OpenCL Programming Guide [Anglais] [Broché]

OpenCL Programming Guide is the 2nd book (to my awareness) being published, which deals with the new and exciting standard by the KHRONOS Group: OpenCL. The goal of this book is to provide the reader with an extensive walkthrough of the standard, providing explanations to complement the standard's specs. The authors of the book dim it "a pragmatic guide for people interested in writing code", and that it is.

The book is at its first edition, and it shows. Throughout the book there are typos, and what can only be explained as 'copy & paste' originated mistakes. Some of the code samples contain generic errors such as memory leaks or incorrect remarks, and some of the figures simply do not convey the intended concept, or are erroneous. The majority of errata I personally reported dealt with these types of errors, which are arguably acceptable (for a first edition) as they are not regarding the focus of the book, however, the book also contains some errata that does touch the actual focus, like an incorrect explanation (e.g. reported issue #14 on pg.132, and reported issue #4 on pg.65), or incorrect usage of returned information (e.g. reported issue #8 on page 88).

On the other hand, the book does provide good insight on a vast portion of the standard. Although it claims to cover the entire spec, the level of this coverage is inconsistent and in some aspects completely lacking (e.g. the explanation of clEnqueueTask() could have been accompanied with a concise example, but in turn ended up as a short sub-section). On the portions with most interest, i.e., OpenCL's support for data-parallel algorithms, the book does provide extended information, and adds to the OpenCL specs, by clarifying the concepts.

The 2nd part of the book, which was added rather close to the final release of the book (from the eyes of a SafariBooksOnline RoughCuts reader), provides 9 case studies of OpenCL usage. Some of these are purely pedagogic (e.g. chapter 15), but some provide more real-world examples of how OpenCL can be used, and optimized (especially for a GPU). These add another dimension to the book, and contribute to its relevancy.

On a closing note, I do think that the book is worth the while. It is currently the best option besides reading the specs, to learn the OpenCL APIs and OpenCL C programming language, and despite the shortcomings I've mentioned, it does manage to provide the gist of OpenCL, and add insight to the standard.

-----------------------------

Proper disclosure: OpenCL and GPU programming is what I do for a living.

The heavy book with the thin green cover is a "must read" book for beginning OpenCL developers. Experienced OpenCL developers could find some chapters interesting too. The book can be used both as day to day reference book and as a manual.

The detailed introduction (chapter 1) enables the reader to understand the essentially important thing - the design of the technology (OpenCL standard v.1.1). It is considered in the book from 4 sides which are reviewed in the connection with each other: platform, execution, memory and programming models. If you are the novice in OpenCL, I urgently recommend you to read introduction firstly.

The narration continues with the "Hello World" example (chapter 2). It is big enough and divided into several subsections. Each of them we can consider as simple steps to create a real OpenCL application. I like it. After 1st an 2nd chapters, I guess, it is hard enough to NOT understand how to write your own OpenCL software. Sure, in this case you need to use some reference book (or the text of OpenCL standard), but from this point you are able to do it without a special assistance.

The next several chapters represent the mentioned by me so called "reference book". In particular, chapter 3 gives a detailed account of OpenCL platform, devices and context; chapters 4 ad 5 represent in detail OpenCL C programming language (operators, types, keywords and other things which you could find in any other programming language); chapters 6, 7, 8 and 9 describe in detail remained components of the technology: kernels, memory objects, images, events and so on. Well, I've concluded that chapters from 3 to 9 could be use as a real day to day reference book. For me it is quite useful.

To tell you the truth, the chapters 10 and 11 and 13 I've skipped, because interoperability with D3D and OpenGL, and embedded development with OpenCL are not interesting for me, but for somebody it could be useful too. The chapter 12 concerns the details of the C++ wrapper API: all staff in cl.hpp header file are briefly but well described.

The second and most interesting for experienced developers part of the book describes real cases of OpenCL using. Here 9 OpenCL-implementations of the specific algorithms are well described. It looks like all of them have been taken from the real author's software. Good. Some of the algorithms are illustrated with explicative figures. I think this part of the book with case studies could be a separated book. I just want to noted that I would glad to see more case studies, especially with vectors and matrices operations.

In addition, I want to mention chapter 20 "Using OpenCL with PyOpenCL". In recent times adopting of python in high performance applications is more and more wide. There are many techniques to improve the performance of python applications, but using GPGPU in intensive computing parts of application could become real help for python developers.

Also, it could be interesting to add chapter concerning using WebCL. Sure, nowadays WebCL is not so widely used, but the next year (2012) promises to be an year of WebCL. I hope.

To sum up, what I like in this book: organization of material, good introduction in OpenCL, detailed "Hello, world" chapter, appropriate case studies, detailed "reference book" part, print quality. Dislike: too soft cover, the lack of WebCL chapter, the number of case studies could be bigger.

Prior to this book I have never heard of OpenCL. I've heard of OpenGL used for graphics, but had never imagined that such a push to use similar parallel-programming techniques in graphics existed for applications outside of the realm of graphics.

From the beginning in chapter 1, the book does a great job introducing the history of how OpenCL came about and it's significance it'll play as the world of computers turns to parallel systems. They get you up to speed on the various terms and techniques used in OpenCL.

It was helpful that as you read into chapter 2 (the hello world example), they go through the process of getting your development set up. It was refreshing to read a programming book that didn't choose to only go through one OS setup but detail Windows, Mac OSX, and Linux. Many programming books that I've read in the past only chose to pick one, at most two platforms and most of the time very spotty in the way they described their setup for the development environment. Which led me to do many searches on Google before diving into the meat of the book, which is the programming.

The quality of the book is definitely written in a form of a reference book. It provides various examples on parallel programming. I could see this book serving as a textbook in a upper level computer science special topics course. As an undergraduate student myself in computer science I found the topic of OpenCL very intriguing but the book a bit tough to follow past the first few chapters.