At 20:53 07/06/2008, David Schleef wrote: >Also, since arithmetic coding is an entirely serial operation, GPUs >are not particularly good at this either. Hopefully newer generations >of video cards will have dedicated units for this.
Although David is correct that arithmetic coding is entirely serial, nevertheless GPU might help (at least with Cuda). In Dirac each subband is separately arithmetically encoded. From memory there are several dozen independent arithmetically coded subbands (4 level wavelet transform gives 13 subbands and there are separate subbands for YUV, plus the motion information is also arithmetically coded). If we put each of these on a separate processor we should get a speed up. There is a difficulty in that high frequency subbands have a dependency on previously decoded lower frequency sub-band. Nevertheless with appropriate pipelining I think it should be possible to get significant speedups using a multiprocessor array (as on Nvidia's newer GPUs) Tim ------------------------------------------------------------------------- Check out the new SourceForge.net Marketplace. It's the best place to buy or sell services for just about anything Open Source. http://sourceforge.net/services/buy/index.php _______________________________________________ Schrodinger-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/schrodinger-devel
