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

Subject: Re: [boost] Fwd: [SORT] Parallel Algorithms
From: Steven Ross (spreadsort_at_[hidden])
Date: 2015-04-18 10:38:40

On Thu, Apr 16, 2015 at 6:20 AM Francisco José Tapia <fjtapia_at_[hidden]>
wrote:
>
>  I just revised the new version, and the documentation is much better than
> the older. ( The index.html don't run well because don't exist the folder
> doc/html/) .
>

The html is generated from the code and a template.  In 1_58_0 boost
release (which just came out), the generated output is included.  Do you
want to know how to generate the html?

>
> As see in the sort.pdf, the spreadsort function is in the boost::sort
> namespace. This can be coherent with my previous message.
>

That's an obsolete file, and I've deleted it.  Thanks for pointing it out.
You should use the html docs.

>
>  My initial idea is to put the functions in the boost::sort namespace with
> the names
>
>
>    - introsort
>    - parallel_introsort
>    - smart_merge_sort
>    - parallel_stable_sort
>    - sample_sort
>
>  I include the sample sort because in the parallel sorting of strings the
> sample sort is a 27% faster than parallel_stable_sort , and use only a 4%
> more of memory.
>
>  The code can be in a folder with any name, by example,
> boost/sort/generalsort/algorithms.hpp. What's your opinion?
>

The namespace needs to match the folder, that's the boost convention.
The only reason to include introsort (i'd put it in the detail folder) is
for comparison; std::sort usually performs comparably or better, and is
easier to find, in which case "parallel" works (or more specifically,
parallel_comparison) as a name.  I don't like "generalsort" because
spreadsort could be described the same way; it isn't specific enough.

>
>  According to your recommendations, I create the benchmarks, and the data
> are extracted from a file generate with a random number generator.
>
> I am trying to check in machines with many cores, I am talking  with
> friends of universities of Madrid.
>
> I include the spreadsort in the benchmarks, and the results are impressive.
>
> I changed the name of the objects ( now is int_array) , and use the default
> copy constructor and the default operator =, the comparison is with the sum
> of all the values in the array, and all the positions of the array are
> filled with the random numbers of the file
>
>  I am writing the documentation and changing the test programs to adapt to
> the boost format. In a few days, I will create a git repository, with all
> the information
>
> Sounds good.