LibTORCH + LIBEFP

Added routines :

nnlib -> contains the scripted neural network potential files

efpmd/src/torch.c

efpmd/src/torch.h

efpmd/torch -> contains the wrapper libraries for torch-gradient computation routine

CMakeLists.txt -> connection between torch and EFP

Required modules :

cmake/3.18 or higher

openblas/0.3.8

netlib-lapack/3.8.0

gcc/9.3 or higher

The modules can be placed in module.sh. Run "source module.sh" to load these modules.

Setup LibTorch and LibEFP variables

Install LibTorch (C++ frontend) of PyTorch. If not, just gathering the lib/ and include/ directories are good enough.

Next we need to setup a few environment varibales. Libtorch and LibEFP directory addresses needs to be changed. In "setup.sh" change the following :

Turn the TORCH_SWITCH = ON for configuration with the torch-routines

setenv LIBEFP_DIR "/PATH/TO/libefp"

setenv LIBTORCH_INCLUDE_DIRS "/PATH/TO/libtorch/include/;/PATH/TO/libtorch/include/torch/csrc/api/include"

setenv TORCH_INSTALLED_DIR "/PATH/TO/libtorch"

**for bash shell use export command instead of setenv>

Compile and run

Run "./compile.sh" to compile.

Executable "efpmd" are in build/efpmd/efpmd

Run test files as :

"build/efpmd/efpmd tests/input.in"