I’ve been using a modified version of Abe Stephen’s CMake script to compile my CUDA code for some time now. I’m having difficulties getting it to work when compiling code with "-arch=sm_13’ now. Has anyone else had any luck with this?
With CUDA 2.0, I could edit the CMAKE_CXX_COMPILE_OBJECT variable to make sure that it used nvcc as the compiler for compiling C++ files, which allowed me to pass the -arch flag. However, as of CUDA 2.1, this no longer works. Something seems to have changed with nvcc. When I add --dryrun to the command line for compiling, I also see no mention of the architecture flag, although if I manually add it (a second time) to the command line, it complains about having the flag twice.
Anyone have any ideas?