C++ wrappers for SIMD intrinsics

SIMD (Single Instruction, Multiple Data) is a feature of microprocessors that has been available for many years. SIMD instructions perform a single operation on a batch of values at once, and thus provide a way to significantly accelerate code execution. However, these instructions differ between microprocessor vendors and compilers.

xsimd provides a unified means for using these features for library authors. Namely, it enables manipulation of batches of numbers with the same arithmetic operators as for single values. It also provides accelerated implementation of common mathematical functions operating on batches.

You can find out more about this implementation of C++ wrappers for SIMD intrinsics at the The C++ Scientist. The mathematical functions are a lightweight implementation of the algorithms used in boost.SIMD.

xsimd requires a C++11 compliant compiler. The following C++ compilers are supported:

The following SIMD instruction set extensions are supported:

A package for xsimd is available on the mamba (or conda) package manager.

mamba install -c conda-forge xsimd

A package for xsimd is available on the Spack package manager.

spack install xsimd
spack load xsimd

You can directly install it from the sources with cmake:

cmake -D CMAKE_INSTALL_PREFIX=your_install_prefix
make install

To get started with using xsimd, check out the full documentation

http://xsimd.readthedocs.io/

Here is an example that computes the mean of two sets of 4 double floating point values, assuming AVX extension is supported:

#include <iostream>
#include "xsimd/xsimd.hpp"

namespace xs = xsimd;

int main(int argc, char* argv[])
{
    xs::batch<double, 4> a(1.5, 2.5, 3.5, 4.5);
    xs::batch<double, 4> b(2.5, 3.5, 4.5, 5.5);
    auto mean = (a + b) / 2;
    std::cout << mean << std::endl;
    return 0;
}

Do not forget to enable AVX extension when building the example. With gcc or clang, this is done with the -march=native flag, on MSVC you have to pass the /arch:AVX option.

This example outputs:

(2.0, 3.0, 4.0, 5.0)

The same computation operating on vectors and using the most performant instruction set available:

#include <cstddef>
#include <vector>
#include "xsimd/xsimd.hpp"

namespace xs = xsimd;
using vector_type = std::vector<double, xsimd::aligned_allocator<double>>;

void mean(const vector_type& a, const vector_type& b, vector_type& res)
{
    std::size_t size = a.size();
    constexpr std::size_t simd_size = xsimd::simd_type<double>::size;
    std::size_t vec_size = size - size % simd_size;

    for(std::size_t i = 0; i < vec_size; i += simd_size)
    {
        auto ba = xs::load_aligned(&a[i]);
        auto bb = xs::load_aligned(&b[i]);
        auto bres = (ba + bb) / 2.;
        bres.store_aligned(&res[i]);
    }
    for(std::size_t i = vec_size; i < size; ++i)
    {
        res[i] = (a[i] + b[i]) / 2.;
    }
}

We also implement STL algorithms to work optimally on batches. Using xsimd::transform the loop from the example becomes:

#include <cstddef>
#include <vector>
#include "xsimd/xsimd.hpp"
#include "xsimd/stl/algorithms.hpp"

namespace xs = xsimd;
using vector_type = std::vector<double, xsimd::aligned_allocator<double>>;

void mean(const vector_type& a, const vector_type& b, vector_type& res)
{
    xsimd::transform(a.begin(), a.end(), b.begin(), res.begin(),
                     [](const auto& x, const auto& y) { (x + y) / 2.; });
}

Building the tests requires the GTest testing framework and cmake.

gtest and cmake are available as a packages for most linux distributions. Besides, they can also be installed with the conda package manager (even on windows):

conda install -c conda-forge gtest cmake

Once gtest and cmake are installed, you can build and run the tests:

mkdir build
cd build
cmake ../ -DBUILD_TESTS=ON
make xtest

In the context of continuous integration with Travis CI, tests are run in a conda environment, which can be activated with

cd test
conda env create -f ./test-environment.yml
source activate test-xsimd
cd ..
cmake . -DBUILD_TESTS=ON
make xtest

xsimd's documentation is built with three tools

• doxygen
• sphinx
• breathe

While doxygen must be installed separately, you can install breathe by typing

pip install breathe

Breathe can also be installed with conda

conda install -c conda-forge breathe

Finally, build the documentation with

make html

from the docs subdirectory.

We use a shared copyright model that enables all contributors to maintain the copyright on their contributions.

This software is licensed under the BSD-3-Clause license. See the LICENSE file for details.