[libcxx] pdqsort - a faster std::sort

[Eric Fiselier]

I'll try and get to this next week. Sorry for the delay.

/Eric

On Sun, Apr 5, 2015 at 7:21 PM, Marshall Clow <mclow.lists at gmail.com> wrote:
> On Sun, Apr 5, 2015 at 12:29 AM, Orson Peters <orsonpeters at gmail.com> wrote:
>>
>> I’ve added “[libcxx]” to the subject of this email.  It was accidentally
>> omitted from my original post, and this may have caused this post to be
>> accidentally overlooked by the libc++ development team.
>>
>> A while back I shortly discussed this with M. Clow, today I post it here
>> to poll public interest.
>>
>> I'd like to present pattern-defeating quicksort, a novel sorting
>> algorithm. It is as fast or faster than std::sort on any implementation I've
>> tested (libstdc++, libc++, VS2013) for both random data and a variety of
>> patterns (equal, asc/desc, pipe organ, etc).
>>
>> A short explanation of the design of pdqsort can be found in the readme:
>> https://github.com/orlp/pdqsort .
>>
>> The performance chart included in the readme is benchmarked against
>> libstdc++, a comparison with libc++ (compiled with clang) can be found here:
>> http://i.imgur.com/JRiJk0g.png .
>>
>> The code of pdqsort is very short (~300LOC) and clean. Understanding
>> pdqsort and integration should be painless.
>>
>> I'm willing to prepare a patch and do the legal wizardry required to
>> integrate pdqsort as std::sort in libstdc++, if the interest is there. This
>> would also solve this bug I found:
>> https://llvm.org/bugs/show_bug.cgi?id=20837
>>
>> If you have any questions after reading the readme, please do reply and I
>> will do my best to answer them.
>>
>
> Orson --
>
> FWIW, the reason I haven't moved on this is that we (ok, Eric) is working on
> a set of benchmarking tests for a bunch of libc++ (including std::sort).
>
> It's not that I'm not interested.
>
> -- Marshall
>