Monitor active warps and threads during a divergent CUDA run

I implemented some CUDA code. It runs fine but the alogrithm inherently produces a strong thread divergence. This is expected.

I will later try to reduce divergence. But for the moment I would be happy to be able to measure it.

Is there an easy way (prefereably using a runtime API call or a CLI tool) to check how many of my initially scheduled warps and/or threads are still active?

from Recent Questions - Stack Overflow https://ift.tt/r9u2LYw8M
https://bit.ly/3GblJNq