This is the implementation of the algorithm IPS²Ra presented in the paper Engineering In-place (Shared-memory) Sorting Algorithms, which contains an in-depth description of its inner workings, as well as an extensive experimental performance evaluation. Here's the abstract:
We present new sequential and parallel sorting algorithms that now represent the fastest known techniques for a wide range of input sizes, input distributions, data types, and machines. Somewhat surprisingly, part of the speed advantage is due to the additional feature of the algorithms to work in-place, i.e., they do not need a significant amount of space beyond the input array. Previously, the in-place feature often implied performance penalties. Our main algorithmic contribution is a blockwise approach to in-place data distribution that is provably cache-efficient. We also parallelize this approach taking dynamic load balancing and memory locality into account.
Our new comparison-based algorithm In-place Superscalar Samplesort (IPS⁴o), combines this technique with branchless decision trees. By taking cases with many equal elements into account and by adapting the distribution degree dynamically, we obtain a highly robust algorithm that outperforms the best previous in-place parallel comparison-based sorting algorithms by almost a factor of three. That algorithm also outperforms the best comparison-based competitors regardless of whether we consider in-place or not in-place, parallel or sequential settings.
Another surprising result is that IPS⁴o even outperforms the best (in-place or not in-place) integer sorting algorithms in a wide range of situations. In many of the remaining cases (often involving near-uniform input distributions, small keys, or a sequential setting), our new In-place Parallel Super Scalar Radix Sort (IPS²Ra) turns out to be the best algorithm.
Claims to have the -- in some sense -- "best" sorting algorithm can be found in many papers which cannot all be true. Therefore, we base our conclusions on an extensive experimental study involving a large part of the cross product of 21 state-of-the-art sorting codes, 6 data types, 10 input distributions, 4 machines, 4 memory allocation strategies, and input sizes varying over 7 orders of magnitude. This confirms the claims made about the robust performance of our algorithms while revealing major performance problems in many competitors outside the concrete set of measurements reported in the associated publications. This is particularly true for integer sorting algorithms giving one reason to prefer comparison-based algorithms for robust general-purpose sorting.
The radix sort algorithm IPS²Ra is an adaption of the samplesort algorithm IPS⁴o. The implementation of IPS⁴o is also available on GitHub. In the sequential case, IPS²Ra outperforms IPS⁴o for many input distributions and data types. However, IPS⁴o may be faster for very skewed key distributions. In the parallel case, IPS⁴o outperforms IPS²Ra most of the time.
An initial version of IPS⁴o has been described in our publication on the 25th Annual European Symposium on Algorithms.
Clone this repository and check out its submodule
git clone --recurse-submodules https://github.com/ips4o/ips2ra.gitor use the following commands instead if you want to include this repository as a submodule:
git submodule add https://github.com/ips4o/ips2ra.git
git submodule update --recursive --initIPS²Ra provides a CMake library for simple usage:
add_subdirectory(<path-to-the-ips2ra-repository>)
target_link_libraries(<your-target> PRIVATE ips2ra)A minimal working example:
#include "ips2ra.hpp"
// sort sequentially
ips2ra::sort(begin, end[, Extractor = ips2ra::Config<>::identity]);
// sort in parallel (uses OpenMP if available, std::thread otherwise)
ips2ra::parallel::sort(begin, end);The parallel version of IPS²Ra requires 16-byte atomic compare-and-exchange instructions to run the fastest.
Most CPUs and compilers support 16-byte compare-and-exchange instructions nowadays.
If the CPU in question does so, IPS²Ra uses 16-byte compare-and-exchange instructions when you set your CPU correctly (e.g., -march=native) or when you enable the instructions explicitly (-mcx16).
In this case, you also have to link against GCC's libatomic (-latomic).
Otherwise, we emulate some 16-byte compare-and-exchange instructions with locks which may slightly mitigate the performance of IPS²Ra.
If you use the CMake example shown above, we automatically optimize IPS²Ra for the native CPU (e.g., -march=native).
You can disable the CMake property IPS2RA_OPTIMIZE_FOR_NATIVE to avoid native optimization and you can enable the CMake property IPS2RA_USE_MCX16 if you compile with GCC or Clang to enable 16-byte compare-and-exchange instructions explicitly.
IPS²Ra uses C++ threads if not specified otherwise.
If you prefer OpenMP threads, you need to enable OpenMP threads, e.g., enable the CMake property IPS2RA_USE_OPENMP or add OpenMP to your target.
If you enable the CMake property DISABLE_IPS2RA_PARALLEL, most of the parallel code will not be compiled and no parallel libraries will be linked.
Otherwise, CMake automatically enables C++ threads (e.g., -pthread) and links against TBB and GCC's libatomic. (Only when you compile your code for 16-byte compare-and-exchange instructions you need libatomic.)
Thus, you need the Thread Building Blocks (TBB) library to compile and execute the parallel version of IPS²Ra.
We search for TBB with find_package(TBB REQUIRED).
If you want to execute IPS²Ra in parallel but your TBB library is not accessible via find_package(TBB REQUIRED), you can still compile IPS²Ra with parallel support.
Just enable the CMake property DISABLE_IPS2RA_PARALLEL, enable C++ threads for your own target and link your own target against your TBB library (and also link your target against libatomic if you want 16-byte atomic compare-and-exchange instruction support).
If you do not set a CMake build type, we use the build type Release which disables debugging (e.g., -DNDEBUG) and enables optimizations (e.g., -O3).
Currently, the code does not compile on Windows.
IPS²Ra is free software provided under the BSD 2-Clause License described in the LICENSE file. If you use IPS²Ra in an academic setting please cite the paper Engineering In-place (Shared-memory) Sorting Algorithms using the BibTeX entry
@misc{axtmann2020engineering,
title = {Engineering In-place (Shared-memory) Sorting Algorithms},
author = {Michael Axtmann and Sascha Witt and Daniel Ferizovic and Peter Sanders},
howpublished = {Computing Research Repository (CoRR)},
year = {Sept. 2020},
archivePrefix ={arXiv},
eprint = {2009.13569},
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent