I'm using EVD (LAPACKE_cheev), I'd like to know if there is a openMP accelerated version in the ARM PL.
Thanks!
Hi,
In Arm PL 24.10, if you link to libarmpl_mp, the EVD implementation in Arm PL will parallelize some of the calls to BLAS routines called as part of the EVD algorithm, depending on the sizes of the calls to BLAS being made. We will be looking into adding parallelism in the symmetric EVD algorithm itself for inclusion in future releases.
Best Regards,
Chris.
Hi Chrs,
I tried linking with both libarmpl and libarmpl_mp, but I didn't notice any difference in runtime.
libarmpl
libarmpl_mp
Is there any way to tell if this function supports multithreading or multi-core processing?
# without openmp target_link_libraries(test_armpl ${CUDA_LIBRARIES} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_CUSOLVER_LIBRARIES} cublas cusolver armpl_lp64 pthread m) # with openmp target_link_libraries(test_armpl ${CUDA_LIBRARIES} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_CUSOLVER_LIBRARIES} cublas cusolver armpl_lp64_mp pthread m)
There is no way to tell other than to notice a speedup in performance, but in this case it sounds like your problem does not currently benefit from multithreading on your system.