A simple example to compare coarrays and openmp

[ivanpribec]

@adenchfi

PS: Is there a ‘Hide’ or ‘Hide Spoiler’ syntax on here? So that anyone who doesn’t want to see the output, can just not click ‘Show’? The blur spoiler is nice but space-wasting.

You can enclose the output in a “Details section”. See Click-to-show code snippets - #2 by ivanpribec

---

@JohnCampbell

This straightforward example would not be effective on the i7/ryzen hardware that I use.
For n small, the omp region initiation would swamp the computing time.
For n large, larger than cache, the memory delays would be the same as for a single thread /serial case.
The inner loop calculation is just insufficient to justify multiple threads. It would be similar to using omp for dot_product.
Modifying the code to target AVX would be the better outcome.

This was really just a toy demo. I agree with your concerns about the overheads and delays.
As your question toward the end indicates, the payload also affects the choice of parallelization strategy. I also like to keep the 80-20 principle in mind; often it’s hard to justify the increased complexity of a program which reaches close to the theoretical-max speed-up. The following comic comes to mind: