triangular solver

[Jen_B_]

I am using the mkl_?coosv, specifically one with d, and compiled the program with 'ifort -openmp -mkl'. I have set 'mkl_omp_num_threads', and also 'omp_proc_bind=true'. I have tested it for three different thread numbers 4, 8, 16. I am getting the following timings: 0.36, 0.3, and 0.32. I am running it on a machine with 16 cores. are these timings reasonable? or are there anything else I should be doing before doing the runs. thanks.

[Gennady_F_Intel]

all of triangle solvers routines are not threaded for now.In the case if you call this routine many time into your application, we would recommend you to make the convertion from coo -> csr format and then call mkl_?csrsv. It would faster because of internally, we make conversion coo->csr.

--Gennady

[Jen_B_]

Thanks Gennady. Is mkl_?csrsv multi threaded? I.e., would there be any effect if I set "mkl_num_threads"? Also, what are the minimal environmental settings needed before I call mkl functions? Thanks!

[Gennady_F_Intel]

Jen, 

Not! All sparse triangle solvers ( including CSR, DIA and BSR ) are not threaded, but when you call mkl_?coosv, internally we make the conversation from COO to CSR format ).

[Jen_B_]

Gennady,

1) Is it the same for MKL 11.2?

2) So there is no use of playing with MKL_NUM_THREADS for all of mkl_?{csr/bsr/csc/coo}sv?

3) Is 2) same for Xeon PHI too?

Thanks!

[Gennady_F_Intel]

yes, all of these statements apply to version 11.2.

All of these routines ( mkl_?{csr/bsr/csc/coo}sv? )  is single - threaded only 

yes, this is the same for all IA including Xeon Phi.

Gennady

[Jen_B_]

Gennady,

Thanks! Last question does it really matter for the csrsv case, whether the row elements column indices are sorted or not in ascending order?

[Jen_B_]

Gennady,

One more, are there any plans to make these multithreaded? Thanks!