I thought I would pass this along.
At the top of my kernel.cu I have a macro called version defined to be “Revision: 1.169 ”
nvcc release 3.2, V0.2.1221 fails to compile kernel.cu
Instead it gives an error message: cuda.h(1803): error: expected a “)”
It turns CUDA 3.2 has a routine called cuCtxGetApiVersion which returns an integer
to an argument called version.
This does not compile if you have a macro also called version.
Work around: My macro is now called Version
ps: discoved by using gcc’s -e switch, which nvcc also understands