On 3/8/2011 9:02 AM, Ingo Schmid wrote: > Dear all, > > there are certainly two major features which impact PDL performance: > > 1. Threading should be aware of multiple CPUs and cores > automagically and use them.
This would be nice to have. > 2. 2GB hard limit due to declaration of dimensions and sizes > as 32bit integers. > > While 2 is not strictly about performance in terms of speed, > it is annoying for a number crunching engine. This has to be > resolved in the near future to stay competitive. I am hit by > those already. I'd be willing to help addressing these issues. There are a number of issues involved in this task. See the feature request tracker item at sf.net where I've been adding notes as to the things that need to be done including: (1) convert the pdl object size from int to size_t (2) verify that uses of ->nelem are ok with long long type (3) make sure the 64bit support doesn't break for 32bit perls (4) update the PP threading to support long long indexing The link to the sf.net ticket is: http://sourceforge.net/tracker/?func=detail&aid=3039433&group_id=612&atid=350612 Another thing that is needed is a way to verify the operation of the new and old functionality via a set of tests. I'm using cygwin now so I'm limited to less than 512MB or so for PDL so someone with a large memory machine and 64bit perl will need to be involved with the development. Ingo, even something as simple as some tests that fail now and would work in a 64bit PDL would be helpful.... Cheers, Chris > It outperforms matlab, which most of our group use, easily. > Best > Ingo _______________________________________________ Perldl mailing list [email protected] http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
