CUDA driver-level debugging API?

One of the feature bullets for CUDA 3.0 was debugging support for driver API applications. With this in place, I am now investigating how this might be used in PyCUDA. I’ve dug around the CUDA Toolkit package a bit, but I couldn’t find anything that would point me in the right direction, in the way of header files or some such.

Also, does something like this exist for OpenCL?

Any clue on where to look would be much appreciated!

Thanks,
Andreas

Uh, I think it just means that you can debug CUDA applications that use the driver API with cuda-gdb. There’s not any API for it.

Uh, I think it just means that you can debug CUDA applications that use the driver API with cuda-gdb. There’s not any API for it.

Ok–I guess I was misreading this then. Last year at GTC, I spoke with a few Nvidia folks who said that there would be a third-party debugging API.

Thanks for the reply,
Andreas

Ok–I guess I was misreading this then. Last year at GTC, I spoke with a few Nvidia folks who said that there would be a third-party debugging API.

Thanks for the reply,
Andreas

It’s coming eventually, but not yet.

It’s coming eventually, but not yet.

Incidentally, if full context recovery after a failure is somehow out of the question, some way to read memory contents from a failed context would already help by making a simulated recovery possible without constantly mirroring all data to ordinary memory…