Greetings, I would like advice if any of the current benchmark harnesses would be suitable for my use case. None of them looked spot-on to me, so I would appreciate your thoughts.
I am writing a high-performance WebGPU primitive library (think: sort, reduce, scan, etc.). I am hoping to do performance tests on this library across a variety of platforms, with the specific goal that I expect the fastest implementation of a particular primitive may differ across platforms, and I would expect that the final library will reflect this. Telemetry looks like it may be a good fit. I would like to gather runtime statistics as well as more detailed information (e.g., GPU timestamps for particular operations) in my tests.
I looked at the Chrome benchmark harnesses and none of them appear to be clearly the perfect options, so wanted to ask your opinion if my use case fits well into any of these.
Thanks!
JDO