problem regarding remote profiling on Jetson

I am using the cross-compilation method to develop and profile my application on Jetson dev-kit - as explained in the following link: http://devblogs.nvidia.com/parallelforall/nvidia-nsight-eclipse-edition-for-jetson-tk1/

In the simple case, I can run the visual profiler on the host to get the overall GPU usage without any problem. However, when I try to use the “Examine Individual Kernels” option to get more detailed information, I get non-stopping error signals regarding some fifo operations and the process doesn’t terminate unless I stop it manually. Here is one of such signals:

“gk20a gk20a.0: fifo_error_isr: channel reset initated from fifo_error_isr”

Can anyone please explain what is causing this problem and if there is any work around? or if there is a way to run the profiler directly on the board itself? I know that I can use “nvprof” on the board to get general performance metrics, but I don’t know how to do it to collect more detailed information (similar to what “Examine Individual Kernels” does)!

I would appreciate any suggestion.

On the board you can run nvprof with --analysis-metrics option to collect the events and metrics required for Visual Profiler Guided Analysis. You can then import this nvprof generated data in Visual Profiler. Please refer http://docs.nvidia.com/cuda/profiler-users-guide/index.html#import-nvprof-session for additional details.

For the remote profiling issue you see can you file a bug with a self-contained piece of code? It will also help if you can add following to the bug report-
a. Your L4T version.
b. Toolkit version on host and device
c. Host operating system

Thanks