Boost.SIMD is a portable SIMD programming library to be proposed as a Boost library. It's a sensible subset of the features proposed by bSIMD - NumScale's closed-source library for SIMD programming - that we saw fit for open sourcing.
Boost.SIMD was developed to simplify the difficult, tedious and often error-prone process of developing SIMD programs. Boost.SIMD is designed to seamlessly integrate into existing projects so that you can quickly and easily start developing high performance, portable and future-proof software.
Single instruction, multiple data (SIMD) instructions or multimedia extensions have been available for many years. They are designed to significantly accelerate code execution, however they require expertise to be used correctly and they depend on potentially fragile compiler support and the use of low-level intrinsics, or vendor-specific libraries. The introduction of these instruction sets has allowed developers to exploit the latent data parallelism available in applications by executing a given instruction simultaneously on multiple data stored in dedicated cpu registers. The SIMD unit is built as an independent computation unit in the processor, it comes in addition to the regular computation unit complete with a special register file, dispatching and pipelining unit.
The use of SIMD instructions requires developers to program into a very verbose manner due to the low-level nature of SIMD instruction sets or to rely on fragile, black-box auto-vectorizating compilers. Furthermore, it can be necessary to re-write code for each revision of each target architecture, accounting for each architecture's vendor provided API as well as architecture dependent implementation details. This greatly complicates the design and maintenance of SIMD code, significantly increasing the time required to develop, test and deploy software as well as increasing the scope for introducing bugs.
Boost.SIMD is designed to be user-friendly and easy to integrate into existing projects. To make life even easier, Boost.SIMD includes a large number of commonly used functions and constants, standardizing the use of common SIMD programming idioms. Boost.SIMD allows you to focus on the important part of your work: the development of new features and functionality.
Boost.SIMD provides the following:
- a proper value semantic wrapper for SIMD registers;
- an automatic system to detect and exploit architecture specific optimization opportunities;
- standard compliant ranges and algorithms to process contiguous range of data in a SIMD compatible way;
- over 350 operators and mathematical functions usable on scalar and SIMD values
| Architecture | Extensions |
|---|---|
| x86 | SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, AVX, FMA3, AVX2 |
Boost.SIMD requires a C++11 compliant compiler and Boost 1.61 and is thoroughly tested on the following compilers:
| Compiler | Version |
|---|---|
| g++ | 4.8 and above |
| clang++ | 3.5 and above |
| Microsoft Visual Studio | 15.0 and above |
We start out on the premise you already have a proper installation of Boost 1.61 or superior setup on your machine.
You can retrieve the current status of the library by cloning the repository:
git clone https://github.com/NumScale/boost.simd.git
By default, the develop branch is fetched. You can change that to master if you want to use the stable version instead of the development version.
Once cloned, the library headers are located in the include/ folder and are ready to be used. A very simple example like:
#include <boost/simd/pack.hpp>
#include <iostream>
namespace bs = boost::simd;
int main()
{
bs::pack<float,4> p{1.f,2.f,3.f,4.f};
std::cout << p + 10*p << "\n";
return 0;
}can then be compiled via:
g++ main.cpp -std=c++11 -O3 -msse4.2 -I<path/to/boost/1.61/include> -I<path/to/boost/simd/include> -o main
and display:
(11, 22, 33, 44)
Now, you can have a look at our documentation to follow one of our tutorials on how to use Boost.SIMD.