Access to internal GPU performance counters Does CUDA 2.0 add support for these?


Is there a way to access the internal GPU performance counters from kernel code? I saw a forum post by Mark saying that these will be added in the future? Has this already happened?


Older versions of the PTX manual listed the relevant registers (though not how to use them). Newer versions don’t even mention them anymore. They’re still there, and exposed via the relatively new Profiler. (Was that what Mark was referring to?) But I don’t know if anyone is planning on making them accessible via intrinsics/device functions. I rather doubt it.