Hello, I don't know if this topic was discussed earlier in some of the GSL lists. What about the possibility of the parallelization of the GSL CBLAS implementation via OpenMP? Nowadays, the multicore processors are the most common hardware and recent compilers (gcc since 4.2 series) supports the OpenMP standard. I think that it will be easy to parallelize at least the CBLAS level 3 (since they perform the most computations) functions. Is clear that who needs powerful must use an optimised CBLAS implementation as ATLAS or GotoBLAS (now free software under the BSD license: http://www.tacc.utexas.edu/tacc-projects/gotoblas2/) but I think that using OpenMP in GSL CBLAS would improve the speed of the standard GSL distribution and it wouldn't suppose a big effort of programming. Attached I send a patch for the source_gemm_r.h file in order to parallelize dgemm and sgemm functions. It is only a patch in order to see the simplicity of adding OpenMP support. I didn't test it as now I'm not have here a multicore here.
Thanks -- ***************************************** José Luis García Pallero [email protected] (o< / / \ V_/_ Use Debian GNU/Linux and enjoy! *****************************************
source_gemm_r.h.patch
Description: Binary data
_______________________________________________ Help-gsl mailing list [email protected] http://lists.gnu.org/mailman/listinfo/help-gsl
