CruNchers XviD P4 optimized Build

[CruNcher]

Cruncher's XviD build 5/04/03 for P4:
- based on CVS HEAD from 5/04/03
- Disabled qpel. It produces odd results in it's current state IMHO - even with SimpleIDCT.
- Disabled fdct_sse2. Was leading to worse quality on P4s. Useing fdct_mmx instead.
- Disabled idct_sse2. Was leading to worse quality on P4s. Useing idct_simpleidct_mmx2 instead.
This build is optimized for P4 testing only.

Update:
As these changes now in Koepis build no need more for this release thx :) get Koepis one

[Nic]

Anymore info on what makes your builds P4 optimised?

-Nic

[CruNcher]

@Nic
not in speed more in quality speed was the second issue but fdct_sse2 and idct_sse2 give odd results in speed and quality here on my p4 1.8 williamete core system, so these two are completly deactivated look in my test table in the "SSE2 fixed- first tests" thread. Simpledidct is used instead for all Processor's same as in your build except the better psnr without useing fdct_sse2 instead normal fdct_mmx is used, it's a little slower but psnr is better and speed decrease for that issue acceptable :)

[Nic]

If you think thats definitely the case for fdct_sse2 then ill make sure it uses fdct_mmx for sse2 in my build next time :)

Thanks for all your tests :)

Cheers,
-Nic

[Koepi]

Ok, so the only difference to our builds is that you threw away the line fdct=fdct_sse2 in xvid.c?

Will be in CVS soon i think ;-)

Btw., your update (in blue) is commited by Michael yesterday - you make it look as if those are all your changes. The "so called" P4-optimization isn't really one. And disabling qpel is crippling xvid, not fixing it.

I don't want to badmouth you, but I think you should be a little more precise and don't use marketing blabla for things you did.

Regards
Koepi

[CruNcher]

@ Koepi

i expected you would react like that, but you have to understand im not a bad guy i don't try to sell something here or to make the XviD project bad in peoples mind, so please don't get mad at me i respect you and what you did in Nandub times and what you do now in XviD times for this Scene.

I also did a preanounce of this release on the Channel and all my testing progresses so everytime anyone could said "hey CruNcher stop this don't release it we don't want that", but nobody did so i thought it is ok.

Koepi i don't want any honor i only see this as giving back something to the XviD project, sure i could do this as a dev and could try optimize fdct_sse2 by myself but im far away from that.

GomGom said he wouldn't like to take out fdct_sse2 and idct_sse2 by default it would be better if someone would work on it and optimize it and that is exaclty the same what i think would be the best in this case.

The updates i only do this too keep track that the modifications i do are compareable with the CVS standard Builds, the Blue color i do this only, because as you can see i don't open a new thread everytime, this is also by intention people who use P4 check out this thread download the build and try it out compare my results with their own (maybe i did something wrong in my compares as with b-frames psnr measurment) so that we can be absulutely sure that fdct_sse2 and idct_sse2 are infact inoptimized and in terms of "quality vs speed" not as good as fdct_mmx and idct_simpleidct_mmx2.

If they interested, they come back and look in the same thread. Then the blue color signalise them "there is a update i need the new build to compare correct". If it goes out of the threadview and a new update is made only then i would start a new thread i won't make a new thread for every Update that's uneeded.

it wasn't my intention to get the honor for making any updates that i doesn't made for XviD and i will correct this imidiatly so excouse me.

Koepi P4 optimized build is half as correct the modifications where actualy done in purpose to get the best quality vs speed for P4 Systems so i think it is infact "optimized" against other builds P4 usage but i don't want that this is integrated in other builds as yours or Nics or even CVS not for now, first it must be absolutely sure that these changes for the moment would be better for the current builds and other P4 users can help to test that and that was my whole intention to start all this.

Sure their alot of people out there who wan't to make profit out of the Devs work but you should know me a little that im not one of these guys :(

[Koepi]

Cruncher,

all I wrote is that you used marketing words for simple things, I can give an example how I would have written this:

Cruncher's XviD build <Date> for P4:
- based on CVS <branch> from <date>
- Disabled qpel. It produces odd results in it's current state IMHO - even with SimpleIDCT.
- Disabled fdct_sse2. Was leading to worse quality on P4s.
(- Doesn't crash on P4s.)
This build thus is optimized for P4 only.


I didn't took any offense in anything. But read your words, you make it sound as if this is the only valid build for P4s. While that "qpel disabled"-point is a VERY hard step IMO.

That's all I wanted to mention with my post. I'm not at all mad at you.

Best regards
Koepi

[CruNcher]

@ Koepi
hope you didn't mind i took your template now and true the text looked like it came straight from the brain off an DXN employe, sorry for that :P but again to get this clear because the Qpel issue is not related to the purpose to test idct_sse2 and fdct_sse2 and so i decided to deactivate it Qpel is tested for now in many other builds like yours and Nics so i see no use for Qpel activated in this build, which purpose its only to test idct_sse2 and fdct_sse2.

[Koepi]

Cool, now it all makes sense and we (hm. well. I. ;) ) know what it's all about =)

Thanks for your efforts!

Regards
Koepi

[trbarry]

Disabled idct_sse2. Was leading to worse quality on P4s. Useing idct_simpleidct_mmx2 instead.

Funny. At one time I think idct_sse2 and idct_mmx (not simple) would give exactly the same results. Maybe a new bug has been introduced?

- Tom