An introduction to OpenCL (Lutecia Week 2020 TP39).

Course from https://github.com/amusant/tpt39.

The target was supposed to be an odroid xu4 (with an ARM Mali T628 GPU), but I changed a few things to have it run on my own computer (whose GPU is an Nvidia GTX 1070).

It should run on any nvidia card and any Linux with Nvidia and OpenCL libs installed.

If you want to run it on another device you probably only need to change the libs in the makefile. You can also change the version of C++ used.

Each subdirectory has a Makefile, running make compiles and run the .cpp file (which itself compiles and run the .cl file).

Writes "hello world !", it's mainly just to see how to setup an OpenCL program.

Adds two vectors together, compares times between CPU and GPU.

If MAPPED is defined, the memory used is mapped from GPU space to CPU space.

If not, the memory is copied.

Multiplies two matrix together, and compares times between CPU and GPU.

If GROUPS is defined, it uses groups to have cache speedup.

Video filtering calculations in GPU (opencv stuff) to modify a video.

In particular write a function to highlight contours in an image and one to blur an image.

You can define GROUPS and MAPPED to use groups and to map memory instead of copy it.

Compare runtimes between OpenCV and home made OpenCL functions.