how to profile cycle program with nvprof.

when ctrl+C to kill my program,i will get the following messages:
“==28033== Error: Internal profiling error 4087:15.”

using “nvprof --profile-child-processes --timeout 60 --kernels “ewMulti” ./start_cmd.sh” to solve this error,i will get this:

==27893== Execution timeout, stopping the application…
==27893== Profiling application: ./BTMiner_NebuTech -url btm.f2pool.com:9221 -user bm1qj3v6qfj7dn9glgrp3aretgn6tcwdrm9n8ujla6 -RUN
==27893== Profiling result:
No kernels were profiled.
No API activities were profiled.
==27893== Warning: Some profiling data are not recorded. Make sure cudaProfilerStop() or cuProfilerStop() is called before application exit to flush profile data.
Miner exit normally …

Hi,

Does your script (start_cmd.sh) launch multiple processes or only a single process?
Note that nvprof cannot profile processes that fork() but do not then exec().
https://docs.nvidia.com/cuda/profiler-users-guide/index.html#multiprocess-profiling

Does your app run infinitely, unless you kill it?
Can you try running nvpof without timeout and see if you get the profiling data?


Thanks,
Ramesh