My Cuda 3, openCL etc were working fine, following the Getting Started for Cuda 4 on OSX fails to correctly populate at least /usr/local/cuda , nvcc missing. The guide says, to begin with, to clean the cuda and /Developer/GPU Computing directories, I took that to mean
sudo rm -rf /usr/local/cuda
Then install driver and install GPU SDK. Curiously the guide does not explicitly say to reboot after the driver installation, and it does not say when to install the little cudatools package, I tried several variations on the theme here, with reboot, without reboot etc, but in all cases I was left with a sing lib file in the cuda directory, no cuda/bin at all. Using updatedb and locate showed that nvcc does not end up anywhere in the system anyway.
It looks like the installer has some Big Bug (I think the SDK installation finishes too quickly anyway on my system), and the manual needs even stricter wording, just in case. The private download area for Cuda 4 has a tiny .xml file for OSX, does anyone know what’s its purpose?