Heterogeneous Computing with OpenCL, [Paperback]

If this is what AMD has to provide developers no wonder OpenCL hasn't caught on yet.
First thing that caught my attention were the blatant spelling and grammatical errors. As though the book wasn't even proofread?
The first chapter starts off easy enough but the book warns that concurrent programming is hard and if you thought serial programming was hard you ain't seen nothing yet. They weren't kidding! The amount of code and steps just to write a basic hello program in OpenCL is ridiculous. Granted a lot of it is boilerplate code that only has to be written once and reused but the book doesn't even do a great job of making it any easier. The code samples also look like they were each written by a different person using C sometimes and C++ at other times.
Other than the nice illustrations and diagrams probably the best thing about this book is that is was so bad that it made me learn more about OpenCL than I ever cared to know by forcing me to Google for information missing from it and referring to other better resources. So that's why I'm even giving it 3 stars.
Like the OpenCL Programming Guide for Mac OS X. Apple's developer documentation which is from 2009 is seems to exactly the same thing as the first chapter in this book but Apple's is way better since they actually have code comments on just about every line walking you step-by-step through the code. So in the end this book showed me how good Apple's documentation really is. I always heard it was pretty good but didn't have an idea until now.
Later chapters seem to be a bunch of random case studies on what OpenCL is good at and some advanced ways to use it. The typical matrix, histogram and image processing examples you normally see in any parallel programming book, nothing that really stands out. Later chapters provide only minimal explanation of the code and that probably explains why the book is so short. Many of the details like OpenCL types, etc. are left for the reader to look up in some other book. If you are thinking of using this book as an introductory textbook they also leave the reader to come up with exercises/problems since there are absolutely no exercises included. The book also spends some time talking about various obscure loop unrolling methods which I'm not sure how relevant they are since this is now built into IDE's like XCode via auto-vectorization option.
As everyone has already stated the book is pretty AMD and Windows centric so none of the code worked on my Apple with MacOSX. Granted the book never said anything about Macs but isn't the whole point of OpenCL to be able to run on any hardware and OS? Also, the last several chapters deal exclusively with tools only available on Windows so not so useful on other operating systems.
I eventually did get the code examples to work on my Mac after some hard reworking of the code in the book. This was not as easy task, since depending on who you ask and where you look Macs either support or don't support OpenCL. Mac OS X, Mountain Lion, just released supports the latest version OpenCL 1.2 according to their documentation. Seems to me they should've worked closer with Apple since it seems that they are making more use of OpenCL than Windows which is introducing C++ AMP to compete with it.
I don't see OpenCL gaining much momentum as suggested in this book in the future until they make it easier to program than mpi,pthreads,not harder than competing alternatives and improve support across all platforms.
As the book states there are some embarrassingly parallel problems that OpenCL is very good at. Unfortunately, it is also embarrassingly hard to program and this book doesn't help much. Don't take my word for it as the majority of the OpenCL information in the book and better examples, video's, etc... can easily be found on AMD's developer website or any other number of places on the web.

Other reviews have uniformly raved about this book and in terms of content matter, I agree with them: It's a nice introduction to the Open CL language itself and the examples are well-chosen. There is a strong emphasis on use with AMD hardware, but that's to be expected given that the first two authors are AMD folks, and that AMD has adopted Open CL enthusiastically, and (I believe) discontinued all development of FireStream, it's earlier proprietary language.

This isn't the book to read if you're interested in any sort of comparison between CUDA, Open CL and MS's language; the purpose of this book is purely Open CL. This should be the book you've selected if you've already decided to use Open CL and to program natively. I mention this because if you're using a professional development tool, such as commercial compilers, the support is much greater in those tools for nVidia's CUDA: many of them will generate CUDA code without programming specifically for it. (This isn't a judgment of CUDA vs anything else, so much a function of CUDA having been around longer.)

The one thing that bothers me is that program code is printed in what appears to me to be gray rather than black text. I understand that this was done to emphasize that one is viewing program code, but given the use of a different font and indentation, I think that this would have been obvious without using a difficult-to-read gray text.

This may not bother you, but be prepared for it.

OpenCL is one of those things that I'd always wanted to get some experience with, so when I saw this book I figured "why not?". I'm glad I did. I found it straightforward and easy to understand, and I think this book is a great way to launch oneself into the world of OpenCL.

The book is an introduction, not a reference, but contains enough material and information to allow the reader to get a good start in OpenCL. Subject material is comprehensive. Chapters cover the basics of OpenCL and the API as well as providing explanations and discussions of how the design and organization of the computing hardware affect the design and performance of OpenCL code. There are lots of informative code examples, and four chapters of case studies, giving the reader some experience in seeing how hardware and program requirements affect the design of OpenCL code.

One minor nitpick: it's not obvious from the cover, but the book is actually from AMD. AMD was one of the first companies (perhaps *the* first) to offer an OpenCL API for their hardware. Most of the hardware-related discussion and examples focus on AMD products. This is fine for giving the reader an understanding of general differences in hardware architecture, but you'll need another source for details on hardware from other vendors.