cuADMM

CUDA-based ADMM solver for SDPs

cuADMM is a CUDA-based implementation of the Alternating Direction Method of Multipliers (ADMM) algorithm to solve Semi-Definite Programming (SDP) problems (Groudiev et al., 2025).

Illustration of the implementation of cuADMM. The left side shows the linear system solving, and the right part shows the PSD cone projection.

References

  1. RSS Workshop
    cuADMM: GPU-Accelerated First-Order Optimization for Large-Scale Multi-Block Semidefinite Programs
    Antoine Groudiev, Shucheng Kang, and Heng Yang
    In 2025 RSS Workshop on Fast Motion Planning and Control in the Era of Parallelism, 2025
    PDF Code