0

I have been checking the SMT-COMP 2015 benchmark results for the QF_AUFBV mode and they are shocking. Same thing for 2014.

According to them Z3 is orders of magnitude slower than the rest of the SMT solvers.

My understanding is that Z3 is as fast as CVC4. Am I missing something?

Regards

user1618465
  • 1,813
  • 2
  • 32
  • 58

1 Answers1

3

Yes, that is correct, Z3 is not particular good in this category. We are working on multiple different techniques that will have an impact on that, but they are not ready yet and they are not directly targetted at those 37 benchmarks. Given the low number of benchmarks in the competition, we should be very cautious about drawing any conclusions from those results.

Christoph Wintersteiger
  • 8,234
  • 1
  • 16
  • 30
  • Thank you for your answer. In my case I only need to use QF_AUFBV theory. Do you then, for this case, recommend me to use CVC4?. (I know that Z3 is better in other theories) Thank you for the awesome job you guys do with Z3 – user1618465 May 18 '16 at 16:03
  • That depends entirely on your requirements. If your goal is to win the competition, then the current version of Z3 is not the right tool for you. If you need to solve problems that are not SMT-LIB benchmarks then they need to be looked at to see why exactly Z3 is slow on your problems and whether that can be changed. – Christoph Wintersteiger May 19 '16 at 18:21