ParallelMiniSchur

A parallel implementation in CUDA to get an approximate mini schur complement for large scale Bundle Adjustment Problem

Installation Instructions

• To install the solver, you need to have cuda installed in your system
• Installation command - make
• A binary executable is formed in the bin/ folder

Running instructions

• The solver is run using the executable formed above
• The command line format to run the solver is as follows:

    bin/main [filename] [matrix_descr] [matrix_order] [matrix_index_format] [nparts] [sizeG]

• The command line options are explained below:
  • filename : the complete path to the matrix text file
  • matrix_descr : The code for matrix description (CSR - 0, COO - 1)
  • matrix_order : The code for matrix order (column major - 0, row major - 1)
  • matrix_index_format : The code for matrix index format (zero base indexing - 0, one base indexing - 1)
  • sizeG : The size of the G partition while building the preconditioner

Sample run of the solver