I have read “Compute Command Line Profiler” document and I have set the following environment variables.
In the config_file I have tested with two counters: instructions and branch.
The problem is that I don’t get any output, the file is not even created. Nevertheless, if I run a CUDA Program I get correct profiling output with same configuration.
I am using CUDA Toolkit 5 prerelease and driver version 302.59.
The board I am using is a GeForce GTX 285.
Thanks in advance
Strange, that looks like the right environment variables to set.
I just started using the command line profiler in Linux for CUDA Toolkit 4.2 this week.
The other difference is that I specified the full-path for my config_file, and I didn’t set the *_CSV or *_LOG environment variables, letting those select the defaults.
But other than you using a pre-release Toolkit version as a possible cause, are you sure the current working directory is writable?
Yes, I have permissions to write on that directory because executable is generated on the same directory.
Also I have proved an absolute path as log_file and config_file environment variables.