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 1st October 2007, 01:05   #21  |  Link
Terranigma
*Space Reserved*
 
Terranigma's Avatar
 
Join Date: May 2006
Posts: 953
Quote:
Originally Posted by martino View Post
I'd also like to know as to which patches are present in Cef's build. I don't mean the "dark" or "exp" version. It's just confusing, since there seem to be a few builds, more patches, and to me it looks like a hell of mess where trying to find an answer is rather hard...

Thanks
I guess you didn't see This ?
Cef, you think you could update the exp build for now and include AQ, Thread Pool, and the new ME-Prepass patch, then for future references, include me-prepass in your regular builds?
Terranigma is offline   Reply With Quote
Old 1st October 2007, 01:12   #22  |  Link
Dark Shikari
x264 developer
 
Dark Shikari's Avatar
 
Join Date: Sep 2005
Posts: 8,690
Also add the --subme 7 patch, since its been proven quite thoroughly to increase quality in basically all cases at minimal speed cost.

Faster DIA, IMH, and SATD shouldn't be applied yet. One thought I did have was to instead of making SATD an option for all ME search methods, instead add a 5th ME search method:

DIA
HEX
UMH
ESA
HES (Hadamard Exhaustive Search: Better than all the above methods, but correspondingly slower)

The reason for this is simply that Aku's testing showed that SATD slowed down all the other methods so much that it was better to use SAD ESA than SATD anything else. However, SATD ESA is so heavily optimized that its still useful, and not too much slower than regular ESA.
Dark Shikari is offline   Reply With Quote
Old 1st October 2007, 01:18   #23  |  Link
Terranigma
*Space Reserved*
 
Terranigma's Avatar
 
Join Date: May 2006
Posts: 953
Quote:
Originally Posted by Dark Shikari View Post
Also add the --subme 7 patch, since its been proven quite thoroughly to increase quality in basically all cases at minimal speed cost.
Oh I thought the new subme-7 made it to the svn, seems I was wrong

Quote:
Originally Posted by Dark Shikari View Post
Faster DIA, IMH, and SATD shouldn't be applied yet.
I agree with you on imh and dia (At first I was all for imh, but suddenly I changed my mind ).
Quote:
Originally Posted by Dark Shikari View Post
One thought I did have was to instead of making SATD an option for all ME search methods, instead add a 5th ME search method:

DIA
HEX
UMH
ESA
HES (Hadamard Exhaustive Search: Better than all the above methods, but correspondingly slower)

The reason for this is simply that Aku's testing showed that SATD slowed down all the other methods so much that it was better to use SAD ESA than SATD anything else. However, SATD ESA is so heavily optimized that its still useful, and not too much slower than regular ESA.

I like this idea. Only allowing SATD to be used with the motion search algorithm it gains any real benefit from. I'm all for the new --hes
Terranigma is offline   Reply With Quote
Old 1st October 2007, 02:07   #24  |  Link
akupenguin
x264 developer
 
akupenguin's Avatar
 
Join Date: Sep 2004
Posts: 2,393
Not sure I like HES, too similar to HEX. Maybe TES (same "T"ransform as in SATD). Or ESH.
But I'm not sure that SATD is only useful in ESA: There are good reasons that integral-based successive elimination for SAD can only be efficient in ESA, but it's possible that SAD-based successive elimination for SATD could work in other search patterns. The cost of a SAD or a SATD is high enough that the overhead of random access needn't be fatal.

Last edited by akupenguin; 1st October 2007 at 02:16.
akupenguin is offline   Reply With Quote
Old 1st October 2007, 02:14   #25  |  Link
Dark Shikari
x264 developer
 
Dark Shikari's Avatar
 
Join Date: Sep 2005
Posts: 8,690
Quote:
Originally Posted by akupenguin View Post
Not sure I like HES, too similar to HEX. Maybe TES (same "T"ransform as in SATD). Or ESH.
TES seems fine to me; it avoids starting with the same letter as any of the others.
Dark Shikari is offline   Reply With Quote
Old 1st October 2007, 02:42   #26  |  Link
fields_g
x264... Brilliant!
 
Join Date: Mar 2005
Location: Rockville, MD
Posts: 167
Quote:
Originally Posted by Dark Shikari View Post
However, SATD ESA is so heavily optimized that its still useful, and not too much slower than regular ESA.
Here's another solution... set regular ESA to SATD ESA?

Let SAD IMH be the (speed) middle ground between SAD UMH and SATD ESA?

akupenguin, could you update your chart here with Dark Shikari's new ME-prepass found here?

Depending on the results, I might even suggest regular ESA to be prepass SATD ESA.

I see it like this... If you are mad enough to do ESA, you are quite likely going to do SATD and prepass also.
fields_g is offline   Reply With Quote
Old 1st October 2007, 02:43   #27  |  Link
Dark Shikari
x264 developer
 
Dark Shikari's Avatar
 
Join Date: Sep 2005
Posts: 8,690
Quote:
Originally Posted by fields_g View Post
Here's another solution... set regular ESA to SATD ESA?

Let SAD IMH be the (speed) middle ground between SAD UMH and SATD ESA?

akupenguin, could you update your chart here with Dark Shikari's new ME-prepass found here?

Depending on the results, I might even suggest regular ESA to be prepass SATD ESA.

I see it like this... If you are mad enough to do ESA, you are quite likely going to do SATD and prepass also.
Shouldn't force users to use what they don't want to--better to give them the option.
Dark Shikari is offline   Reply With Quote
Old 1st October 2007, 02:51   #28  |  Link
fields_g
x264... Brilliant!
 
Join Date: Mar 2005
Location: Rockville, MD
Posts: 167
Quote:
Originally Posted by Dark Shikari View Post
Shouldn't force users to use what they don't want to--better to give them the option.
I agree.. I am just looking for some other option to making another me-type that really isn't anything but a preset. Make the defaults best practices and allow switches to deviate.

How about normal default esa be SATD and with a switch it can SAD and ditch the new me-type. All choices still remain.
fields_g is offline   Reply With Quote
Old 1st October 2007, 02:54   #29  |  Link
Dark Shikari
x264 developer
 
Dark Shikari's Avatar
 
Join Date: Sep 2005
Posts: 8,690
Quote:
Originally Posted by fields_g View Post
I agree.. I am just looking for some other option to making another me-type that really isn't anything but a preset. Make the defaults best practices and allow switches to deviate.

How about normal default esa be SATD and with a switch it can SAD and ditch the new me-type. All choices still remain.
How is using --fpel-cmp satd on ESA a "best practice"? Its even slower than regular ESA, and so only useful for those who have even more time to waste.

Also note that if TES/whatever its called uses SATD, the --fpel-cmp satd option will be removed.
Dark Shikari is offline   Reply With Quote
Old 1st October 2007, 03:24   #30  |  Link
akupenguin
x264 developer
 
akupenguin's Avatar
 
Join Date: Sep 2004
Posts: 2,393
Quote:
Originally Posted by fields_g View Post
How about normal default esa be SATD and with a switch it can SAD and ditch the new me-type. All choices still remain.
A new value for --me is simpler interface-wise than a new top-level option.
akupenguin is offline   Reply With Quote
Old 1st October 2007, 03:32   #31  |  Link
fields_g
x264... Brilliant!
 
Join Date: Mar 2005
Location: Rockville, MD
Posts: 167
Quote:
Originally Posted by Dark Shikari View Post
How is using --fpel-cmp satd on ESA a "best practice"? Its even slower than regular ESA, and so only useful for those who have even more time to waste.

Also note that if TES/whatever its called uses SATD, the --fpel-cmp satd option will be removed.
Using the chart as an approximation roughly:
SAD - UMH - ME32
has the same quality as
SAD - ESA - ME7
and
SATD - ESA - ME4

However the FPS is 47 vs. 42 vs. 39 respectively. SATD is slowest, by not by too much. Even though SATD is 7-8% slower, the magic is that SATD, with computation (me-range increments), SATD gains quality much quicker and peaks much higher. Additionally, SATD ESA-me6 beats the quality of SAD ESA-me12 at the same FPS! Therefore, SAD ESA only has a place for me range less than 12.

So instead of telling people that ESA is only has benefits from me-7 through me-12, over other ME-types, you could tell them ESA picks up quality-wise where UMH stops. It seems a little more clean to me. I just hope the explanation is understandable.

Maybe I'm a little my willing to throw computation at it than others, but I think the average person ESA would usually do this anyway.
fields_g is offline   Reply With Quote
Old 1st October 2007, 03:44   #32  |  Link
Dark Shikari
x264 developer
 
Dark Shikari's Avatar
 
Join Date: Sep 2005
Posts: 8,690
Quote:
Originally Posted by fields_g View Post
Using the chart as an approximation roughly:
SAD - UMH - ME32
has the same quality as
SAD - ESA - ME7
and
SATD - ESA - ME4

However the FPS is 47 vs. 42 vs. 39 respectively. SATD is slowest, by not by too much. Even though SATD is 7-8% slower, the magic is that SATD, with computation (me-range increments), SATD gains quality much quicker and peaks much higher. Additionally, SATD ESA-me6 beats the quality of SAD ESA-me12 at the same FPS! Therefore, SAD ESA only has a place for me range less than 12.

So instead of telling people that ESA is only has benefits from me-7 through me-12, over other ME-types, you could tell them ESA picks up quality-wise where UMH stops. It seems a little more clean to me. I just hope the explanation is understandable.

Maybe I'm a little my willing to throw computation at it than others, but I think the average person ESA would usually do this anyway.
Except that on some sources, SATD is inferior to SAD as a metric

Anime in particular seems to suffer from this, in my experience.
Dark Shikari is offline   Reply With Quote
Old 1st October 2007, 03:55   #33  |  Link
fields_g
x264... Brilliant!
 
Join Date: Mar 2005
Location: Rockville, MD
Posts: 167
Quote:
Originally Posted by Dark Shikari View Post
Except that on some sources, SATD is inferior to SAD as a metric

Anime in particular seems to suffer from this, in my experience.
I was just about to state that I might be overusing this single chart (source) a bit! I'll be downloading one of these fancy "bundle-o-patches" builds and start going at it! Quite honestly, I'll be able to follow either scheme and will be happy as long as I have prepass SATD ESA around in some form, especially with the improvements listed here.

Last edited by fields_g; 1st October 2007 at 04:08.
fields_g is offline   Reply With Quote
Old 1st October 2007, 11:00   #34  |  Link
Cef
Registered User
 
Join Date: Mar 2007
Posts: 25
Quote:
Originally Posted by martino View Post
I'd also like to know as to which patches are present in Cef's build. I don't mean the "dark" or "exp" version. It's just confusing, since there seem to be a few builds, more patches, and to me it looks like a hell of mess where trying to find an answer is rather hard...

Thanks
As I already said, my builds have AQ and thread pool applied. x264_xxx_dark was including all Dark_shikari's patches at the time it was posted (except faster first pass iirc), and x264_xxx_exp was a build requested by Sagittaire with some Dark's patches and hrd.

I completly agree this is confusing, my organization is terrible on this, but I don't have much time to dedicate to it, and I usually spend it fixing conflicts with new rev's or between patches. If you have any suggestion it's welcome.
Cef is offline   Reply With Quote
Old 1st October 2007, 16:57   #35  |  Link
martino
masktools2 (ab)user
 
martino's Avatar
 
Join Date: Oct 2006
Location: PAL-I :(
Posts: 235
Quote:
Originally Posted by Terranigma View Post
I guess you didn't see This ?
I did in fact, and heck. At this point there are two AQ patches. ;_;

But I can say at this point that it's the "old" one. And thanks Cef for explaining.

I'm not sure whether I'd have any good suggestions, but perhaps just a small txt in the directory where your builds are located (on x264.nl) which would state which patch(es) was/were applied to which build(s). Or maybe if morph would be so kind to interpret this into the introductory post in this thread... Whatever works really.
martino is offline   Reply With Quote
Old 1st October 2007, 17:46   #36  |  Link
Terranigma
*Space Reserved*
 
Terranigma's Avatar
 
Join Date: May 2006
Posts: 953
martino, do you know how to complile x264? It doesn't look like noone's too eager to compile an experimental build with aq, thread pool, new subme7, new pre-pass, and keep --fpel-cmp sad/satd like it is as suggested by akupenguin, and if you must implement the new aq, add it as an optional command. Maybe something like aq2-strength. You can find the latest aq2 algortihm by Dark Shikari here. I'm not sure if that's the latest, so he's the only one that can confirm or deny this.
Terranigma is offline   Reply With Quote
Old 1st October 2007, 18:14   #37  |  Link
Dark Shikari
x264 developer
 
Dark Shikari's Avatar
 
Join Date: Sep 2005
Posts: 8,690
Quote:
Originally Posted by Terranigma View Post
martino, do you know how to complile x264? It doesn't look like noone's too eager to compile an experimental build with aq, thread pool, new subme7, new pre-pass, and keep --fpel-cmp sad/satd like it is as suggested by akupenguin, and if you must implement the new aq, add it as an optional command. Maybe something like aq2-strength. You can find the latest aq2 algortihm by Dark Shikari here. I'm not sure if that's the latest, so he's the only one that can confirm or deny this.
New AQ is definitely not ready, and that is quite old IIRC.

Don't add it yet. Its way too experimental.
Dark Shikari is offline   Reply With Quote
Old 2nd October 2007, 00:10   #38  |  Link
DeathTheSheep
<The VFW Sheep of Death>
 
DeathTheSheep's Avatar
 
Join Date: Dec 2004
Location: Deathly pasture of VFW
Posts: 1,149
I think the reason why most people don't have a go at compiling these patches is that the patches themselves are quite troublesome to apply.

I did it with a lot of manual patching, so it's definitely possible. But of course I also messed with a lot of other stuff in the code and then finally deleted the folder. I think it's best at this point to wait it out until more stability/commits/developments occur. Else just use the older build Cef (?) made, there shouldn't be much difference.
__________________
Recommended all-in-one stop for x264/GCC needs on Windows: Komisar x264 builds!
DeathTheSheep is offline   Reply With Quote
Old 2nd October 2007, 00:35   #39  |  Link
Terranigma
*Space Reserved*
 
Terranigma's Avatar
 
Join Date: May 2006
Posts: 953
Quote:
Originally Posted by DeathTheSheep View Post
I think it's best at this point to wait it out until more stability/commits/developments occur. Else just use the older build Cef (?) made, there shouldn't be much difference.
Quote:
Originally Posted by Dark Shikari View Post
Speed: 25% faster (25% less impact on speed as compared to the old ME-prepass)
Quality: 42% better (42% more increase in quality as compared to the old ME-prepass)

Not surprisingly, eliminating the qpel aspect of the search gave a huge speed boost with an actual slight increase in quality.
42% is a huge difference, or so I would think

Last edited by Terranigma; 2nd October 2007 at 00:38.
Terranigma is offline   Reply With Quote
Old 2nd October 2007, 00:56   #40  |  Link
Dark Shikari
x264 developer
 
Dark Shikari's Avatar
 
Join Date: Sep 2005
Posts: 8,690
Quote:
Originally Posted by Terranigma View Post
42% is a huge difference, or so I would think
Let's say the original gave a 2% quality boost.

42% better quality over the original ME Prepass = 2.84% quality boost.
Dark Shikari 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 15:25.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2017, vBulletin Solutions Inc.