[x264-devel] Hpel interpolation - marginal speed increase

[Mathieu Monnier]

Hi,

I played a little more with the hpel interpolation, and manage to gain 
some speed while cleaning the code a little. At subme 1 - analyse 
"none", the speed goes from ~13.10 fps to ~13.40 fps, so it's only a 3% 
speed increase ( interpolation was taking ~10% of the encoding time with 
these settings ), and it won't be noticeable at higher subme / analyse 
settings.

The speed gain wasn't obtained through intruction reordering nor 
prefetching ( I did add a prefetch, but the speed boost it gives is 
really negligible ), but only thanks to a better loop counter 
management. I was rather astonished to get a speed gain in the first 
place, since I tweaked the loops only to be able to prefetch.

Prefetch didn't give the results expected, though I didn't try that 
hard, and I didn't make all the code modifications needed for it to be 
plainly effective.

These results were obtained on an Athlon XP 2000+ ( 3 years old ).

So here is the patch ( against rev 220 )

Regards,

Mathieu




-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: x264_hpel4.patch
Url: http://mailman.videolan.org/pipermail/x264-devel/attachments/20050507/836abab3/attachment.txt