[vlc-devel] Phosphor timing tests (reworked += 5)

[Juha Jeronen]

Hi,

On 03/30/11 15:57, Rémi Duraffort wrote:
> Hello,
>
>> Timing the new version ("reworked += 5"), DarkenField() takes about
>> 500us per call in 4:2:2 mode on my Atom. It used to take 750us, so the
>> simpler MMX code is a significant improvement. It also makes the code a
>> bit shorter and easier to read.
>>
>> The performance scaled surprisingly linearly: 12 instructions vs. 23 per
>> loop produced a 100% speed increase in the 4:2:2 chroma handler. It
>> seems it was CPU bound.
>
> Only one word: congrats !

Thanks :)

> That's a good news so thanks for your work on that and keep up the good
> work (I like when we have some benchmarks:))

My thanks goes to Laurent for figuring out what my MMX code wants to do, 
and then doing it twice faster ;)

(Both in this case and in CalculateInterlaceScore() in the upcoming IVTC 
patch...)

I'll keep benchmarking my creations. I think the performance impact is 
relevant, especially on low-end machines. Netbooks and other mobile 
equipment are a nice portable video source, and it's important to get as 
much out of the available CPU cycles as reasonably possible.

  -J