GBLA -- Gröbner Basis Linear Algebra Package
share
This is a system paper about a new GPLv2 open source C library GBLA implementing and improving the idea of Faugère and Lachartre (GB reduction). We further exploit underlying structures in matrices generated during Gröbner basis computations in algorithms like F4 or F5 taking advantage of block patterns by using a special data structure called multilines. Moreover, we discuss a new order of operations for the reduction process. In various different experimental results we show that GBLA performs better than GB reduction or Magma in sequential computations (up to 40 much better than GB reduction for a higher number of cores: On 32 cores we reach a scaling of up to 26. GBLA is up to 7 times faster than GB reduction. Further, we compare different parallel schedulers GBLA can be used with. We also developed a new advanced storage format that exploits the fact that our matrices are coming from Gröbner basis computations, shrinking storage by a factor of up to 4. A huge database of our matrices is freely available with GBLA.
READ FULL TEXT
