Looking for feedback on a possible inplace_stable_sort replacement
richard at metafoo.co.uk
Tue Mar 18 13:46:22 PDT 2014
On Thu, Mar 13, 2014 at 11:06 AM, Mike McFadden <mike at ivideoapp.com> wrote:
> Hi, I used a paper called "Ratio based in-place stable merging" to
> implement a hybrid stable sort algorithm, and have been validating the
> results against stable_sort and inplace_stable_sort. What's interesting is
> that it matches 80-90% of the speed of stable_sort with random data, is
> much faster in many other cases, and does so while using O(1) memory.
> Simply giving it more memory allows it to closely match stable_sort in
> speed. And compared to inplace_stable_sort it's anywhere from 3-15x faster,
> with random data being about 5x faster.
> I have fully documented the algorithm and have a C++ implementation
> available here:
> The underlying algorithm was vetted and proven back in 2008 (there's a
> link to the published paper on the GitHub project page), and the simplified
> version used here is constantly tested against existing algorithms for
> correctness and speed, but toying with a core feature of libc++ is
> obviously not something to be taken lightly so I'm hoping for some
> additional vetting and consideration.
Looks interesting to me. You should be able to accelerate std::rotate by
using the cache even if neither range fits within it (perform a gcd rotate,
moving a cache-sized chunk of the data at a time).
Incidentally, your current Rotate function is cheating. You can't use
std::memcpy/std::memmove on objects of class type in general.
> My ultimate goal would be to replace inplace_stable_sort at the very
> least, and ideally stable_sort as well. Maybe including it as a separate
> sort function for now would be wise?
> - Mike
> cfe-commits mailing list
> cfe-commits at cs.uiuc.edu
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the cfe-commits