Loops

Loops is a minimalistic, fast, cross-platform C++ JIT engine with register allocation, mainly intended for generating vectorized code.

License

Loops is an open-source project distributed under the permissive Apache 2 license.

Compatibility

Supported architectures:

• Aarch64. NEON vectors.
• x86_64. AVX2 vectors.
• Risc-V64. No vector support yet.

Supported operating systems:

• Linux: Aarch64, x86_64, Risc-V64.
• MacOS: Aarch64.
• Windows: x86_64.

Of course, many compiler versions can build the library. There we are enumerating only verified compilers versions.
Verified supported compilers:

• GCC 12.2.0, 13.2.0
• Clang 15.0.0
• MSVC for Visual Studio 2019, Visual Studio 2022

How to build

Linux and Macos

You need git, CMake, and a C++ compiler installed on your system.

1. Clone the repository and enter the directory:

git clone https://github.com/4ekmah/loops
cd loops

2. Create a build directory and configure the project:

mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ../

3. Build the project:

make -j8

Windows

You need VSCode, CMake, and Visual Studio installed, or Visual Studio Build Tools(https://aka.ms/vs/stable/vs_BuildTools.exe). You also need to appennd the CMake extension to VSCode.

1. Clone the repository using Git Bash:

git clone https://github.com/4ekmah/loops

2. Open the loops directory in VSCode. There you can configure, build, and run examples and tests.

Samples and tests

There are two simple runnable examples:

1. minmaxloc - finds the position and value of the maximum element in an array of integers.
2. bresenham - an implementation of the Bresenham line drawing algorithm..

Both examples use only scalar instructions. For SIMD examples, see the test code:

1. testregress - a large set of regression tests demonstrating different ideas and approaches.
2. testperf - see next paragragh.

Loopslayers

Loopslayers is a small library of generators for neural network kernels.

Currently includes:

1. Depthwise convolution with padding. Strides 1 or 2. No dilations. Activations: none, ReLU, ReLU6, LeakyReLU. AArch64 only.
2. Max pooling with padding. Strides 1 or 2. No dilations. Activations: none, ReLU, ReLU6, LeakyReLU. Works on AArch64, and partially (stride 1 and FP32 only) on Intel.

testperf checks kernels supported on the current architecture and measures execution time for several cases used in well-known neural network models.

How to append the library to your project

The built library can be linked to another project via:

1. Adding the include path <loops_dir>\include
2. Adding the library search path <build_dir>\
3. Linking <build_dir>\libloops.a and, if needed, <build_dir>\libloopslayers.a (or .lib if your system is Windows)

Manual

More information about basic concepts, interfaces, usage, and approaches can be found in the reference manual: "docs/Loops_Manual_en.md"