OpenCL.tex

\chapter{OpenCL}

In 2006 NVIDIA announced CUDA architecture along with APIs based on C language
dedicated for GPU. OpenCL specification was published in 2009 and, unlike CUDA,
is implemented in hardware produced by different companies, not only by NVIDIA.



Matthew Scarpino's A Gentle Introduction to OpenCL,:
\url{https://freecontent.manning.com/wp-content/uploads/a-gentle-introduction-to-opencl.pdf}

In the era of heterogenous computing, one of the main advantage of OpenCL is portability.

OpenCL defines a set of standardized APIs for evoking kernel/function to run on
a computing architecture, e.g. CPU, GPU, FPGA.
OpenCL-coded routines, called kernels, can execute on GPUs and CPUs from such
popular manufacturers as Intel, AMD, Nvidia, and IBM.


New OpenCL-capable devices appear regularly.
There are efforts underway to port OpenCL to embedded devices, digital signal
processors, and field-programmable gate arrays.

IMPORTANTLY: a single application can dispatch kernels to multiple devices at
once.

CHALLENGE: OpenCL has one significant drawback: it’s not easy to learn

\begin{itemize}
  \item   OpenCL isn’t derived
from MPI or PVM or any other distributed computing framework

   \item  Its overall operation resembles that of NVIDIA’s
CUDA, but OpenCL’s data structures and functions are unique.
\end{itemize}

Like a CUDA-enabled app, you need to write host-code (which run on x86 CPU), and
then write OpenCL-based kernels (which can run on an OpenCL-capable device).

\section{Host code}

In developing an OpenCL project, the first step is to code the host application, which can be in C or C++.

IMPORTANT:
every host application requires five data structures: 
\begin{verbatim}

cl_device_id, cl_kernel,
cl_program, cl_command_queue, and cl_context.

\end{verbatim}

At first, it  is hard to remember these structures and how they work together