Hi,
I would like to parallelize the computation of an histogram.
- Let's suppose that you have a std::vector<double> of elements in between 0 and 1 (think of a size of 1 billion)
- Let's suppose that you have an integer n known at runtime that subdivide [0, 1] in [0, 1/n[, [1/n, 2/n[, ..., [(n-1)/n, 1].
For every k in {0,1,...,n-1}, I would like to compute the number of elements of the vector in [k, (k+1)/n[. I have managed to program that in OpenMP, with an histogram for every thread and a summation of all the histogram at the end. However, as I want to use a Xeon Phi with many threads, I would like the summation at the end to be in log(nb_of_threads) instead of being linear. Is there a way to do that using OpenMP 4?
Best regards,
Francois