Profiler error message when profiling JCuda application

Hello

With previous versions of the Visual Profiler (namely, up to version 4.2) it was possible to profile Java applications that use http://www.jcuda.org/ to call CUDA functions via JNI. This was possible by simply specifying a .BAT file (for starting the Java Application on Windows) as the executable of the program.

In the most recent version (5.0 final), the attempt to launch such an application only results in the following error message:
======== Warning: No CUDA application was profiled, exiting

(The same error message was reported in CUDA profiling through Matlab in Linux - CUDA Programming and Performance - NVIDIA Developer Forums , but this seemed fairly unrelated, so I’m starting a new thread).

Have there been any changes in the Visual Profiler that could explain this behavior? Are there any known workarounds?

Thanks in advance
Marco

EDIT: Sorry, this might belong into Visual Profiler and nvprof - NVIDIA Developer Forums

Still the same error with version 5.5.

(Also mentioned here Profiling Java application with nvvp - Visual Profiler and nvprof - NVIDIA Developer Forums in a similar context)

To emphasize this once more: This worked like a charm with version 4.2 of the Visual Profiler!

(Actually, in version 5.5, the error message seems to be omitted, but it is shown when the VIPER_DEBUG environment variable is set)

EDIT: The message
======== Warning: No CUDA application was profiled, exiting
is printed by nvprof. It seems to have some problem with Java spawning child processes or so. With the
–profile-child-processes
argument, nvprof seems to run properly, but despite some effort, I did not (yet) find a way to let the Visual Profiler use the same argument.

Still the same with version 6.0 and 6.5 (gonna test 7.0 later). It can’t be THAT hard, so I guess it’s intentional.