SIMD-Collection

A collection of highly optimized, SIMD-accelerated (SSE, AVX, FMA, NEON) functions written in C.

Includes Google Benchmark and Google Test support (C++).

Categories

• Dot product

  • using SSE, AVX, FMA and NEON intrinsics
  • for every data type combinaison: (u)int8, int16, int32, float, double
  • comparison with compiler auto-vectorized and naive implementations
  • optimization options: data alignement, vector size multiple, number of accumulators

• Sort 8-elements

  • based on optimal sorting networks
  • using SSE, AVX intrinsics
  • for every data type: int8, int16, int32, float, double
  • comparison with 'qsort' and 'std::sort' implementations (already ordered and random inputs)
  • optimization options: data alignement, early exit check

Benchmark results

See corresponding 'bench' sub-folders for graphs.

Dependencies

This project uses git submodule to include Google Benchmark and Google Test repositories:

$ git clone https://github.com/gaujay/simd_collection.git
$ cd simd_collection
$ git submodule init && git submodule update
$ git clone https://github.com/google/googletest lib/benchmark/googletest

Building

Support GCC/MinGW and MSVC in release mode only (see 'CMakeLists.txt' and 'src/Utils/compiler_utils.h').

On Linux/Unix:

$ mkdir build
$ cd build
$ cmake ..
$ make <target> -j

On Windows:

You can use command line or QtCreator for simplicity:

$ open 'CMakeLists.txt'
$ in 'Projects', select each kit 'Build' tab
$ under 'Build Steps->Targets', uncheck 'all' and check 'Current executable' (avoid building both x86/ARM targets)

License

Apache License 2.0