OpenMP optimization for specific number of threads

Question

I'm developing a scientific code, so of course speed is of the essence. Now because of that portability is not really an issue and so I know how many openmp threads I will have available already when compiling the program. Can I use this information to perform any additional optimization? If yes, how do I do so?

Since it was pointed out that this question is very broad. I want to reduce it a bit to automatic, i.e. compiler optimization. So setting compiler flags or similar things.

Cheers

-A

It's still too broad. For now the best answer is: **it depends on the code**. — Hristo Iliev, May 07 '12 at 07:44
Yes, you should certainly set compiler flags. I'd add to what @Hristo Iliev has already written: it also depends on the compiler. Read The Fortran (compiler) Manual. — High Performance Mark, May 08 '12 at 08:24

score 0 · Answer 1 · answered May 07 '12 at 08:04

0

Well,you can modify the code such that it can be divided into n independent regions(n=no. of threads). U should prefer using sections.They give better speedups as compared to parallel for loops due to reduced inter processor communication.

answered May 07 '12 at 08:04

haxor

211
1
2
11

Sections are no faster than `for schedule(static)` work-share construct. – Hristo Iliev May 07 '12 at 14:20
Well,I had implemented some codes where i used both parallel for and sections (odd and even loops) and sections seemed to be a little bit faster.It was quite a while ago,maybe it did so for that particular code. – haxor May 08 '12 at 17:18

OpenMP optimization for specific number of threads

1 Answers1