GPU General-Purpose Computation
	A note:  nVidia Tesla cards
	Popular Methods:
		Hacks with OpenGL, feedback or perhaps compute shaders
		CUDA
		OpenCL
		There are other methods of course, but these are "popular"
			According to me
	CUDA vs. OpenCL
		CUDA is first to support features of nVidia hardware
		If you are programming an nVidia card, it is better
		Very well supported by nVidia
		Not supported on any other vendor's hardware!
			No ATI/AMD, the crypto mining value champions
		OpenCL:  Khronos backing, works everywhere
			May not expose certain nVidia card features
	OpenCL versions:
		1.2 (current on nVidia), 2.0, 2.1, 2.2
		Many 2.0 features are supported on nVidia
		Probably a driver issue rather than hardware
		Later versions support C++ on the GPU
		Can be used together with OpenGL
			More tempting before compute shaders

OpenCL:
	Instead of a rendering pipeline, we have a kernel
	A kernel encapsulates a program
	Really, this is not that different from a compute shader
	Language:  C dialect
		Kinda like GLSL, but not exactly
		Khronos would like to merge OpenCL into Vulkan
		That way, you only have to learn this stuff once

OpenCL demo, using functions (all in man except as noted):
	clGetPlatformIDs
	clGetDeviceIDs
	clGetDeviceInfo
	clCreateContext
	clCreateCommandQueueWithProperties
		Unlike the others, this one doesn't have a manpage
		Read clCreateCommandQueue instead
	clCreateBuffer
	clCreateProgramWithSource
	clCreateKernel
	clSetKernelArg
	clEnqueueNDRangeKernel
	clEnqueueReadBuffer
	man on isoptera is for opencl-1.2

A blog tutorial that might be helpful:
	https://anteru.net/blog/2012/11/03/2009/index.html
	It is C++, and not just in a trivial way