I have a strange problem! I have a GTX 560 Ti card (compute capability 2.1) where I develop my CUDA apps.
Now after some hundred times of compiling, I got an error. That is, the compilation phase stops, with the cicc process not terminating. But when I change my number of threads per block number, it sometimes may compile. (For 64 threads it crashes for 1024 it works).
Does anyone had such a strange error. The other thing is, that with sm_13 it compiles every time, but with -arch=sm_21 it won't. But as I said it ran hundreds of times with this option, but now it will not work any more. Any suggestions?
Haha!
Here it runs again!
Commenting everything (global function) -> then compile -> uncomment again -> compiles as nothing happened before!!!
Such a strange behaviour! Maybe some intern compiler flags were set false, or something else.