Dynamic launch works differently on different systems?

Feb 1, 2014 at 11:59 PM
On one system I can use the new "gpu.Launch().add(2, 7, dev_c)" syntax (from Chapter03/simple_kernel_params), but on another system there is a runtime error that "add" could not be found. However, if I use the older syntax "gpu.Launch(1, 1, "add", 2, 7, dev_c)" the example does work fine.

These are both Windows 7, x64, with the same source base, CUDAFY, and CUDA, but probably different additional software installed. However, they both have VS2010, VS2012, and VS2013 Ultimate installed.

I have CUDAfy.1.25 installed, CUDA 5.5 SDK, and the 331.58 NVIDIA drivers. One system is a notebook (where the new syntax works), while the system where the old syntax has to be used is a desktop system.

I have a couple more experiments to try, but any suggestions?
Feb 2, 2014 at 3:14 PM
Try dumping your CudafyModule object to file (Serialize() ) and examining in either XML viewer or CudafyModuleViewer.exe. Check if add function is actually present.