Building CUDPP

CUDPP has currently been tested in Windows XP, Windows Vista, Mac OS X and Linux. See Release Notes for release specific platform support.

Building CUDPP on Windows XP

CUDPP can be built using either or MSVC 8 (2005) or MSVC 9 (2008). To build, open cudpp/cudpp.sln. Then you can build the library using the "build" command as you would with any other workspace. There are four configurations: debug, release, emudebug, and emurelease. The first two are self-explanatory. The second two are built to use CUDA device emulation, meaning they will be run (slowly) on the CPU.

Building CUDPP on Linux and Mac OS X

CUDPP can be built using standard g++ and Make tools on Linux, by typing "make" in the "cudpp/" subdirectory. Before building CUDPP, you should first build the CUDA Utility Library (libcutil) by typing "make; make dbg=1" in the "common/" subdirectory. This will generate libcutil.a and libcutilD.a.

The makefile for CUDPP and all sample applications take the optional arguments "emu=1" and "dbg=1". The former builds CUDPP for device emulation, and the latter for debugging. The two flags can be combined. "verbose=1" can be used to see all compiler output.

Building CUDPP Sample Applications

The sample applications in the "apps/" subdirectory can be built exactly like CUDPP is--either by opening the appropriate .sln/.vcproj file in MSVC in Windows, or using "make" in Linux.

On some Linux installations you will get linker errors relating to "-lXi" and "-lXmu". To fix this, you will need to install libXi and libXmu. On Debian and Ubuntu, for example, you can simply run "sudo apt-get install libxi-dev", and "sudo apt-get install libxmu-dev"