Nic's xvid build 2004.12.14

[Sharktooth]

Well, it has been a while hasn't it?
New XviD build up, nothing special. Intel 8.1 Compiled.
(Has been over a year afterall... ;) )

Download it from http://nic.dnsalias.com/

Are there more info about it?

[Koepi]

Yupp. Straight CVS (so some fancy things missing like VBV, selectable curve compression and [don't know for sure] _maybe_ dshow auto AR correction). Fastest build possible for P4. Will be slower on AMD machines though :-/ (ICL8.1 is evil).

Finally, Nic is back to offer builds, too! :) It's always better to have a choice.

Regards
Koepi

[Sharktooth]

Well, i can compile some builds too (GCC compiles) but i'm sure nic's and your builds are faster and there is already ppl (like the "binarymaker" celticdruid :D ) that offer compiles with gcc.

[Caleb666]

Originally posted by Koepi
Yupp. Straight CVS (so some fancy things missing like VBV, selectable curve compression and [don't know for sure] _maybe_ dshow auto AR correction). Fastest build possible for P4. Will be slower on AMD machines though :-/ (ICL8.1 is evil).

Finally, Nic is back to offer builds, too! :) It's always better to have a choice.

Regards
Koepi

Hey Koepi,

Are there going to be any improvements in seeking speed in upcoming XVID decoders?

DIVX 5's seek speed is really really fast and XVID is sometimes very slow.

Also, most talk is done about the encoder.. but are there any talks about the xvid decoder ? as in speed improvements or any others?

For example, a few notes from the latest DIVX alpha:
CoreBurn includes numerous fixes to the encoder GUI and functionality. Also bundled with this release is a newer version of the DivX DirectShow decoder, which includes performance and quality enhancements.

[Bogalvator]

I assume this build doesn't contain skal's fix for the trellis bug suggested here
http://forum.doom9.org/showthread.php?s=&threadid=84999&perpage=20&pagenumber=2
as it has not been applied in the CVS (I notice it was last suggested on 10/12/04 on the mailing list)

Is the fix not safe?

[Nic]

(I was wondering when someone would notice :) )

I'm going to try and update lots on my site (here's hoping I'll have time).

Take Care,
-Nic

[Leak]

Originally posted by Nic
I'm going to try and update lots on my site (here's hoping I'll have time).

One word: *WHEEEE*! :)

np: Autechre - Draun Quarter (Envane)

[lithoc]

I'm not a programmer but something strike me.

I'm sure AMD fans would love to have.


http://groups.google.ca/groups?dq=&hl=en&lr=&ie=UTF-8&oe=UTF-8&threadm=a13e403a.0402091438.14018f5a%40posting.google.com&rnum=1&prev=/groups%3Fdq%3D%26hl%3Den%26lr%3D%26ie%3DUTF-8%26oe%3DUTF-8%26group%3Dcomp.arch%26selm%3Da13e403a.0402091438.14018f5a%2540posting.google.com

Anyone try it yet? What's the % improvement for XVID?


:devil:

[Nic]

@lithoc:
Interesting post. The AMD FX chip can handle most of the intel optimisations (Can't it? I thought, the only one I think it doesn't is SSE3). So the intel compilers routine must be checking for an Intel Chip as well as Processer Capabilities. A better hack would remove the Intel Chip detection but leave the Processor Capabilities check in. :)

-Nic

[lithoc]

As far as I know, ICL 8.1 crippled Athlon64 by disabled SSE2

[bond]

Originally posted by Koepi
some fancy things missing like VBVhopefully it will be stable soon :)

[SeeMoreDigital]

Nice one!

More "profiles" I see...

http://img145.exs.cx/img145/5337/nicsxvidbuild4ft.gif


Cheers

[celtic_druid]

Just tried it with /QaxN, it found and removed the checks and the program does not crash despite me not having SSE2.

[dragongodz]

the real test would need to be on an Athlon64 or AthlonFX of course. to see whether its speed changed or if it still used the generic code(generated by the "a" in /QaxN).
since some of xvid is compiled with nasm these parts dont recieve any benefit from this at all.

[celtic_druid]

Well I just compiled xvidcore.dll with QaxN and QxN. In both cases the patched dll's were identical to the originals.

The previous test I did was with ffdshow.ax which I use ICL8.1 for.

[SeeMoreDigital]

Sadly not even Nic's build will not allow me to generate XviD Mpeg4 B-VOP encodes that will play with my Sigma Xcard :(

Not that I was really expecting it too!


Cheers

[sysKin]

Yay :)

Originally posted by Koepi
Yupp. Straight CVS (so some fancy things missing like VBV, selectable curve compression and [don't know for sure] _maybe_ dshow auto AR correction).
VBV is there, you just need to use corresponding profile/level and you'll get VBV active.
"Unrestricted" is the default again, just to disable VBV by default.

Radek

[BoNz1]

sysKin, is there anyway that you can make these VBV boxes in the Levels tab editable? This would be ideal so that if you wanted to encode with say unrestricted profile using quarter pixel or whatever you could still use the VBV for compatibility for your standalone (if it supported qpel). And you wouldn't be forced to use the DXN HT profile which won't allow you to do turn on things like qpel and gmc. This would be really handy I think.

[celtic_druid]

I could do a build with say a new MTK profile. GMC disabled, packed bitstream off by default, VBV, etc.

GMC and VBV? Are there any chipsets that support 3 warp point GMC?

[SeeMoreDigital]

Originally posted by celtic_druid
GMC and VBV? Are there any chipsets that support 3 warp point GMC? It seems not even the new NeroDigital stand-alone supports 3WP GMC :eek:


Cheers

[celtic_druid]

Did a new profile for MTK based players. Enables qpel and unticks packed bitstream if max bframes is greater than 1 (checked from registry). Otherwise it is the same as DXN HT.

@BoNz1, making VBV editible is real easy, however (from what I can tell) the values aren't stored in the registry, just grabbed from the profile, so editing it wouldn't do you much good.

[BoNz1]

Originally posted by celtic_druid

@BoNz1, making VBV editible is real easy, however (from what I can tell) the values aren't stored in the registry, just grabbed from the profile, so editing it wouldn't do you much good.

Yup, I know how to do that. But you have to do that at build time. Usually, I just edit the unrestricted profile to my liking.

[bond]

does anyone know the vbv settings useful for the different chipsets? who knows whether the dxn vbv settings together with qpel aso are manageable, aso...?
i would just stick to the official mpeg-4 profiles and the dxn ones (and maybe nero, once they get officially published) to avoid confusion

[QNX]

i've ripped AVP movie with AutoGK 1.83 and 2 builds - Koepi's and Nic's

with same settings:

Nic's build - 1 hour and 40 minutes
Koepi's build - 1 hour 49 minutes

both were the last 1.1 builds avaliable


:D

[el divx]

Originally posted by Nic
The AMD FX chip can handle most of the intel optimisations (Can't it? I thought, the only one I think it doesn't is SSE3).

As the guys from AMD said, there is going to be support for the SSE3 instruction set in the revisions of the Athlon 64 core that will come out in 2005.

[sysKin]

Originally posted by QNX
i've ripped AVP movie with AutoGK 1.83 and 2 builds - Koepi's and Nic's

with same settings:

Nic's build - 1 hour and 40 minutes
Koepi's build - 1 hour 49 minutes

both were the last 1.1 builds avaliable
What do you mean, last 1.1 available - I've been adding speedups like crazy recently, there is no way for relatively old Koepi's build to be as fast as much newer Nic's build, and then again they can't possibly be as fast as today's build :)

Radek

[chilledoutuk]

Does anyone know where I can find out what changes have been made to 1.1 cvs. I for the life of me cant find anything on the main site.

Do I need to subscribe to a mailing list?

I used to use gamr's site for this but thats down now.

cheers

[celtic_druid]

I have been keeping a log recently of changes:
http://celticdruid.no-ip.com/phpBB2/viewtopic.php?t=4

Also you can read the mailing list archive without registering if you want.

Well I guess for instance the Latest speed up with bvop-vhq, wouldn't make any difference if they didn't run the test with bvop-vhq.

[SeeMoreDigital]

Originally posted by celtic_druid
....Well I guess for instance the Latest speed up with bvop-vhq, wouldn't make any difference if they didn't run the test with bvop-vhq. I'm still at a total loss as to why my Xcard can't play Xvid encodes with one (or more) B-VOP.

I'm curious, when packed bit-stream and Closed GOV is taken out of the equation... How many different B-VOP implementations are there?


Cheers

[QNX]

Originally posted by sysKin
What do you mean, last 1.1 available

XviD-1.1.-127-06112004.exe from koepi
and this Nic's build

last binaries avaliable :P

if you know newer - tell me :rolleyes:

[BigDid]

Originally posted by QNX

if you know newer - tell me :rolleyes:
Maybe this one from Celtic_Druid (04/12/18):

http://celticdruid.no-ip.com/xvid/XviD.cvs.head.exe :)

Did

[dragongodz]

Well I just compiled xvidcore.dll with QaxN and QxN. In both cases the patched dll's were identical to the originals.
celtic_druid - and you use IC 8.1 right ? the original article was for changing IC 8.0 compiled files so maybe intel changed something fo IC 8.1 so the old method doesnt work.

- I've been adding speedups like crazy recently, there is no way for relatively old Koepi's build to be as fast as much newer Nic's build, and then again they can't possibly be as fast as today's build
sysKin - not exactly true. they can encode at roughly the same speed if the input (whether decoding or filter of etc) is a limiting factor. in that case they are not running to the full speed but being made to wait. :)

[celtic_druid]

That is true. Now I guess I could reinstall ICL8, but I really couldn't be bothered. It definatly did something to an ICL8.1 compile of ffdshow though.

[Sharktooth]

ICL8.1 is EVIL for AMD users...
Nic's build will be slower than Koepi's for my Athlon64.

[BigDid]

Originally posted by celtic_druid
Did a new profile for MTK based players. Enables qpel and unticks packed bitstream if max bframes is greater than 1 (checked from registry). Otherwise it is the same as DXN HT ...
Hello,

Apologies if dumb questions or formulating: :)
-Is the MTK profile available and selectable like the DXN HT PAL ? because I can't find it in your CVS. :confused:
-If not, will you make one ? and in that case will the user be able to choose a custom matrix?, because it seems people (not me I use AGK atm) can not select custom matrix in HT PAL.

Thanks

Did

[celtic_druid]

I never posted the link. Main changes over the DivX HT profile are, qpel is enabled, as are custom matrices, also if it detects more than 1 bframe then it turns off packed bitstream. Think that was about it and it can be found here: http://celticdruid.no-ip.com/test/mtkxvidvfw.7z

Probably DivX's bitrate/VBV settings could be raised a bit though, but I left them as is.

[BigDid]

Originally posted by celtic_druid
I never posted the link. Main changes over the DivX HT profile are, qpel is enabled, as are custom matrices, also if it detects more than 1 bframe then it turns off packed bitstream. Think that was about it and it can be found here: http://celticdruid.no-ip.com/test/mtkxvidvfw.7z

Probably DivX's bitrate/VBV settings could be raised a bit though, but I left them as is.

Thank you Celtic,

I will post the link in unite-video, french forum where Kurtnoise & other(s) doom9 french speaking members exchange. :)
I will follow up feedback if any (145 hits on the mtk profile subject in 3 days).
I would also like the bitrate raised a bit ( my SAP H&B 3220 accepts 6000kbps) but others told me that my SAP may be a good exception and that 4000kbps was safer ... maybe 5000 ? :D

Did

[bond]

i have two questions:

1) whats the status of reduced resolution? i saw that it has been removed for 1.1. isnt it stable, compliant...?

2) talking about the divxnetworks profiles:
first of all great that xvid is able to set them!
second: shouldnt the max number of consecutive b-frames also be hardset to 1 and packed bitstream not "disableable" (as divx5 doesnt allow more than 1 b-frames with htp set)
also shouldnt the fourcc be set to DIVX in that case?

[sysKin]

Originally posted by bond
i have two questions:

1) whats the status of reduced resolution? i saw that it has been removed for 1.1. isnt it stable, compliant...?
It wasn't working, it wasn't useful (now nor in the future), and it was bloating the code everywhere.
So, it was removed ~1 month ago. Removed for good, I hope.

2) talking about the divxnetworks profiles:
first of all great that xvid is able to set them!
second: shouldnt the max number of consecutive b-frames also be hardset to 1 and packed bitstream not "disableable" (as divx5 doesnt allow more than 1 b-frames with htp set)
also shouldnt the fourcc be set to DIVX in that case?
Hmm perhaps. The profiles were programmed into VfW long before DivX 5.1 so basically they don't have these capabilities *yet*, because noone revised this code, not because we don't want to.

Radek

[bond]

thx :)

[dream_falcon]

My benchmark with a small videoclip,
P4 Prescott 3.0

Nic's 14-12-04
98.554
45.951

Koepi´1.0.2
99.567
38.687

a good increase of speed in the 2n pass :cool:

[celtic_druid]

You are comparing a 1.0.x branch build to a cvs head branch build though, as well as the different compilers.

[dragongodz]

celtic_druid and anyone else interested in the SSE2 on Athlon64 with Intel compiler etc that was discussed earlier can find some more information here
http://forum.doom9.org/showthread.php?s=&threadid=87299