[sldev] VWR-940, VWR-962, VWR975

[Dzonatas]

Paul TBBle Hampson wrote:
> I would however be particularly interested in the method of profiling
> you use, so I can try applying it to my build here, and work out why
> _it_ is so painfully slow.
>   
I used VTune and also added benchmarks. I've also tested code that runs 
in a single thread and compared it to the live situation in SL.

It's interesting that a few spin locks come up as hotspots. Their 
threshold could be optimized.

> I think so, yes. Over the course of this discussion, a voice in the back
> of my head has been helpfully suggesting that we've got two classes that
> are trying to do the same thing...
>   
One is raw data from the hardware while the other is what the OS reports.

I've sorted these patches. I'm compiling today for final tests.


--