This PR modifies the clause reduction heuristic. With the changes, we will keep the binary clauses in addition to the clauses that are used in the BCP and the half of the clauses kept according to the activity score.
I did some quick experiments -- 3 mins timeout, 8GB memory. Overall, we see improvement. However, one regression (QF_NIA) test timeout with this PR -- that test is disabled in thiS PR.
Coverage: 64.244% (-0.006%) from 64.25% when pulling 4e18eb1cfc912c630838551171364fed8fca751e on mcsat-keep-binary-clauses into eb46358378f88496bb8457e628bbf9131f130cae on master.
This PR modifies the clause reduction heuristic. With the changes, we will keep the binary clauses in addition to the clauses that are used in the BCP and the half of the clauses kept according to the activity score.
I did some quick experiments -- 3 mins timeout, 8GB memory. Overall, we see improvement. However, one regression (QF_NIA) test timeout with this PR -- that test is disabled in thiS PR.
QF_UF: 6879 -> 7084 QF_AX: 519 -> 522 QF_NRA: 10841 -> 10848 QF_NIA: 15683 -> 15827