5. Conclusion

Four points are relevant:

• The Partial Factorization designed and implemented by Corinne ANCOURT and Fabien COELHO proves to be an improvement for implementations of Chernikova algorithm, in term of run time performance and number of exceptions;
• Existence of some tests that are not suitable for the Partial Factorization, therefore the incoherence of PerfectClub sampled and full database (see the sampling problem in example_sampling_problem);
• 64-bit is preferable to 32-bit computation for this set of tests;
• We plan to work on the conversion of New POLKA's convex hull implementation for POLYBENCH.