If you just want to use Warp, all tutorials and binaries can be found at http://www.warpem.com. Continue reading this README only if you're interested in compiling Warp from source.

CUDA SDK 10.0 with $PATH and $CUDA_PATH environment variables set correctly
cuDNN 7.4
Visual Studio 2017 with the Windows SDK 10.0 included during installation

Modified TensorFlow 1.10 source code

GTOM
liblion

Make sure the Warp, GTOM and liblion folders are located in the same parent folder, or modify the include paths accordingly.

Compile GTOM and liblion using the same configuration (Debug/Release) you intend to use for Warp's compilation.

Compile TensorFlow 1.10. This is a tricky thing to do on Windows, but our modified version should help. If it still doesn't compile, some googling will be required. Once compiled, create an environment variable $TENSORFLOW_LIBS that points to TF's 'build' folder. This will be required to configure all paths correctly in Warp's TFUtility project.

Open Warp.sln in Visual Studio. If everything is configured correctly, you shouldn't see any error messages. Now build the 'Warp' project, which will also build all of its dependencies. That's it!

Warp's GUI definitely won't build on Linux or OS X as there are no ongoing efforts to port WPF to these platforms. There is a good chance you will manage to build WarpLib, or any other WPF-free parts of the solution. However, we haven't tried it. We definitely intend to make a head-less, cross-platform version for processing in HPC clusters, but that is still a long way down the road.

Warp is being developed by Dimitry Tegunov ([email protected]) in Patrick Cramer's lab at the Max Planck Institute for Biophysical Chemistry in Göttingen, Germany.