Click here to Skip to main content
15,916,835 members
Articles / GPGPU

Introductory Tutorial to OpenCL™

Rate me:
Please Sign up or sign in to vote.
4.90/5 (11 votes)
8 Jul 2010Apache9 min read 164.1K   48   12
This article helps make OpenCL™ easier to understand and implement.

Benedict R. Gaster, AMD Architect, OpenCL™

OpenCL™ is a young technology, and, while a specification has been published (, there are currently few documents that provide a basic introduction with examples. This article helps make OpenCL™ easier to understand and implement.

Note that:

  • I work at AMD, and, as such, I will test all example code on our implementation for both Windows® and Linux®; however, my intention is to illustrate the use of OpenCL™ regardless of platform. All examples are written in pure OpenCL™ and should run equally well on any implementation.
  • I have done my best to provide examples that work out-of-the-box on non-AMD implementations of OpenCL™, but I will not be testing them on non-AMD implementations; therefore, it is possible that an example might not work as expected on such systems. If this is the case, please let me know via our Stream Computing forum, and I will do my best to rectify the code and publish an update.

The following "Hello World" tutorial provides a simple introduction to OpenCL™. I hope to follow up this first tutorial with additional ones covering topics such as:

  • Using platform and device layers to build robust OpenCL™
  • Program compilation and kernel objects
  • Managing buffers
  • Kernel execution
  • Kernel programming - basics
  • Kernel programming - synchronization
  • Matrix multiply - a case study
  • Kernel programming - built-ins

The "Hello World" program in OpenCL™

Here are some notes of caution on how the OpenCL™ samples are written:

  • OpenCL™ specifies a host API that is defined to be compatible with C89 and does not make any mention of C++ or other programming language bindings. Currently, there are several efforts to develop bindings for other languages (see the links at the end of this article), and, specifically, there has been a strong push to develop C++ bindings. In this and subsequent tutorials, I use the C++ bindings exclusively and describe OpenCL™ in these terms. See the OpenCL™ 1.0 specification for the corresponding C API. Alternatively, you can view the source for the C++ bindings to see what underlying OpenCL™ function is used, and with what arguments by the particular C++ binding.

  • OpenCL™ defines a C-like language for programming compute device programs. These programs are passed to the OpenCL™ runtime via API calls expecting values of type char *. Often, it is convenient to keep these programs in separate source files. For this and subsequent tutorials, I assume the device programs are stored in files with names of the form, where name varies, depending on the context, but the suffix does not. The corresponding device programs are loaded at runtime and passed to the OpenCL™ API. There are many alternative approaches to this; this one is chosen for readability.

For this first OpenCL™ program, we start with the source for the host application.

Header files

Just like any other external API used in C++, you must include a header file when using the OpenCL™ API. Usually, this is in the directory CL within the primary include directory. For the C++ bindings we have (replace the straight C API with cl.h):

#include <utility>

#define __NO_STD_VECTOR // Use cl::vector instead of STL version
#include <CL/cl.hpp>

For our program, we use a small number of additional C++ headers, which are agnostic to OpenCL™.

#include <cstdio>

#include <cstdlib>

#include <fstream>

#include <iostream>

#include <string>

#include <iterator>

As we will dynamically request an OpenCL™ device to return the "Hello World\n" string, we define it as a constant to use in calculations.

const std::string hw("Hello World\n");


A common property of most OpenCL™ API calls is that they either return an error code (type cl_int) as the result of the function itself, or they store the error code at a location passed by the user as a parameter to the call. As with any API call that can fail, it is important, for the application to check its behavior correctly in the case of error. For the most part we will not concern ourselves with recovering from an error; for simplicity, we define a function, checkErr, to see that a certain call has completed successfully. OpenCL™ returns the value CL_SUCCESS in this case. If it is not, it outputs a user message and exits; otherwise, it simply returns.

inline void
checkErr(cl_int err, const char * name)
    if (err != CL_SUCCESS) {
        std::cerr << "ERROR: " << name
                 << " (" << err << ")" << std::endl;

A common paradigm for error handling in C++ is through the use of exceptions, and the OpenCL™ C++ bindings provide just such an interface. A later tutorial will cover the use of exceptions and other optional features provided by the C++ bindings. For now, let’s look at the one remaining function, "main," necessary for our first OpenCL™ application.

OpenCL™ Contexts

The first step to initializing and using OpenCL™ is to create a context. The rest of the OpenCL™ work (creating devices and memory, compiling and running programs) is performed within this context. A context can have a number of associated devices (for example, CPU or GPU devices), and, within a context, OpenCL™ guarantees a relaxed memory consistency between devices. We will look at this in detail in a later tutorial; for now, we use a single device, CL_DEVICE_TYPE_CPU, for the CPU device. We could have used CL_DEVICE_TYPE_GPU or some other support device type, assuming that the OpenCL™ implementation supports that device. But before we can create a context we must first queue the OpenCL runtime to determine which platforms, i.e. different vendor’s OpenCL implementations, are present. The class cl::Platform provides the static method cl::Platform::get for this and returns a list of platforms. For now we select the first platform and use this to create a context. The constructor cl::Context should be successful and, in this case, the value of err is CL_SUCCESS.

    cl_int err;
    cl::vector< cl::Platform > platformList;
    checkErr(platformList.size()!=0 ? CL_SUCCESS : -1, "cl::Platform::get");
    std::cerr << "Platform number is: " << platformList.size() << std::endl;
    std::string platformVendor;
    platformList[0].getInfo((cl_platform_info)CL_PLATFORM_VENDOR, &platformVendor);
    std::cerr << "Platform is by: " << platformVendor << "\n";
    cl_context_properties cprops[3] = 
        {CL_CONTEXT_PLATFORM, (cl_context_properties)(platformList[0])(), 0};
    cl::Context context(
    checkErr(err, "Conext::Context()");            

Before delving into compute devices, where the ‘real’ work happens, we first allocate an OpenCL™ buffer to hold the result of the kernel that will be run on the device, i.e. the string "Hello World\n." For now we simply allocate some memory on the host and request that OpenCL™ use this memory directly, passing the flag CL_MEM_USE_HOST_PTR, when creating the buffer.

char * outH = new char[hw.length()+1];
cl::Buffer outCL(
    checkErr(err, "Buffer::Buffer()");

OpenCL™ Devices

In OpenCL™ many operations are performed with respect to a given context. For example, buffers (1D regions of memory) and images (2D and 3D regions of memory) allocation are all context operations. But there are also device specific operations. For example, program compilation and kernel execution are on a per device basis, and for these a specific device handle is required. So how do we obtain a handle for a device? We simply query a context for it. OpenCL™ provides the ability to queue information about particular objects, and using the C++ API it comes in the form of object.getInfo<CL_OBJECT_QUERY>(). In the specific case of getting the device from a context:

cl::vector<cl::Device> devices;
devices = context.getInfo<CL_CONTEXT_DEVICES>();
    devices.size() > 0 ? CL_SUCCESS : -1, "devices.size() > 0");

Now that we have the list of associated devices for a context, in this case a single CPU device, we need to load and build the compute program (the program we intend to run on the device, or more generally: devices). The first few lines of the following code simply load the OpenCL™ device program from disk, convert it to a string, and create a cl::Program::Sources object using the helper constructor. Given an object of type cl::Program::Sources a cl::Program, an object is created and associated with a context, then built for a particular set of devices.

std::ifstream file("");
checkErr(file.is_open() ? CL_SUCCESS:-1, "");

std::string prog(

cl::Program::Sources source(

    std::make_pair(prog.c_str(), prog.length()+1));

cl::Program program(context, source);
err =,"");
checkErr(file.is_open() ? CL_SUCCESS : -1, "Program::build()");

A given program can have many entry points, called kernels, and to call one we must build a kernel object. There is assumed to exist a straightforward mapping from kernel names, represented as strings, to a function defined with the __kernel attribute in the compute program. In this case we can build a cl::kernel object, kernel. Kernel arguments are set using the C++ API with kernel.setArg(), which takes the index and value for the particular argument.

cl::Kernel kernel(program, "hello", &err);
checkErr(err, "Kernel::Kernel()");

err = kernel.setArg(0, outCL);
checkErr(err, "Kernel::setArg()");

Now that the boiler plate code is done, it is time to compute the result (the output buffer with the string "Hello World\n"). All device computations are done using a command queue, which is a virtual interface for the device in question. Each command queue has a one-to-one mapping with a given device; it is created with the associated context using a call to the constructor for the class cl::CommandQueue. Given a cl::CommandQueue queue,kernels can be queued using queue.enqueuNDRangeKernel. This queues a kernel for execution on the associated device. The kernel can be executed on a 1D, 2D, or 3D domain of indexes that execute in parallel, given enough resources. The total number of elements (indexes) in the launch domain is called the global work size; individual elements are known as work-items. Work-items can be grouped into work-groups when communication between work-items is required. Work-groups are defined with a sub-index function (called the local work size), describing the size in each dimension corresponding to the dimensions specified for the global launch domain. There is a lot to consider with respect to kernel launches, and we will cover this in more detail in future tutorials. For now, it is enough to note that for Hello World, each work-item computes a letter in the resulting string; and it is enough to launch hw.length()+1, where hw is the const std::string we defined at the beginning of the program. We need the extra work-item to account for the NULL terminator.

cl::CommandQueue queue(context, devices[0], 0, &err);
    checkErr(err, "CommandQueue::CommandQueue()");

    cl::Event event;
    err = queue.enqueueNDRangeKernel(
         cl::NDRange(1, 1), 
    checkErr(err, "ComamndQueue::enqueueNDRangeKernel()");

The final argument to the enqueueNDRangeKernel call above was a cl::Event object, which can be used to query the status of the command with which it is associated, (for example, it has completed). It supports the method wait() that blocks until the command has completed. This is required to ensure the kernel has finished execution before reading the result back into host memory with queue.enqueueReadBuffer(). With the compute result back in host memory, it is simply a matter of outputting the result to stdout and exiting the program.

    err = queue.enqueueReadBuffer(
    checkErr(err, "ComamndQueue::enqueueReadBuffer()");
    std::cout << outH;
    return EXIT_SUCCESS;

Finally, to make the program complete an implementation for the device program (, requires defining the external entry point, hello. The kernel implementation is straightforward: it calculates a unique index as a function of the launch domain using get_global_id(), it uses it as an index into the string, hw, then writes its value to the output array, out.

#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
__constant char hw[] = "Hello World\n";
__kernel void hello(__global char * out)
    size_t tid = get_global_id(0);
    out[tid] = hw[tid];

For robustness, it would make sense to check that the thread id (tid) is not out of range of the hw; for now, we assume that the corresponding call to queue.enqueueNDRangeKernel() is correct.

Building and running

On Linux, it should be enough to use a single command to build the OpenCL™ program; for example:
gcc –o hello_world –Ipath-OpenCL-include –Lpath-OpenCL-libdir lesson1.cpp –lOpenCL

To run:
LD_LIBRARY_PATH=path-OpenCL-libdir ./hello_world

On Windows, with a Visual Studio command window, an example is:
cl /Fehello_world.exe /Ipath-OpenCL-include lesson.cpp path-OpenCL-libdir/OpenCL.lib

Let’s assume that OpenCL.dll is on the path, then, running

outputs the following string pm stdout:
Hello World

This completes our introductory tutorial to OpenCL™. Your feedback, comments, and questions are requested. Please visit our Stream forum.

Useful Links

The following list provides links to some specific programming bindings, other than C, for OpenCL™. I have not tested these and cannot vouch for their correctness, but hope they will be useful:

OpenCL and the OpenCL logo are trademarks of Apple Inc. used by permission by Khronos.


This article, along with any associated source code and files, is licensed under The Apache License, Version 2.0

Written By
United States United States
Member of AMD's Developer Outreach team.

Comments and Discussions

QuestionGreat work Pin
Lutosław24-Jun-20 12:26
Lutosław24-Jun-20 12:26 
QuestionWhere can I find the complete code? Pin
foadsf31-May-18 16:18
foadsf31-May-18 16:18 
QuestionAnd now a less convoluted solution Pin
D. Infuehr18-Jul-17 1:28
D. Infuehr18-Jul-17 1:28 
QuestionQuestion Pin
Member 104475937-Aug-14 14:42
Member 104475937-Aug-14 14:42 
QuestionVery good Pin
ed welch30-Apr-13 3:51
ed welch30-Apr-13 3:51 
GeneralMy vote of 5 Pin
sundareswaran.senthilvel8-Sep-12 8:13
professionalsundareswaran.senthilvel8-Sep-12 8:13 
QuestionFix the formating Pin
arjo12921-Jun-12 16:38
arjo12921-Jun-12 16:38 
QuestionNext tutorials Pin
Shnkc10-Nov-11 7:47
Shnkc10-Nov-11 7:47 
GeneralUsefull article Pin
kj-dev31-Aug-10 8:55
kj-dev31-Aug-10 8:55 
QuestionWhat *is* it? Pin
Lucas Vogel9-Jul-10 6:27
professionalLucas Vogel9-Jul-10 6:27 
AnswerRe: What *is* it? Pin
pnolte649-Jul-10 7:07
pnolte649-Jul-10 7:07 
AnswerRe: What *is* it? Pin
GPUToaster™11-Jul-10 8:07
GPUToaster™11-Jul-10 8:07 

General General    News News    Suggestion Suggestion    Question Question    Bug Bug    Answer Answer    Joke Joke    Praise Praise    Rant Rant    Admin Admin   

Use Ctrl+Left/Right to switch messages, Ctrl+Up/Down to switch threads, Ctrl+Shift+Left/Right to switch pages.