Strange profiling result with pardiso

[dshor1]

Hi all

Our app uses multiple solve with the same factor, when I looked at the call graph results' it seems that about 30% of the time is in the LSAME function.

70% of DTRSM ( most of the SOLVE work) is the blas_errchk_dtrsm, that calls LSAME.

Is that reasonable? Just some profiling artifact? Or a configuration problem ?

The problem size in about 10000.

Thanks

[michael_chuvelev]

Hi!

It shouldn't spend much time a LSAME - it's just a few instructions. What is theprofiler?

[dshor1]

I know It shouldn't ... I have a felling that the problem is calling the BLAS function with very small size, and that the reson for the large overhead.

the profiler is intel vtune