May 26, 2018
C++ library to use SIMD vectorization
Inastemp provides a set of C++ classes to make vectorization with intrinsics easier. It aims at developing numerical kernels by separating the algorithm from the hardware target. Inastemp comes with several examples and patterns related to widespread use-cases.
- The following x86 SIMD types are currently supported
- SSE3, SSSE3, SSE4.1, SSE4.2, AVX, AVX2, AVX512-KNL, AVX512-SKL
- The following Powere PC SIMD types are currently supported
- Power-8 Altivec/VMX
- arithmetic operators */+- are provided
- CPU capacities are detected automatically during the CMake stage
- The compiler capacities are detected automatically during the CMake stage
- The library purely contains of headers, no linkage is necessary.
- CPU detection may use Intel SDE
- Unit-tests may use Intel SDE
- Fast intrinsic exp function if not supported natively by the compiler
- Explicit branches vectorization several patterns which represent many applications are demonstrated