How to build a PyOpenCL application
As for programming with PyCUDA, the first step to build a program for PyOpenCL is the encoding of the host application. In fact, it is performed on the host computer (typically, the user's PC) and then it dispatches the kernel application on the connected devices (GPU cards).
The host application must contain five data structures:
Device: This identifies the hardware where the kernel code must be executed. A PyOpenCL application can be executed on CPU and GPU cards but also in embedded devices, such as Field Programmable Gate Array (FPGA).
Program: This is a group of kernels. A program selects the kernel that must be executed on the device.
Kernel: This is the code to be executed on the device. A kernel is essentially a C-like function that enables it to be compiled for execution on any device that supports OpenCL drivers. A kernel is the only way the host can call a function that will run on a device. When the host invokes a kernel, many work items start...
