Cvs 2002-03-16

[-h]

I finally got around to merging a new VfW frontend. This one has a main dialog, with the extra options hidden behind an "Advanced options" button. There's tooltips (albeit crappy), internal changes (uh oh) and more enabling/disabling of controls.

I've uploaded the .dll here (http://briefcase.yahoo.com/hstink), though I suspect others will upload soon enough :)

A lot of things changed, so please report any bugs. Of *GREAT* assistance would be if, upon finding a bug, you could reinstall the previous version of XviD and see if it still occurs. This will let us know whether the issue is with the core, or the vfw update.

Anyone with ideas for tooltip text (or perhaps a non-english translation?) is welcome to contribute. I tried implementing "normal" popup tooltips, but things went very wrong and I did it the lazy way instead.

-h

[Koepi]

I'll take the time tomorrow to play around with it and create a new installer, but i guess my priority will be the help file/XOE.

It's very likely that the new vfw breaks the DSF, so I'm looking into that as well...

Regards,
Koepi

[mac1929]

If you ever want someone to translate anything to spanish, mail me to mac at theoffice.net I will be glad to do it.

[The People's Elbow]

Nice work -h! Looks more "intelligent" this way and the tooltips make it easier for unexperienced users!

[Nic]

Can now get it at:

http://nic.dnsalias.com

(oohh, doesn't it look fancy :) - thanks to doggie)

The new DShow filter should be the best yet,
if you have problems installing the filter then copy the msvcr70.dll to your system directory & then re-install
(ive tested it on Win98SE and it worked ok)
(you can make sure you've got the latest version installed by looking at the properties...the date is in there)

Thanks to Tom, what used to take up 100% of my CPU now takes up around 50%-70% :D

Cheers,
-Nic

ps
Please post all problems in this thread.

pps
There is a delay in the encoding & decoding (playback) with this core...dont know why???

[lar1r]

Good job!
I like it because I have no need to leave the default panel now.

Great to have Luminance Masking auto disabled on first pass and enabled on second without having to remember to switch it.

[-h]

There is a delay in the encoding & decoding (playback) with this core...dont know why???

Agh that damn delay! It occurs "somewhere" in xvid_init() in core's xvid.c. I placed a bunch of debug outputs to trap which init() function was causing the delay, compiled and linked to vfw.... and the delay was gone!

Very frustrating. I was hoping it was a one-off on my end, but it seems to be happening everywhere (except, now, for me).

I'll look again.

-h

[Nic]

Ill look into it later today (its weird...)
Also my DShow filter calls xvid_init twice (!) so its a bit of a problem for me....-h if you workout what did it...shout me :)

Cheers,
-Nic

[-h]

I tried to get the delay to come back, but it's gone for good on my end (fresh cvs checkout and all).

Can't explain it :)

Edit: I did merge a couple mods into core perhaps after you checked out (turned off debug output for release compiles, and switched on automatic inlining in core.dsp). Though I doubt that would have fixed this issue... should run 0.7% faster though (woohooo).

-h

[Nic]

Ill go re-check it out (3rd time today!!!) & try again :)

Cheers for the info,
-Nic

[Isibaar]

Hi all!

nice forum you have here...

About the delay: I guess it's caused by init_timer(), so make sure that you don't have _PROFILING_ defined. (note to myself: move init_timer into decoder.c/encoder.c).

@-h:
I want my debug output back ;)

[Koepi]

I'll release a new binary soon, it'll contain an unrolled sad_xmm from -h... thanks -h! :)
Should result in a speed gain about 3%, some other compiling options nearly another 1%,...

I'll post again when I uploaded the update!

Btw., might be one of my last releases since now i feel most of the GUI mods i thought they are necessary are incorporated now :)

But you never know what'll happen :)

Regards,
Koepi

[The People's Elbow]

@Nic: WOW! The new DSF is awesome compared to any older build. Respect, very good work Nic & Tom! Now it relly roxx0rs, because quality is way better then the "lousy" DivX4/5 DSF and playback is finnaly smooth enough. (@ my TB1200 and my Duron800 with Resolution 640x352 no more dropped frames and no 100% use of CPU capacity anymore.)

About the delay...on my Win ME machine it doesn't exist and on my XP it seems to be the opposite! :confused: Not that it's really disturbing, but funny in some case!

There's only one thing missing to make XviD "the emperor of video encoding"...I guess it has something to do with B-Frames! ;)

- greetz from the Elbow!

[-h]

I want my debug output back ;)

Agh there is someone reading that weird output! ;)

I thought it would be the branch mispredictions in mbcoding (-128..127 bounds checking) that led to the speedup, more than anything else. But this is what it ended up with. Fun output:


time (ms) d = debug output, i = "any suitable inlining" on

+d-i +d+i -d-i -d+i
dct 9473 9467 9379 9379
quant 3665 3661 3630 3651
idct 4877 4850 4779 4784
iquant 2799 2790 2787 2790
me 47820 47721 47535 47555
mc 15528 15532 15632 15408
interp 14425 14398 14415 14326
rgbyuv 8252 8242 8169 8196
transf 1746 1720 1725 1730
pred 891 887 879 883
mbcod 9890 9890 9737 9706

total 123970 123745 123235 123016


.. or I could just unroll sad_xmm and get a 3% speedup regardless ;)

-h

[Nic]

1) The latest CVS code doesn't have the delay bug??? Weird....already recompiled & uploaded to my site

2) Why haven't you committed that unrolled SAD code? You did it ages ago? huh? huh? :D

Cheers h,
-Nic

[xming]

I just checked out and build it, I cannot notice any delay

[Koepi]

XviD-16032002-1:
- Fresh build from CVS, added -h's unrolled sad_mmx routines for speed gain
- Added Nic's new Directshow Filter which is amazing fast now thanks to trbarry's
enhancements!
- Readded msvcr70.dll again since the DSF is compiled with VC 7 again by Nic

Follow the XviD link in my signature.

Tomorrow or in some days I'll take the time to update the help file / XOE, for now the options have just moved and you must search in those documents to find explanations ;)

Best regards,
Koepi

P.S.: I'd like to thank all involved people again for their hard and nice work!

[Nic]

Where can I get -h's sad_mmx.asm ?

-Nic

[wing1]

Unbelievable! XVID!!!!! video encoding has never been soooo goooood :D Thanks guys.

[Koepi]

@Nic,

I don't know if -h already commited it to CVS, but you should check your mail ;)

Best regards,
koepi

[Nic]

Cheers Koepi....(I knew you were going to do that...I was sitting there with my webmail open, clicking f5 everyso often....I just knew it :) )

Yours Weirdly, :D
-Nic

ps
Recompiled sources & theyre uploaded already with -h's new asm code...cheers, Koepi :)

[Koepi]

You are welcome - I was too lazy to include this as modified_sources.zip as it should show up in the CVS anyways :)

Well, have fun,

cheers,
Koepi

[wing1]

Once again the new build is just simply superb! THANK YOU VERY MUCH for making testing an exciting event.

@-h, isibaar

The codec's speed is awesome; Real time compressing during capture is great; Therefore, deinterlacing option for capture can wait until developers are ready and willing to include it :D ATM, I am quite happy with the capture quality and filesize. I luv that buffering option! JUST AMAZING...simply AMAZING.

@nic, trbarry
DSfilter is beyond word to describe it. It's definitely a masterpiece!
However, I am having a little problem with the install process using your inf file on WIN98. The DSfilter will not install correctly and will not function even if i tried to do the registration manually. I install Koepi's version and the DSfilter works. Win2K doesn't have this problem.

@Koepi,
I like the installer and the uninstaller. It's very easy and cleanup quite nicely :D It would be nice if more software developers would do that with their uninstaller: Most leave all kind of trash behind in the registry and main drive. Not all users are capable of cleaning up registry's trash :D

@all,
Can a universal/starter template for the custom quantization table be included in the next build? It is helpful to have a starting point instead of all zeros in the table to start with.

[tangent]

Holy sh!t... I used to be encoding at ~19.5fps now it's encoding at ~24.5fps with the new version!

[Nic]

What I wanna know is, whose encodes faster, mine or Koepi's build ??

(Mine is compiled in VC7 & Koepi's VC6)

It should be mine....but you never know....they might well be exactly the same :)

Cheers,
-Nic

[wing1]

@nic

Using your build I can reach 29fps max and averaging 25fps, while using Koepi's build the max fps is 26fps and averaging 23fps. For capturing Koepi's build CPU usage runs around 45-75% and maxout at 100%(briefly during major motion) and your build runs around 17-54% and maxout at 91%(briefly during major motion).

Currently, I am using two system for capture (win98). One has Koepi's build and the other has your build. I'll keep you posted :D

I also notice a slight sharpness increase with koepi's build: This is using the same settings for both system. You guys doing any further mod to the source with 1pass CBR?

[Koepi]

Nope. Since Nic and I use the same code now (except for DSF, that's still Nic's only ;) ), the difference must come from different compiler optimizations.
(There were such phenomenons in LAME, so I guess it could be possible here as well).

Well, if Nic's build is even faster, I can definatly stop here. Maybe I'll find some other features to add, maybe not.

Regards,
Koepi

[-h]

CBR encoding is working?!?! It's horribly broken on my side, everything gets quant=31..

Oh well, guess I don't have to post a report :)

I'm going to unroll some more asm and send it around.. I'd rather not put it into CVS as trbarry is playing with core's asm at the moment, and having huge wads of unrolled code suddenly appearing isn't fun. Should be able to get another 2% (more for decoding, since interpolation is called so much).

-h

[Nic]

Just make sure you send me a copy ;)

Cheers,
-Nic

ps
Ive just been out & drunk far too much fosters... :)

[trbarry]

I'm going to unroll some more asm and send it around.. I'd rather not put it into CVS as trbarry is playing with core's asm at the moment, and having huge wads of unrolled code suddenly appearing isn't fun. Should be able to get another 2% (more for decoding, since interpolation is called so much).

-h

Except for some casual browsing, all my efforts so far have been in Nic's deringing code. So don't hold off on anything on my account.

And I always do another CVS update before I start changing anything.

- Tom

[-h]

Just had a spectacular failure with hv interpolation (rounding is my enemy).

And Nic..... how could you drink it!!

-h

[wing1]

@-h

what do you mean it's broken???? It is working fine here.