Radeon GPU Profiler 1.4

While the G in GPU stands for graphics, there are also popular SIMD programming models and associated APIs that map well to the GPU but mostly bypass their fixed graphics-specific logic. Khronos OpenCL™ is one such model and API that we support on Radeon GPUs, and the ability to profile OpenCL workloads in RGP is the big new feature in RGP 1.4.

OpenCL profiling

Most of the major RGP features that you’re used to using for profiling graphics workloads generated by Vulkan® and DirectX® 12 are there when profiling OpenCL applications, including the workload and barrier overviews.

You can see a pair of compute dispatches in the image above, separated by a barrier event that’s been highlighted and shown in the information pane on the right. You can see the first dispatch drain out of the machine before the subsequent dispatch starts up.

You can also see that the wavefronts are coloured by the SE they were dispatched to inside the GPU, highlighting how the GPU was filled up by the particular dispatch, and how it dynamically executes the workload by balancing waves across the available execution resources.

In the profile summary view you can configure it to see how your kernel enqueues progressed on the machine and were able to fill it up.

It’s worth nothing that RGP’s view of barrier sychronisation is different to how you drive them from the OpenCL API. You’ll see that they show up in RGP as CmdBarrier() s, rather than cl_event s or regular barrier() calls in your kernels. Check the documentation for more information.

Lastly, all of RGP’s familiar UI features and idioms from profiling graphics workloads are all there. Things like: you can zoom into specific workload areas, see CPU submission markers, interact with the timeline events as you see them to get more information on the rightmost summary pane, and selecting a region inside the timeline view will show you its duration.

Platform and hardware support

OpenCL profiling support is available on Windows 10, Windows 7 and Ubuntu 18.04.1 LTS on systems with Radeon RX Vega 56 and Radeon RX Vega 64 GPUs, and on AMD Ryzen processors with integrated Radeon Vega graphics.

Driver support

RGP 1.4 requires specific driver support to work correctly, so please sure you that you have at least 18.12.2 installed. In 18.12.2’s specific case that means using this special hotfix driver that we released alongside the regular 18.12.2 release. It fixes a couple of last minute issues we couldn’t roll in due to WHQL QA pressure.

Drivers from 18.12.3 onwards don’t have that special hotfix requirement so you’re good to go after that!

More information

As always, you can find out more about RGP, including links to the release binaries on GitHub and the full release notes list, on our product page. As always: please send us your feedback so that we can keep making RGP the very best developer-focused performance analysis tool for modern graphics work.

Your feedback is incredibly valuable to us and helps drive the RGP roadmap.