nvprof crash (error signal 134)

Hi everyone,

I’m trying to profile my application. It is mainly in fortran, but there is a small portion coded in cuda c. My program runs fine outside of nvprof, but for some reason I get a fortran floating invalid error when the profiler is run. This is probably irrelevant, but the application hangs in nvvp.

Why would my application run fine by itself, but crash during the profile?

==63715== NVPROF is profiling process 63715, command: ./ssk_cuda
forrtl: error (65): floating invalid
Image PC Routine Line Source
libcuinj64.so.6.0 00007FF6DEC66E87 Unknown Unknown Unknown
libcuinj64.so.6.0 00007FF6DEC65C52 Unknown Unknown Unknown
libcuinj64.so.6.0 00007FF6DEC65D67 Unknown Unknown Unknown
libcuinj64.so.6.0 00007FF6DEE65D79 Unknown Unknown Unknown
libpthread.so.0 0000003C350079D1 Unknown Unknown Unknown
libc.so.6 0000003C34CE8B6D Unknown Unknown Unknown
==63715== Profiling application: ./ssk_cuda
==63715== Profiling result:
No kernels were profiled.

==63715== API calls:
Time(%) Time Calls Avg Min Max Name
76.46% 211.03us 83 2.5420us 234ns 85.963us cuDeviceGetAttribute
12.49% 34.484us 1 34.484us 34.484us 34.484us cuDeviceTotalMem
9.91% 27.351us 1 27.351us 27.351us 27.351us cuDeviceGetName
0.81% 2.2320us 2 1.1160us 407ns 1.8250us cuDeviceGetCount
0.32% 891ns 2 445ns 379ns 512ns cuDeviceGet
==63715== Warning: Some profiling data are not recorded. Make sure cudaDeviceReset() is called before application exit to flush profile data.
======== Error: Application received signal 134

Are you doing proper cuda error checking on all kernel calls and CUDA API calls?
What happens when you run your application with cuda-memcheck?