I am just starting to try and use Thrust in my CUDA software, but have hit a big stumbling block.
I am using Visual Studio 2008, and want to use its built-in debugger.
If I set the “Generate GPU Debug Information” option (under Configuration Properties->CUDA Runtime API->GPU) to Yes, the code will crash. This is the flag that sets the -G0 option in the NVCC compiler. If I do not set this flag, the code runs fine. Obviously, I want to be able to use this flag!
This problem is easy to reproduce - I simply took saxpy.cu from the Thrust examples and built a visual studio 2008 project around it. Also linked in cudart.lib, and am building 64 bit versions.
Any help would be greatly appreciated!