We got this error message “Error: Const space Overflowed” when compiling a CUDA code on a system where no CUDA compliant device is installed.
On the opposite, compiling our code with a Quadro FX 5600 installed gives no error. Moreover our code works pretty well.
Is there any check for a present CUDA device in the system when compiling CUDA code ?
All right, we figured out what happened. It might be interesting for some people here.
Actually, the compilation platform is the key of our issue. In our kernel, we use several float* stored in const space memory. The compilation worked perfectly well on a 32 bits plateform and we had this error message “Error: Const space Overflowed” on a 64 bits platform. That’s just because our pointers in 64 bits take twice the size they take in 32 bits.
Finally, our issue was not CUDA related.