Expose self-standing parallel routines
Closed this issue · 2 comments
So far, the interface to parallel routines is the standard interface of the sequential routine with an extra ParSeqHelper::Parallel argument at the end. Moreover, these routines assume that they will be called within a PAR_BLOCK, which translates in openmp as a openmp parallel
followed by an openmp single
.
We should also offer a self-standing set of parallel routines who take care of creating the PAR_BLOCK and set the number of threads to a value optionnally given by the caller.
@ZHG2017 : can you implement this? As a starter, this is just the content of the patch we created in
https://trac.sagemath.org/ticket/27444
Then, you should also add PLUQ, Det, Rank, EchelonForm, etc when these parallel routines will have a standardized interface with parallel helpers. (See issue #228 )