Re: [UFRaw-Devel] OpenMP support for VNG interpolation

[Bruce G. <lis...@br...>]

On Wed, Aug 12, 2009 at 09:30:18AM +0000, Omari Stephens wrote:
> Bruce Guenter wrote:
> >   Are there any other areas I should
> > be looking at?
> 
> Denoising can add a significant amount of time when it's enabled (that is, 
> amount is set to non-zero value)

The denoising algorithm already has parallelism enabled.  Each of the 3
or 4 sensor colors is run independently, and will run in separate cores
if there are enough threads enabled.  The rest of the algorithm is
fairly interdependent, and cannot be parallelized easily.  Smaller parts
may be possible for systems with more than 4 cores, but the benefits are
likely to be lost in the cost of thread synchronization.

-- 
Bruce Guenter <br...@un...>                http://untroubled.org/