster by 36.6% for Denoise 3D normal x86. This filter is useless because it's quality is too low. Denoise 3D HQ should be used.
ICL is 13 times faster for Denoise 3D HQ x64. MSVC does not support MMX intrinsics for x64. ICL seems to be OK.

EDIT
Core2 Quad Q9550 2.83GHz
Windows 7 Professional sp1
GeForce 7900 GS
ffdshow rev 4298
__________________
[ Download ffdshow | Wiki ]

Last edited by haruhiko_yamagata; 4th February 2012 at 11:45.
haruhiko_yamagata is offline   Reply With Quote
Old 4th February 2012, 11:23   #14644  |  Link
khagaroth
Registered User
 
khagaroth's Avatar
 
Join Date: Feb 2006
Posts: 103
Quote:
Originally Posted by clsid View Post
I am making preparations for a new official release that is planned for later this month, once remaining regressions have been fixed.
Up to update the Czech translation I guess. What a pity it's such a pain to find changed/new strings, it would be great if the whole thing got gettextized. As probability of that ever happening is pretty low, it would be nice if this ticket was followed.
khagaroth is offline   Reply With Quote
Old 4th February 2012, 11:26   #14645  |  Link
nevcairiel
Registered Developer
 
Join Date: Mar 2010
Location: Hamburg/Germany
Posts: 10,336
Quote:
Originally Posted by haruhiko_yamagata View Post
MSVC does not support MMX intrinsics for x64. ICL seems to be OK.
So rewrite them using SSE2, its really just the same commands with 128-bit instead 64...
__________________
LAV Filters - open source ffmpeg based media splitter and decoders
nevcairiel is offline   Reply With Quote
Old 4th February 2012, 12:17   #14646  |  Link
haruhiko_yamagata
Registered User
 
Join Date: Feb 2006
Location: Japan
Posts: 1,560
Quote:
Originally Posted by nevcairiel View Post
So rewrite them using SSE2, its really just the same commands with 128-bit instead 64...
That's a bother.
That's something compilers must do. MSVC is simply inferior to ICL in that regard.
__________________
[ Download ffdshow | Wiki ]

Last edited by haruhiko_yamagata; 4th February 2012 at 12:36.
haruhiko_yamagata is offline   Reply With Quote
Old 4th February 2012, 12:33   #14647  |  Link
haruhiko_yamagata
Registered User
 
Join Date: Feb 2006
Location: Japan
Posts: 1,560
Quote:
Originally Posted by khagaroth View Post
Up to update the Czech translation I guess. What a pity it's such a pain to find changed/new strings, it would be great if the whole thing got gettextized. As probability of that ever happening is pretty low, it would be nice if this ticket was followed.
Tags such as <translate> in the svn log may help. I think it's a good idea.
As far as I can find,
  • Image buffer: in Subtitles::Text page
  • Primary output color space: in Output page
Both have tool-tip hint.
__________________
[ Download ffdshow | Wiki ]
haruhiko_yamagata is offline   Reply With Quote
Old 4th February 2012, 13:19   #14648  |  Link
ipanema
Registered User
 
Join Date: Apr 2009
Posts: 93
Quote:
Originally Posted by haruhiko_yamagata View Post
ipanema, please upload the file anyway. The file I mentioned above has the same timestamp as the older build.
Which source/splitter filter do you use?
Sorry you've had some problems finding a file - it seemed to do this to all 3 files that I examined. Perhaps this does have something to do with the splitter. The splitter being used is the Microsoft MPEG-2 Demux that is part of Windows which works fine with all decoders I've tried, including ffdshow - but it's just the recent ffdshow revs that cause the problem. The Microsoft splitter produces fixed sized 8192 byte samples, where only the sample which STARTS a frame carries a start/end timestamp - all other samples have no timestamp. Also the sample at the end of a frame may be less than 8192 bytes in size.

Here's a sample file - it's pretty small but is enough to show the problem. It is the "File 3" in my previous message.

http://www.mediafire.com/?88db0kvuw83ubh5
ipanema is offline   Reply With Quote
Old 4th February 2012, 14:23   #14649  |  Link
haruhiko_yamagata
Registered User
 
Join Date: Feb 2006
Location: Japan
Posts: 1,560
Thanks for the sample.
Quote:
Originally Posted by ipanema View Post
The splitter being used is the Microsoft MPEG-2 Demux that is part of Windows
Is it "MPEG-2 Demultiplexer" or "MPEG-2 Splitter"?
What is the OS?
"MPEG-2 Demux" isn't installed on my PC.
__________________
[ Download ffdshow | Wiki ]
haruhiko_yamagata is offline   Reply With Quote
Old 4th February 2012, 15:55   #14650  |  Link
egur
QuickSync Decoder author
 
Join Date: Apr 2011
Location: Atlit, Israel
Posts: 916
@clsid
The QS check doesn't work on non-SandyBridge machines. It fails on my 9400T (Penryn) which is capable of QS. I'm afraid it will also fail on IvyBridge.
Please remove the existing check untill a new one is found.
If it makes sense to you, I can export a check function from my dll.

If these checks are too complicated or time consuming (from you), please remove them altogether or at least make them pass on all Intel processors.
__________________
Eric Gur,
Processor Application Engineer for Overclocking and CPU technologies
Intel QuickSync Decoder author
Intel Corp.
egur is offline   Reply With Quote
Old 4th February 2012, 16:46   #14651  |  Link
clsid
*****
 
Join Date: Feb 2005
Posts: 5,640
Quote:
Originally Posted by egur View Post
@clsid
The QS check doesn't work on non-SandyBridge machines. It fails on my 9400T (Penryn) which is capable of QS. I'm afraid it will also fail on IvyBridge.
Please remove the existing check untill a new one is found.
If it makes sense to you, I can export a check function from my dll.

If these checks are too complicated or time consuming (from you), please remove them altogether or at least make them pass on all Intel processors.
A simple check function in the DLL would be fine. Let it return for example a DWORD with a bitmask of capabilities.
__________________
MPC-HC 2.1.7.2
clsid is offline   Reply With Quote
Old 4th February 2012, 16:55   #14652  |  Link
ipanema
Registered User
 
Join Date: Apr 2009
Posts: 93
Quote:
Originally Posted by haruhiko_yamagata View Post
Thanks for the sample.

Is it "MPEG-2 Demultiplexer" or "MPEG-2 Splitter"?
What is the OS?
"MPEG-2 Demux" isn't installed on my PC.
Sorry, it's the "MPEG-2 Demultiplexer" http://msdn.microsoft.com/en-us/libr...=vs.85%29.aspx

OS is Windows 7. I haven't checked other versions of Windows.
ipanema is offline   Reply With Quote
Old 4th February 2012, 17:37   #14653  |  Link
khagaroth
Registered User
 
khagaroth's Avatar
 
Join Date: Feb 2006
Posts: 103
Updated Czech translation.
khagaroth is offline   Reply With Quote
Old 4th February 2012, 20:43   #14654  |  Link
egur
QuickSync Decoder author
 
Join Date: Apr 2011
Location: Atlit, Israel
Posts: 916
Quote:
Originally Posted by clsid View Post
A simple check function in the DLL would be fine. Let it return for example a DWORD with a bitmask of capabilities.
OK, I'll add a simple check with a DWORD bitmask. First version will have 2 properties - HW support and SW support (Media SDK installed).

Update:
Done at r4301.
__________________
Eric Gur,
Processor Application Engineer for Overclocking and CPU technologies
Intel QuickSync Decoder author
Intel Corp.

Last edited by egur; 4th February 2012 at 22:00.
egur is offline   Reply With Quote
Old 5th February 2012, 01:26   #14655  |  Link
haruhiko_yamagata
Registered User
 
Join Date: Feb 2006
Location: Japan
Posts: 1,560
Quote:
Originally Posted by ipanema View Post
Sorry, it's the "MPEG-2 Demultiplexer" http://msdn.microsoft.com/en-us/libr...=vs.85%29.aspx

OS is Windows 7. I haven't checked other versions of Windows.
I cannot use "MPEG-2 Demultiplexer". There is the filter but it requires a source filter.
I tried "TS File Source". It connects but the screen is black whatever the decoder is.
Any other source filters I have do not connect.
How do you use the filter? Would you tell me the application and the source filter?

EDIT
I gave up MPEG-2 Demultiplexer. I tried DVBViewer and reproduced the problem.
__________________
[ Download ffdshow | Wiki ]

Last edited by haruhiko_yamagata; 5th February 2012 at 03:58.
haruhiko_yamagata is offline   Reply With Quote
Old 5th February 2012, 09:31   #14656  |  Link
Liisachan
李姗倩 Lǐ Shān Qiàn
 
Liisachan's Avatar
 
Join Date: Nov 2002
Posts: 1,340
@haruhiko_yamagata
About r4303: if it's not too much trouble could you please make this optional? Maybe "Frame timestamps" (original one) and "Frame timestamps (friendly)" (new format), or perhaps "Frame timestamps" (new format) and "Raw frame timestamps" (old format) in the "OSD" menu? I sometimes want to see the frame time stamp down to rt (or even ns), and the old feature in ffdshow was quite handy.
Liisachan is offline   Reply With Quote
Old 5th February 2012, 10:26   #14657  |  Link
haruhiko_yamagata
Registered User
 
Join Date: Feb 2006
Location: Japan
Posts: 1,560
Quote:
Originally Posted by Liisachan View Post
@haruhiko_yamagata
About r4303: if it's not too much trouble could you please make this optional? Maybe "Frame timestamps" (original one) and "Frame timestamps (friendly)" (new format), or perhaps "Frame timestamps" (new format) and "Raw frame timestamps" (old format) in the "OSD" menu? I sometimes want to see the frame time stamp down to rt (or even ns), and the old feature in ffdshow was quite handy.
OK, then I'll create a new item.
__________________
[ Download ffdshow | Wiki ]
haruhiko_yamagata is offline   Reply With Quote
Old 5th February 2012, 12:16   #14658  |  Link
ipanema
Registered User
 
Join Date: Apr 2009
Posts: 93
MPEG-2 demultiplexer can be a little tricky to set up. The output media type must be set manually. It's part of a test environment here which I think has its own source filter (I didn't write it).

Quote:
I tried DVBViewer and reproduced the problem.
Great.

Last edited by ipanema; 5th February 2012 at 12:18.
ipanema is offline   Reply With Quote
Old 5th February 2012, 13:08   #14659  |  Link
haruhiko_yamagata
Registered User
 
Join Date: Feb 2006
Location: Japan
Posts: 1,560
@ipanema,
It has been fixed at rev 4302. Please try the latest build.
__________________
[ Download ffdshow | Wiki ]
haruhiko_yamagata is offline   Reply With Quote
Old 6th February 2012, 12:56   #14660  |  Link
haruhiko_yamagata
Registered User
 
Join Date: Feb 2006
Location: Japan
Posts: 1,560
libswscale: migrate from FFmpeg to Libav

To fix the regression of performance of resize filter with Accurate rounding checked, I changed the base of ffdshow's libswscale from FFmpeg to Libav.
The performance is OK with Libav's libswscale plus ffdshow's custom multithreading.
As we are preparing for the next release, this kind of big change should be avoided if possible. But because it was a regression, I couldn't help...
Please test new builds.
x86
x64
patch

EDIT
The patch reverts rev 4156.
Accurate rounding is enabled by default. The new builds and current SVN builds use a different registry value for "Accurate rounding". So please be careful if you compare performance or quality.
__________________
[ Download ffdshow | Wiki ]

Last edited by haruhiko_yamagata; 6th February 2012 at 13:15.
haruhiko_yamagata is offline   Reply With Quote
Reply

Tags
ffdshow, ffdshow tryouts, ffdshow-mt, ffplay, icl

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 03:22.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.