cuda-memcheck Could not start the application (7)

2
votes

when I use cuda-memcheck ,it's shows

./cuda-memcheck  ~/work/local_pc_code/bin/test/test_parm 
========= CUDA-MEMCHECK
========= Could not start the application (7)
========= No CUDA-MEMCHECK results found

my compile flag is:

nvcc -G -g  -Xcompiler -rdynamic -gencode arch=compute_61,code=sm_61  -L /usr/local/cuda-8.0/lib64/ -lcublas -lcudnn -lcudart

what's error number 7?

1
cudagpu

1 Answers

3
votes

According to my testing, cuda-memcheck must be run in a directory that your user account has write access to, or else it will report the error you indicate.

It seems evident that you are likely doing that:

./cuda-memcheck  ...
^^

This command invocation (./) suggests to me that you are attempting to run cuda-memcheck from the directory where it was installed. This is not typical or expected usage. In a typical CUDA install, and assuming you are not root, you would not have write access to the directory that the cuda-memcheck executable is located in.

  • Make sure you have properly updated your PATH environment variable so that you can properly run cuda-memcheck from any location. The instructions for that are here.

  • change your directory to one that you have write access to. Presumably you could do e.g. cd ~/work/local_pc_code/bin/test/

  • run cuda-memcheck from there, e.g. cuda-memcheck ./test_parm