Cuda runtime API error Error occurs when used -dir flag


I’m having problems trying to execute programs compiled with a devcode directory (with generated ptx code) on both nvcc 4.0 and nvcc 4.1

Using nvcc 4.0 the error is:

cudaSafeCall() Runtime API error 10301: unspecified driver error.

Using nvcc 4.1 the error is:

cudaSafeCall() Runtime API error 30: unknown error.

Using nvcc 3.2 it works fine, but I’m unable to compile newer SDK samples. I’m using fedora 15, GCC 4.5.1 20100924 and nvidia dev driver 285.05.15. Should I try the official cuda 4.0 driver?

A compile line command example for lineOfSight:

/nvcc -arch=compute_20 -int=all -ext=all -dir=lineOfSight.devcode -DUNIX --compiler-options "-fpermissive -O2" -I./ -I../../../ -I/home/undead/sdk/shared/inc -I/home/undead/sdk/C/common/inc -I/home/undead/cuda/include -L/home/undead/cuda/lib64 -L/home/undead/sdk/C/lib -L/home/undead/sdk/shared/lib -L/home/undead/sdk/C/common/lib/linux -lcutil_x86_64 -lshrutil_x86_64 -lGL -lGLU -lglut -lGLEW -o lineOfSight

Before I used the flags -int=none -ext=all on a fedora 13. When I switched to fedora 15, even on sdk 3.2 failed with those flags. Switching to -int=all -ext=all it worked.



hi sampaio,if you remove cudaSafeCall() code in your programs then this problem will resolve…