I have been surfing through the forums all day and I have tried all the options I found but nothing has worked so far.
So I am working with a Fedora 10 (that cannot handle the computeprof, so that is not an option), cuda toolkit 4.0 and driver 280.13. My hardware is one Tesla C2070 and my target sm is 2.0.
Now, if I set the compile options as “-cl-nv-verbose” nothing happens. (I have tried deleting the ~/.nv/ folder, the code is compiled but there is nothing on the log).
It is strange, since not even warnings appear on the log (I have a directive cl_amd_printf, that, when the compiler crashes, it appears as a warning).
Then, I have also tried the clcc compiler on my code with exactly the same result (no output).
Does anyone has a remote idea of how to fix it??? I just want to know how many registers and memory am I using, a solution (but not feasible) is just to go to the ptx code and count…