| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
This fixes two issues for preprocessing:
(1) The term preregistration visitor was calling preRegister on terms multiple times in a SAT context, which the linear arithmetic solver is sensitive to.
(2) It was possible for non-preprocessed terms to appear in assertions if they were on the RHS of substitutions learned by non-clausal simplification, and substituted into assertions post-theory-preprocessing.
To fix (1), the SharedTermsVisitor is update to track which theories has preregistered each term, as is done in the PreRegisterVisitor. To fix (2), we no longer apply-subst after theory preprocessing.
These two fixes are required to fix #6071.
Note: we should performance test this on SMT-LIB.
|
|
This PR does some more cleanup of the includes.
|
|
|
|
This refactors the term registration visitors (PreRegisterVisitor and SharedTermsVisitor), which are responsible for calling Theory::preRegisterTerm and Theory::addSharedTerm.
The purpose of this PR is to resolve a subtle bug in the implementation of PreRegisterVisitor (see explanation below), and to improve performance in particular on the Facebook benchmarks (where preregistering terms is currently 25-35% of runtime on certain challenge benchmarks).
This PR makes it so that that SharedTermsVisitor now subsumes PreRegisterVisitor and should be run when sharing is enabled only. Previously, we ran both PreRegisterVisitor and SharedTermsVisitor, and moreover the former misclassified when a literal had multiple theories since the d_theories field of PreRegisterVisitor is never reset. This meant we would consider a literal to have multiple theories as soon as we saw any literal that had multipleTheories. After this PR, we run SharedTermsVisitor only which also handles calling preRegisterTerm, since it has the ability to anyways (it computes the same information as PreRegisterVisitor regarding which theories care about a term).
The refactoring merges the blocks of code that were copy and pasted in term_registration_visitor.cpp and makes them more optimal, by reducing our calls to Theory::theoryOf.
FYI @barrettcw
|
|
|
|
This makes the shared terms database use a proof equality engine as a layer on top of its equality engine, analogous to how this done in theories.
|
|
This PR adds the definition for the "shared solver", it does not connect it to TheoryEngine/CombinationEngine yet. This is an object that behaves like a "glue" theory solver and has a special role in TheoryEngine.
In the current architecture, the "SharedTermsDatabase" is the "shared solver", although in the central architecture, the shared solver will be required to behave differently. This PR adds the abstract definition of shared solver, where notice SharedSolverDistributed is a thin layer on top of SharedTermsDatabase.
In a followup PR, CombinationEngine will maintain a shared solver and ~5 blocks of code in TheoryEngine will be callbacks to the SharedSolver. Also, eventually the code for SharedTermsDatabase should be split: the parts involving equality reason/propagation/explanation should be migrated into SharedSolverDistributed, and the parts related to registration will remain in SharedTermsDatabase (to also be used in the planned SharedSolverCentral).
FYI @barrettcw
|
