Re: [boost] [SORT] Parallel Algorithms

Subject: Re: [boost] [SORT] Parallel Algorithms
From: Steven Ross (spreadsort_at_[hidden])
Date: 2014-12-22 06:08:56

Francisco,

On Thu, Dec 18, 2014 at 10:22 PM, Steven Ross <spreadsort_at_[hidden]> wrote:
> Francisco,
>
> Have you looked at:
> https://software.intel.com/en-us/articles/a-parallel-stable-sort-using-c11-for-tbb-cilk-plus-and-openmp
>
> For comparison?

I used the free tbb parallel stable sort referenced above for
comparison, and found that as soon as I randomized the contents of the
entire struct and made the copy constructor copy over the entire
struct, that the tbb version was 37% faster on randomized data all the
way up to 256 bytes relative to countertree::parallel_merge_sort.
I've attached my modified "original" directory where I tested this
out (see: Original/benchmark/parallel_stable_sort/build/make_tbb_benchmark.sh).
Unless you can get your library close to the speed of this tbb
sort, I don't see how we'd be benefiting people by pointing to it
instead of the tbb library.

What I am interested in is your idea for indirect sorting: can you
come up with an easy-to-use API to handle efficient indirect sorting?
That would probably be worth including in the boost::sort library,
especially if it is compatible with different sort functions.