SIMD Module

SIMD Module Overview

The HIPP::SIMD module includes the templated version of SIMD instrunctions. Currently it only supports the Intel x86 platform where AVX vector instructions are implemented (see Intel’s Guide and check the Wiki page to see the processors that supports SIMD). Some of the operations need AVX2, AVX512F, AVX512VL supports.

All the definitions in this SIMD module can be accessed by including the header file <hippsimd.h>. All SIMD classes are defined within the namespace HIPP::SIMD. This is a pure-header module, no library is needed to link. Refer to the manual of the compiler you use to see how to enable SIMD extensions. For example, -mavx2, -mavx512f, -mavx512vl enables the AVX2, AVX512F and AVX512VL extension in GCC, respectively. To achieve the best performance, optimization flags need to be enabled in the compiling process, e.g.:

$ g++ -std=c++17 -O3 -mavx2 src.cpp

To avoid confusion, we list the conventions used in this API reference:

  • The example code that uses C++ standard library may ignore the std:: namespace specifier (namely, assume a global using namespace std is used). This is not good in practice, but makes the example code more compact.

API References