Interesting iDCT test

[Fishman0919]

I did this test a few days ago but thought some people who read this thread might be interested in the result.

http://forum.doom9.org/showthread.php?p=682312#post682312

Really not much of a diff between any of the iDCT's

[Video Dude]

Great test.

Your results seem to match Donald Graft's documentation in DGDecode. Interesting to see the small error percentage that separates them.



Qualitywise: IEEE-1180 Reference > 64-bit Floating Point > Simple MMX (XviD) > Remaining iDCTs.

Speedwise: SSE2/MMX and SSE/MMX (Skal) are usually the fastest. The IEEE-1180 Reference is easily the slowest.


IDCT SSIM Time
--------------------------------------------
1 32-bit MMX 99.63% 2:15
2 32-bit SSEMMX 99.63% 2:14
3 32-bit SSE2MMX 99.63% 2:14
4 64-bit Floating Point 99.88% 2:41
5 64-bit IEEE-1180 Reference 99.93% 2:52
6 32-bit SSEMMX (Skal) 99.67% 2:11
7 32-bit Simple MMX (XviD) 99.74% 2:12

[Carpo]

so would this be a reason other than bug fixes to ask jdobbs to support newer versions of DGDecode - or does it not really make a diff over all ?

[Rizlaw]

I'm not really sure what all this means, but I did find an enlightening post on Doom9 about this subject, which I think may be more helpful to those of us who either don't understand this arcane stuff, or just don't care.

http://64.233.161.104/search?q=cache:2Yo7sIPCxFkJ:www.doom9.org/idct.htm+idct&hl=en&client=firefox-a

http://www.doom9.org/idct.htm

Edit: added second link since first link seems only to work for Firefox browser.

[Guest]

@Rizlaw

Bad link.

[Rizlaw]

@Rizlaw

Bad link.

Neuron2,

I tried it from my post and it works using Firefox 1.0.5, I would assume it should work for all browsers. Anyway, here is the link to the actual Doom9 page, or so Google says:

http://www.doom9.org/idct.htm

[Guest]

I tried it from my post and it works using Firefox 1.0.5, I would assume it should work for all browsers. Perhaps not when your link specifies the client as Firefox!