C++ High Performance

Using a custom kernel in Boost Compute

As mentioned earlier, this is not intended as a course in OpenCL, but we will now have a look at how to move away from the algorithms and abstractions of Boost Compute and invoke regular for-loops where elements are read at random positions in an array. In other words, the following example is almost bare OpenCL, rather than Boost Compute.

As OpenCL evolves from a computer graphics background, it does contain a multitude of operations for handling textures and filtering, but in order to see the resemblance of regular C++ code, we will use traditional vectors to operate on.

Remember that a GPU is very good at executing many tasks in parallel, and OpenCL correspondingly needs to be informed of what exactly ...