xsimd Documentation

xsimd provides C++17 wrappers for SIMD (Single Instruction, Multiple Data) intrinsics. It offers a unified abstraction for different microprocessor vendors and compilers, allowing developers to write vectorized code that is portable across architectures.

The library is built around:

• Batch Types: Parameterized vector types.
• Operations: A set of arithmetic, data transfer, and mathematical functions designed for batch processing.

To use xsimd in your own project, use find_package to locate the configuration files and link against the xsimd target. Linking to the xsimd target automatically handles include directories and requirements like the minimum C++ version.

find_package(xsimd REQUIRED)
target_link_libraries(myproject PRIVATE xsimd)

To build and run tests, you need cmake. You can install cmake via conda if it is not present on your system.

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

You can build and install xsimd directly from the source code using cmake. Ensure you provide a valid CMAKE_INSTALL_PREFIX.

cmake -D CMAKE_INSTALL_PREFIX=your_install_prefix .
make install

Version 8.x introduces significant API changes due to a redesign of batch types. The primary change is that xsimd::batch is now parameterized by a type (e.g., double) and an optional architecture (e.g., avx512).

Key Changes:

• Batch Parameterization: Use xsimd::batch<T, arch> to target specific architectures, or xsimd::batch<T> to let the library select the best available architecture.
• Batch Loading: xsimd::batch<T>::load* methods are now static. You can no longer update an existing batch via load; use the assignment operator instead.
• Indexing: operator[] has been replaced by .get(size_t). Note that .get() performs a register load on every call and should be avoided in performance-critical loops.
• Complex Types: Loading a batch of xtl::xcomplex<T> now returns an xsimd::batch<std::complex<T>>.

Use the Spack package manager to install and load xsimd.

spack install xsimd
spack load xsimd

The documentation is built using doxygen, sphinx, and breathe. You can install breathe via pip or conda.

1. Install breathe: pip install breathe or conda install -c conda-forge breathe
2. Navigate to the docs directory.
3. Run make html.

make html

If you are on a Windows platform and need to use nmake, specify the generator during the configuration step.

cmake -B build/ -G "NMake Makefiles" -D CMAKE_INSTALL_PREFIX=/path/to/install/dir
cmake --build build/
cmake --install build/

Use the spack package manager to install and load xsimd.

spack install xsimd
spack load xsimd

When loading or storing values from contiguous dynamically allocated memory into SIMD registers, use xsimd::batch::load_unaligned and xsimd::batch::store_unaligned.

To improve performance, use xsimd::aligned_allocator with STL containers to ensure memory is aligned, allowing for faster memory transfer operations.