cudaprof in CUDA 5.0 and how to use it on linux system

Asked Nov 27 '12 at 08:08

Active Nov 27 '12 at 09:48

Viewed 947 times

I wanted to run the profiler on my code. I have CUDA 5.0 installed and have CentOS 6.0 as my operating system. Firstly, I could not find cudaprof anywhere on my system. I searched on internet and came to know that nvprof can be used to profile from command line. So, I compiled my code using -g and -G options and ran it using the command:

nvprof -o profileOutput -s ./exe args.

But I am getting the following error:

terminate called after throwing an instance of 'thrust::system::system_error'
  what():  unspecified launch failure
======== Warning: Application received signal 6
======== Error: failed to read result file

I am not understanding what wrong am I doing.

Also, What is the location of cudaprof on Linux system with CUDA 5.0 installed?

edited Nov 27 '12 at 09:48

asked Nov 27 '12 at 08:08

user1439690

Don't compile your code with debugging flags - it is causing thrust to break and your GPU code isn't running – talonmies Nov 27 '12 at 08:25
So can I use nvprof without debugging flags? – user1439690 Nov 27 '12 at 09:17
Ok I got it. I compiled without the debugging flags and it is working now :). Thank you – user1439690 Nov 27 '12 at 09:22
What is the location of cudaprof on Linux system with CUDA 5.0 installed? – user1439690 Nov 27 '12 at 09:47
Sometime back the name of cudaprof was changed to nvvp standing for Nvidia Visual profiler. – user1439690 Nov 27 '12 at 09:58
1

For profiling it is useful to add the option -lineinfo or --generate-line-info to your release compiles. This will generate line table information but not affect optimizations. – Greg Smith Nov 28 '12 at 05:28

cudaprof in CUDA 5.0 and how to use it on linux system

0 Answers0