Nic's latest Xvid Decoder is Slower

[Dali Lama]

Hi Nic,

I tried out your latest Xvid Decoder, but its slower than the previous one. Before I could use one deringing filter, but now, even just blocking filter is slow.

Is this because of the new compiler? Could you post your old decoder?

Thanks,

Dali

[Nic]

I dunno whats going on, the code should now be its fastest.....

Ill release a new version today, I think, from what your saying the brightness is now slowing it down (the deblocking shouldn't change between versions as its assembly which doesn't make much of a difference....)

The brightness is now controlled by good MMX, but I wonder if its the function jump slowing it down...I really don't know, im going to release a VC7 one later here, could people then test it for speed....

Thanks for all the input,
Cheers,
-Nic

[Nic]

Get new VC7 version only at:

http://nic.dnsalias.com

Please test for speed, if this doesn't work then I know its someother changes that are slowing it down. in which case ill revert back to old version.....

Cheers,
-Nic

(ps VC6 version from yesterday still available from:
http://xvid.stormpages.com )

[vxvas]

Hi nic.

just tried your new vc7 compiled filter

it is a lot faster than your vc 6 compiled 12-03

from my tests it is almost as fast as your old one vc7 build 05-03(the difference in speed is very small )



great work

[Nic]

It should be faster ????? It has MMX brightness, three less Malloc & free commands and the dering filter is slightly optimised.....

I have no idea why/how it could be slower.....bizzare (?)

Cheers for the input though,

-Nic

[primitive]

When I attempt playback with the VC7 version on my machine, it complains about not being able to find MSVCR70.dll and playback crashes. When I revert to the old version it works normally. Is this a bug in the playback filter, or is this a .dll i should have but don't?

-p

[Nic]

Doh! Forgot about that.....

Please download:
http://nic.dnsalias.com/msvcr70.zip

unzip it, then copy the file to your System directory (System32 for win2k and above, System for win98)

Then all will be well....

Cheers,
-Nic

[primitive]

It works now without issue.

-p

[wing1]

Just an observation, Win98 1st edition somehow doesn't required the missing DLL, but Win2K needs it. Furthermore, VC7 compiled DS filter seems to play much faster on the Win98 then the Win2K and the opposite is true with the VC6 version????

Win98 systems (2) have AMD 1.467Ghz (Epox w/ViaKT133 MBrd)

Win2K system has AMD 1900+XP (Shuttle w/ SiS745 MBrd)

Memory are 1Gb SDRAM PC133 for all systems.
Video card are ATI AIW128 16Mb (Win98)
Video card are ATI Rage Xpert98 + TNT32 (dual monitors).

Codecs installed for all systems are:

Huffyuv2.11
Divx3.11VKI
MPG4V3VKI
Divx4.12
DSResizer Filter
xvid (most current build .... may be different every few days :D)
ATI's codecs for video capture purpose

[Nic]

Well...In an hour or so, youll have MMX, SSE Deringing (not finished....bot Tom's doing an amazingly good job....)

....so get excited :)


Cheers,
-Nic

[rui]

"Like Homer Simpson would have said":

Hummm....Beer.... (ds filter) :D

[Koepi]

Homy would've said:


hmmmmmmmm.... delicious.....

(I'm quite sure :) )

I'm a little unsure, do only the new Duron and Athlon cores support SSE or do they support sse2 and the old cores just sse?

(I'd really love to benefit from it ;) )

Regards,
Koepi

[rui]

Koepi, if i am not mistaken, the old duron and athlon cores only supported integer SSE.
The new cores support integer and floating point SSE. But none of them supports SSE2, only the hammer line will be hable to do that.

Like always, please someone correct me if i am wrong

[Koepi]

thanks rui, that sounds like i remembered correctly :-/

so, stuff sse/sse2, max. use integer sse ;)

Regards,
Koepi

[Nic]

No, Homer would say:
"Your optimising the what with the what now?" :D

And yup what rui said about instruction sets is correct. :)

The new code, is now pretty fast (compared to normal deringing, it's not very hard!), but its hard for me to be sure its working as well (not just Tom's part but the changes ive made)

Cheers,
-Nic

[trbarry]

It's just MMXSSE, integer only. I've almost never done any floating point assembler and probably wouldn't be any good at it.

But only parts of the deringing are asm optimized so far.

- Tom

[Nic]

Well, the end of the weak draws ever near, which means I won't be able to update in VC7 (I should really take a copy home :) )

<<::: Rummages around at work, asks a few people ::::>>

Ok ive got the DVD of Visual Studio .net now, so I will be able to work on it at home :D

Even still if you re-download the version up at:
http://nic.dnsalias.com (still marked 13/03/02)

Its got Tom's SSE MMX code in for deringing (only part of the dering filter is in MMX + SSE at present so its not a huge leap forward, but it definitely helps)

Could people both test the speed & quality for me,

Cheers,
-Nic

ps
Ive taken some other stuff out, so if it glitches thats why.

[rui]

Nic, i don't know how to say (post) this but:

The new ds filter crashes windows media player (6.4 version) when i choose deringing :(

And, when only choose the deblock options, the quality seems a little inferior to the other ds filters, in spite of seeming faster.

I have Win95, P2-350 here (no SSE). Don't know if it is something related with my comp.

[Nic]

Yup, sorry rui, your computers not upto the job of SSE..... :(

I dont know what to say......

(The deblocking code hasn't changed, but ill look into it anyway)

Cheers,
-Nic

[rui]

Well, seems no deringing to me, at least here in my work.

But in home i have an AMD XP :D

Nic, if the deblocking code hasn't changed, then don't worry about it, maybe it's my eyes that are tricking me ;)

[Nic]

Ill look into it,

The brightness now comes before the post-processing....this might effect deblocking....I don't know,

Ill do some tests tonight :)

Cheers,
-Nic

[avih]

yes, it does :)
since the threshes change. from my experience (and i tried both and thought of the implications) brightness/contrast should be applied at the end of the chain.

if u use many filters, it's turning to be a headache deciding which filter comes first :)

[Nic]

I thought that as I was doing it... :)

The problem is, with the present code, its quicker that way :)

Ill change it back to have brightness at the end. :)

Cheers,
-Nic

[avih]

tell me about it lol
;)

yup, as u get more filters crumping them together makes it faster, but harder to configure/modify code.. :(