visual profiler error codes

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

i keep receiving an error while profiling my code :

Error: application being profiled returned a non-zero return code …

I am not sure what the problem is. The Code is healthy and memcheck returns o errors. Why is this happening? I have been able to profile the same code before i switchied into Cuda 5.5 rc …

best,
Thanasio

2

On most operating systems a non-zero process exit code indicates an error in the application. The Visual Profiler is warning that it detected a non-zero process exit code from the target application.

  1. Does the target application return a non-zero process exit code when run outside of the Visual Profiler? If yes, then fix the target application.

  2. Does the target application return a non-zero process exit code only when launched from the Visual Profiler? If yes, then debug what error is occurring in the application. These types of errors are often cause by a difference in the process launch environment including but not limited to:

  • Incorrect working directory
  • Incorrect command line arguments
  • Incorrect environment variables

These three settings can be specified in the Visual Profiler session settings.

3

hello guys,

Did you solved this problem? how can I solve this problem, HELP!
I also encounter this error:

==3481== NVPROF is profiling process 3481, command: ./a.out
Internal profiler error (8589934593:999)
==3481== Profiling application: ./a.out
==3481== Profiling result:
No kernels were profiled.
======== Error: Application returned non-zero code 1

nvprof version is 5.5; pgf90 version is 17.4;

4

Hi thanasio,

Did you solved this problem later? really hope to your reply.

5

try using CUDA 8, rather than CUDA 5.5

1 Like
6

Thank you for your rely, I have already solved this problems.Actually I installed cuda 8 and cuda tookit 8, but may be a wrong environment setting is made. Thank you for your kind help

7

yuan-ubuntu:
How to solve the problem?
I have a problem same as you.
the error message is “The application being profiled returned a non-zero return code”
I already debugged my code.
but when I use the nvvp, that error occurs…

8

make sure you have a

return 0;

statement at the end of your main routine, or wherever your application is exiting.

9

Thank you for your reply.
It already fed into the main routine. (i.e., return 0 is fed into the main code)
So, you guess that problem is the application exiting part, right?

int main(int argc, char ** argv) {
	parameter_handler param(argc, argv);

	pr_algorithm alg;
	bool set_indicator = false;
	float gpu_cache_percentage = 90;

	engine e(alg, param, set_indicator, gpu_cache_percentage);
	e.run();

	return 0;
}
10

your application may be exiting somewhere else besides that return 0 statement.

Figuring out why your application returns a non-zero error code is really not a CUDA programming question.

11

Thank you I will check my code again!

12

I found a reason for the problem.
A part of my code needs ‘sudo mode’.
So when I execute the code root mode, it works.
but when I execute the code user mode, it doesn’t work.

Do you know the how to execute sudo mode at the nvvp?

13

Switch to a root user.

Run nvvp.

[url]https://www.wikihow.com/Become-Root-in-Ubuntu[/url]

14

Thank you so much, Robert_Crovella!!!
I did it!

15

I resolved this issue.

The root cause is that I have no ‘writing permission’ to the working directory I used in nvvp.

The working directory in nvvp should be the path you ran *.exe without nvvp.

16

issue lies in the fact that you are returning something other than 0 “return 0;” in your code for the program to exit or simply in the settings you are trying to run your .cu file when you should be running the .exe file generated on compilation. For me, it was the latter that fixed it.