• Stars
    star
    307
  • Rank 136,109 (Top 3 %)
  • Language
    C++
  • License
    GNU General Publi...
  • Created over 9 years ago
  • Updated about 9 years ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

The easiest way to get started with OpenCL!

EasyOpenCL

No exposed low-level C, just the C++ standard library

Features

  • Focus on programming the GPU instead of messing about with long C-style OpenCL function calls and manual memory management.
  • CMake support for Linux and Mac - No more linking problems when you have installed the correct driver.
  • Support for scalar values: pass additional structs to your kernel, eg. transformation matrices or custom constants.
  • Chain kernels together in order to create a true pipeline on your GPU in which kernels can depend on multiple others. (example/main.cpp)
  • Human readable OpenCL errors for easy debugging and teaching of the OpenCL basics.

Overview: it's this easy!

// example/simple.cpp
try {
  EasyOpenCL<float> framework (SHOW_DEBUG);
  
  //Register the kernel with the framework
  auto& square = framework.load("squarefloat");
  
  //Bind the input buffer (initial values from a std::vector) and the output buffer
  square.bindInput(0, std::vector<float> { 1.1, 2.2, 3.3, 4.4, 5.5, 11.0 });
  square.bindOutput(1);
  
  //Evaluate the kernel and display the output
  square.evaluate();
  square.showBuffer(1); // [ 1.21, 4.84, 10.89, 19.36, 30.25, 121 ]
} catch (std::exception& e) { std::cerr << "Error: " << e.what() << std::endl; }
// kernels/squarefloat.cl
__kernel void squarefloat(__global float* input, __global float* output)
{
  int i = get_global_id(0);
  output[i] = input[i] * input[i];
}

Getting started

All platforms:

  • Update your graphics drivers
  • Install the drivers with OpenCL support (NVIDIA CUDA Toolkit, AMD APP SDK or Intel OpenCLâ„¢ Code Builder)
git clone https://github.com/Gladdy/EasyOpenCL.git
cd EasyOpenCL
mkdir build && cd build
cmake ..
make
./simple

TODO:

  • High priority

    • More examples - image processing, deep learning (matrix operations) and a renderer/raytracer
    • Asynchronous kernel calls + benchmarks of asynchronous vs synchronous kernel calls
    • Cleaning up the framework, getting public/private right + the different constructors
  • Low priority:

    • Automatic generation of local and global work group sizes
    • Detect circular dependencies of kernels
  • Mehh whenever I have the time:

    • .dot output of your current kernel dependency graph
    • Automatic generation of stub kernel (.cl) files based on the kernel links specified.

Thanks to: