OpenCL on FPGAs for GPU Programmers

Data Parallelism and Kernels Data parallelism is a form of parallelism across multiple processors that is achieved when each processor performs identical tasks on different pieces of distributed data. Data-parallel portions of an algorithm are executed on devices as kernels, which are C functions with some restrictions and a few language extensions. The host launches kernels across a 1D, 2D, or 3D grid of work-items to be processed by the devices. Conceptually, work-items can be thought of as individual processing threads, that each execute the same kernel function. Work-items have a unique index within the grid, and typically compute different portions of the result. Work-items are grouped together into work-groups, which are expected to execute independently from one another.