Simple examples of OpenCL code used to learn heterogeneous and GPU computing with OpenCL.
* -- reproduced to work in 2025 on AMD GPU
- *
add_numbers: add a list of numbers together. Includes detailed error handling which makes the code harder to read and understand - *
square_array: computes array^2 (I am playing mostly with this one) - *
sum_array: sums two arrays cf4cl: testing OpenCL C wrapper- *
Hello_World: OpenCL "Hello World" by Apple - *
mandelbrot: my attempt at a simple Mandelbrot set calculation N-BodySimulation: Apple's N-body simulator which clearly illustrates the speedup gained by using the GPU. Requires XcodeRayTraced_Quaternion_Julia-Set_Example: Apple's ray-traced quaternion Julia set renderer. Requires Xcoderng: Illustrates how to generate random numbers in the host and in the GPU using the libraryclRNGauger: generates random cosmic rays on an isotropic sky- *
waste: compute cycle waster
The examples that clearly demonstrate the computational advantage of using a GPU for processing are N-BodySimulation, RayTraced_Quaternion_Julia-Set_Example (both developed by Apple programmers) and auger. For auger, I got impressive speedups of >200x compared to a serial code on the CPU.
mkdir build && cd build
cmake ..
makeEach target is placed in its corresponding subdirectory under build/. Run
executables from that subdirectory so the .cl kernel files are found:
cd build/square_array && ./squareExamples requiring clRNG
(auger, rng) are built automatically when clRNG is detected; they are
silently skipped otherwise.
Each example also has its own Makefile. Build from within the project directory:
cd <example> && make # build OpenCL version
cd <example> && make run # build and runExceptions: N-BodySimulation uses an Xcode project. RayTraced_Quaternion_Julia-Set_Example has a Makefile and CMakeLists.txt. cf4cl uses make.sh.
To learn about your OpenCL devices, try:
clinfo
To install clinfo on MacOS:
brew install clinfo
- similar repo with CUDA examples
