Tricks and Tips: Compiling CUDA mex files with Visual Studio – The case of dynamic parallelism

With the recent development of CUDA 5.0 and the release of the Kepler architecture, dynamic parallelism makes it possible to call kernels within kernels.
The use of dynamic parallelism requires the generation of relocatable code and the linking of the cudadevrt.lib library.

In the framework of linking CUDA routines under Matlab by mexfiles, the new procedure at  Mathworks Help does not seem to make it possible to link external libraries.

The  downloadable Visual Studio 2010 project shows how the procedure  of our older post enables also to solve this problem.

Leave a Reply

Your email address will not be published. Required fields are marked *