Download instructions for the GLU solver

Introduction

GLU was GPU-accelerated parallel sparse LU factorization solver developed by VSCLAB at UC Riverside. GLU is based on a hybrid right-looking LU factorization algorithm. We show that more concurrency can be exploited in the right-looking method than the left-looking method, especially on GPU platforms.

GLU has the following features and advantages over existing sparse LU solvers:

  • It is base on the hybrid column-based right-looking LU factorization method, which is shown to be more amenable for exploiting the concurrency of LU factorization. The new method preserves the benefit of column-level concurrency and symbolic analysis in the left-looking method meanwhile, it allows more parallelism to be exploited in GPUs.
  • GLU solver allows the parallelization of all three loops in the LU factorization on GPUs. In contrast, the existing GPU-based left- looking LU factorization approach can only allow two-level parallelization.
  • Currently the GLU package does not include the triangular solver. The reason is that we found triangular solver on GPU does not work much faster than the solver on CPU (based on our implementation). We may add the GPU-accelerated triangular solver in the future.

    This software package (GLU) is copyrighted by UCR under BSD license.

    Please see more detailed information about GLU solver at here

    Download instruction

    Please fill the form below to obtain the GLU library and user manual. Please contact Sheldon Tan at stan@ece.ucr.edu for comments

    First name 20 chars max.
    (*required)
    Last name 20 chars max.
    (*required)
    E-mail (*required)
    Phone number
    Package name
    (*required)
    Organization 40 chars max.
    (*required)
    Research area 80 chars max.
    (*required)
    Additional comments and requests? 200 chars max.