This is the first version of cuADMM. cuADMM is a GPU-accelerated SDP solver based on the ADMM algorithm. It supports multiple GPU-efficient PSD projection methods:
- standard projection using cuSOLVER’s eigenvalue decomposition
- efficient projection in low precision using composite polynomials
- low-rank projection using LOBPCG
This version supports TXT input files as well as calls from MATLAB using bindings.