Re: [eigen] FLENS C++ expression template Library has excellent documentation

[ Thread Index | Date Index | More Archives ]

> There are different use cases.
> - many products of many small matrices
> - one product of two large matrices

This kind of question needs thought. You see, I was pretty
disappointed at the rather poor performance of eigen in the Pentium D
benchmarks. I am thinking that that is due to there being machine
dependent constants in the eigen which need to be tuned for each

Perhaps, we need an auto tuner which will run a wide variety of
benchmarks and set those constants in a header file that other
programs include. That header file comes with some sane constants when
you download it so you can run the auto tuners to optimize.
Parallelism is beneficial only if your problem size is greater than a
particular number. That number also should be decided on a per
algorithm basis by that auto-tuner. Now that we are thinking of
parallelism, we should consider this as well.

Rohit Garg

Senior Undergraduate
Department of Physics
Indian Institute of Technology

Mail converted by MHonArc 2.6.19+