A trivial "Hello world" style example is located in this namespace. It defines a kernel to factor two float arrays and shows the required steps to run this on either CPU or GPU devices (by default the most performant is chosen automatically):

;; chooses max performance device
lein trampoline run -m thi.ng.simplecl.test.hello
;; or force usage of given device type
lein trampoline run -m thi.ng.simplecl.test.hello gpu
lein trampoline run -m thi.ng.simplecl.test.hello cpu

Running the command below will run a basic 2d cloth simulation with 40,000 particles & 80,000 springs and generate an image sequence of 120 PNGs in the /export sub-directory.

lein trampoline run -m thi.ng.simplecl.test.verlet

The image sequence can then be converted into an MP4 using this `ffmpeg` command:

ffmpeg -r 30 -f image2 -start_number 5 -i "export/verlet-%04d.png" \
 -vcodec libx264 -preset slow -crf 23 -pix_fmt yuv420p \
 -f mp4 -threads 0 verlet.mp4 -y

This demo is also showcasing usage & integration of the structgen partner library to ease the task of aligning nested Clojure data structures in an OpenCL compatible way (mapping to OpenCL structs).

See thi.ng/raymarchcl for full details.

A sneak peak of an upcoming example kernel pipeline to:

1. compute a 2d strange attractor
2. apply a 2D distortion filter to all points
3. map the result points onto the surface of a sphere and
4. apply a 3D distortion filter to all points
5. do a standard 3d > 2d camera-screen projection

More examples & documentation will be added ASAP.