Dear colleagues!
In early 2009 there was a discussion about fast BLAS library initiated by
Sachin. He reported a faster BLAS library made by Nvidia CUBLAS library. Uwe
Ligges showed an interest for placing the optimized rblas.dll into
windows/contrib section managed by him. Unfortunately there is no any CUBLAS
version of rblas.dll in this section at present. So, is anybody interested
in CUBLAS optimization of rblas now?
In my opinion current directions (packages) of high-performance and
parallel computing with R suffer from grave shortcomings: all GPU
optimization is made by introducing more and more optimized functions
coexisting with original ones. This hard way has been chosen for Matlab too.
It complicates R as a language. So, the idea of optimizing BLAS (especially
by CUBLAS) seems to overcome the mentioned shortcomings. It might be
remarkable to make some efforts to this way of high-performance computing.
Yuri Possokhov
[[alternative HTML version deleted]]
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
