Wavelet-Based Density Functional Theory on Massively Parallel Hybrid Architectures
Résumé
In this chapter, we describe the GPU acceleration of density functional theory (DFT) calculations based on wavelet basis sets as realized in the BigDFT code. Daubechies wavelets have not been traditionally used for DFT calculations, but they exhibit properties that make them attractive for both accurate and efficient DFT simulations. Here we explain how an existing MPI parallel wavelet-based DFT implementation can benefit from the computational power of GPUs by offloading numerically intensive operations to GPUs.