Check out the final blog in the series, on Fine Grain Shared Virtual Memory.
As for every blog in this series, we discuss the example code in the APP SDK completely so you get a solid understanding of how it works. The blog will point you to all the necessary resources so you can be sure you're on compatible hardware, point you to the right driver, and link you to the Beta APP SDK.
By way of summary, since this is the final installment, to familiarize you with the new features of OpenCL 2.0 we published blogs on:
I hope you have found and continue to find these useful. I can look at page hits, and see that these blogs have already been visited 6,000 times. That’s just a number, I think a nice number, but still I’d rather hear from you directly.
Were these useful? How did we do? What do you like and what you think we can do better. Would you like to see more of the same on other technologies? Perhaps, for example, for the heterogeneous system architecture? Let me know. Your feedback guides my work.
We are starting week 3 in our migration from CUDA to OpenCL 2.0. CodeXL is very nice for debugging OpenCL 1.2, but we want to use and debug SVM in OpenCL 2.0.
>>> Were these useful? How did we do?
The examples have helped a little. We have gotten the SVMBinaryTreeSearch to run, but we are not convinced that it is working. We would like to step through the OpenCL 2.0 kernel, but CodeXL only works at the API level. How can we step through and debug OpenCL 2.0 kernels? Is there a Beta version of CodeXL or some other tool that we can use?
Thanks for the feedback. I'll be in a room with the author for the next two days, so if you want something brought to Prakash's attention, I can make sure that happens.
>>> if you want something brought to Prakash's attention, I can make sure that happen
My team likes CodeXL on OpenCL 1.2 and we would *love* to see the same functionality with CodeXL & OpenCL 2.0 If it makes the task easier, we would be happy if CodeXL only worked on a subset of OpenCL 2.0 Partial functionality would let us use it immediately on some of our code (something is infinitely better than nothing), and it would give the developers of CodeXL time to work on the tricky new parts of OpenCL 2.0.
We are porting from CUDA to AMD/OpenCL 2.0 *only* because of SVM, and we are disappointed that we cannot do any debugging on the OpenCL 2.0 code. I think that our productivity is 1/10th what it could be with a functional (or even semi-functional) debugger.
Thanks for the feedback. Actually, the best person for this isn't the article author, but I'll see he hears it too. I'll get this to the CodeXL lead engineer.
>>> I'll get this to the CodeXL lead engineer.
We are coding for a patented 10-year development project that we will begin promoting later this year. The system relies heavily on GPU programming, and we appreciate anything you can do to improve OpenCL 2.0 debugging on AMD APUs in Linux. I'm bringing another AMD development system online today. Thanks for your help.
CodeXL 1.6 provides only API-level debugging for OpenCL 2.0. There are several AMD teams working together with the CodeXL team to add OpenCL 2.0 kernel debugging. This is a top priority feature that we're eager to provide. Company policy prevents the disclosure of future release dates so all I can say is that this is a big feature and we're already working on it.
Development builds can be provided to NDA partners only. For further information about NDA partnership please contact firstname.lastname@example.org