Re: [boost] Fwd: [SORT] Parallel Algorithms

Subject: Re: [boost] Fwd: [SORT] Parallel Algorithms
From: Steven Ross (spreadsort_at_[hidden])
Date: 2015-06-29 06:32:37

On Fri, Jun 26, 2015 at 11:01 AM Francisco José Tapia <fjtapia_at_[hidden]>
wrote:

> Hi
>
>  the code and the documentation are finished
>
>  All is in https://github.com/fjtapia/sort_parallel
>
>  Please, if you see any error, mistake or something to correct, say me in
> order to change as soon as possible
>
> Thanks Francisco:
My preliminary testing on Linux suggests that these libraries are
beneficial in at least one case on an 8-core system:

introsort

smart_merge_sort

parallel stable sort
sample sort

parallel introsort didn't prove itself (based on raw speed).  Do you have a
larger test I can run to verify memory usage (which other users might
like), or should I just hack up the tests you provided for this purpose?
If parallel introsort is better than competitors in RAM, then that would be
its reason for use.  It looks like we can tell users that these library
versions (except parallel introsort) are faster than TBB and GCC for very
large data types (256+ bytes), and for other data types they are roughly
competitive in speed, so at minimum, it'll save people the effort of
setting up their own indirect sort.