As some of you may have already noticed, there is many references to a tool named CUDACG into the nvcc manual provided with CUDA 2.0 Beta. It looks like internal references that was forgotten to be removed for the public release of the file ;-)
A cubin generation phase from Cg is mentioned, and the diagram page 23 introduces cudacg, acting in place of nvopencc and ptxas. If I read correctly this diagram, cudacg seems to be designed to produce directly cubin (true hardware assembly) from Cg source code embedded into cpp source file (like cuda kernel codes).
I exposed my speculations about this in a post on Cg developer forum, but I didn’t get any response yet ( http://developer.nvidia.com/forums/index.php?showtopic=1565).
Anybody have any speculation on the usage of this ? Will this tool be provided with final release of CUDA 2.0 toolkit ?