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:
Compiler | Version |
---|---|
Microsoft Visual Studio | MSVC 2015 update 2 and above |
g++ | 4.9 and above |
clang | 4.0 and above |
The following SIMD instruction set extensions are supported:
Architecture | Instruction set extensions |
---|---|
x86 | SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, AVX, FMA3, AVX2 |
x86 | AVX512 (gcc7 and higher) |
x86 AMD | same as above + SSE4A, FMA4, XOP |
ARM | ARMv7, ARMv8 |
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
xsimd
has an optional dependency on the xtl library:
xsimd |
xtl (optional) |
---|---|
master | ^0.7.0 |
8.x | ^0.7.0 |
7.x | ^0.7.0 |
The dependency on xtl
is required if you want to support vectorization for xtl::xcomplex
. In this case, you must build your project with C++14 support enabled.
The version 8 of the library is a complete rewrite and there are some slight differences with 7.x versions. A migration guide will be available soon. In the meanwhile, the following examples show how to use both versions 7 and 8 of the library?
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, xs::avx2> a(1.5, 2.5, 3.5, 4.5);
xs::batch<double, xs::avx2> 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)
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
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.