GPU Fermion Matrix

[Marcel-Rodekamp]

Make the fermion matrix available for the GPU. Maybe we require #96.

[Marcel-Rodekamp]

A first attempt has been made in feature/GPUBasics in this commit. At least with Lattices that use the default methods the GPU support for the Fermion matrix is available. A example:

#include "NSL.hpp"

int main(){
    // shortcut the real dtype
    typedef double real;
    // shortcut the complex dtype
    typedef NSL::complex<real> complex;

    // Space Time Lattice size:
    NSL::size_t Nt = 64;
    NSL::size_t Nx = 8;
    complex beta = {2.,0.};

    // Create a lattice, the Ring lattice uses only the default methods from lattice.tpp allowing a pull copy to GPU
    // so far it is undetermined what happens with Square (more tests required)
    NSL::Lattice::Ring<complex> Lattice(Nx);
    Lattice.to(NSL::GPU());

    // Create a test configuration
    NSL::Tensor<complex> phi(NSL::GPU(),Nt,Nx); phi.rand();

    // Construct the test fermion matrix matrix
    NSL::FermionMatrix::HubbardExp M(Lattice,phi,beta);

    // Create a test vector to apply the matrix on
    NSL::Tensor<complex> psi(NSL::GPU(),Nt,Nx); phi.rand();

    // mat vec application runs on the GPU
    NSL::Tensor<complex> chi = M.M(psi); 

    std::cout << chi << std::endl;

    return EXIT_SUCCESS;
}

[evanberkowitz]

112 feature/GPUBasics merged this in!