In the nvcc 2.0 documentation, there is a switch “–machine” that allows nvcc to target 32 or 64-bit platforms. However, it also says that it is only available when compiling .cubin files on Linux.
Are there plans to support this switch for normal compilation on Windows? My development box is running Windows XP 64-bit, but I’ll need to compile a separate 32-bit version of my program for distribution. Currently, it looks like the only way to do this is to make a separate installation of the 32-bit version of nvcc.
It would be nice to have the feature in CUDA 2.1 so that only one copy of the CUDA toolkit is installed, and we can use the switch to compile for our target platform (x86 or x64).