Question

Considering the fact that openmp uses simd model i.e. each instruction is executed by all the threads, so one should be careful while writing openmp parallel application, why should we prefer openmp over multi-threaded version of same algorithm ?

Answer 1

OpenMP isn’t SiMD only anymore from 3.0 onwards.

And common reasons are:

• Ease of use/Amount of code.
• One can compile the same code without -fopenmp flag and get “legacy code retro-compatibility”. Obviously this has to do with the use of #pragma statements.