ippFilterRow runs slower when using static libraries

[mikew123]

Hello,

During my testing of the Intel libraries I've noticed that some functions take significantly longer (up to 5x) when statically linking vs dynamic linking. One in particular is ippiFilterRow. When linking with merged static libraries, it takes about 5x longer to execute then it does when linking dynamically. This happens with IPP 3.0, 6.1, and 7.0. Is this normal?

Thanks,

Mike W

[pvonkaenel · Accepted solution]

Hi Mike,

By default the dynamic linked versions of IPP have OpenMP threading enabled - this allows certain routines to run in parallel across multiple cores. The static version can be configured for either threaded or non-threaded execution. Are you using the threaded static libs?

Peter

[mikew123]

I am using the non-threaded static libs. I understand that the dynamic linked versions use threading in version 7.0, but is this also the case for older versions? I have reproduced this issue using the static libs using IPP versions 3.0 and 6.1 as well.

[igorastakhov]

All ippiFilterXXX functions were threaded in Q1 2002, IPP 3.0 was released in 2003 - so all these IPP version have OMP code for 2D filtering functions.

Regards,
Igor

[mikew123]

Ah, thank you. That explains it.