mirrors/z3
3
0
Fork 0
mirror of https://github.com/Z3Prover/z3 synced 2025-07-29 07:27:57 +00:00
Code Activity

Update PARALLEL_PROJECT_NOTES.md

Browse source
This commit is contained in:
Nikolaj Bjorner 2025-07-25 15:12:13 -07:00 • committed by GitHub
parent 202807b317
commit f6fc5045d2
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 4 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

8
PARALLEL_PROJECT_NOTES.md
View file

@ -15,9 +15,9 @@ We track notes for updates to smt\_parallel.cpp and possibly solver/parallel\_ta
* Lookahead solvers: * Lookahead solvers:
* lookahead in the smt directory performs a simplistic lookahead search using unit propagation. * lookahead in the smt directory performs a simplistic lookahead search using unit propagation.
* lookahead in the sat directory uses custom lookahead solver. * lookahead in the sat directory uses custom lookahead solver.
They both proxy on a cost model where the most useful variable to branch on is the one that \_minimizes\_ the set of new clauses maximally They both proxy on a cost model where the most useful variable to branch on is the one that _minimizes_ the set of new clauses maximally
through unit propagation. In other words, if a literal \_p\_ is set to true, and \_p\_ occurs in clause $\\neg p \\vee q \\vee r$, then it results in through unit propagation. In other words, if a literal _p_ is set to true, and _p_ occurs in clause $\neg p \vee q \vee r$, then it results in
reducing the clause from size 3 to 2 (because $\\neg p$ will be false after propagating \_p\_). reducing the clause from size 3 to 2 (because $\neg p$ will be false after propagating _p_).
* VSIDS: * VSIDS:
* As referenced in Matteo and Antti's solvers. * As referenced in Matteo and Antti's solvers.
* Variable activity is a proxy for how useful it is to case split on a variable during search. Variables with a higher VSIDS are split first. * Variable activity is a proxy for how useful it is to case split on a variable during search. Variables with a higher VSIDS are split first.
@ -57,7 +57,7 @@ The existing algorithm in <b>smt_parallel</b> is as follows:
5. Share unit literals learned by each thread. 5. Share unit literals learned by each thread.
6. Compute unit cubes for each thread $T$. 6. Compute unit cubes for each thread $T$.
7. Spawn $T$ solvers with $F \wedge \ell$, where $\ell$ is a unit literal determined by lookahead function in each thread. 7. Spawn $T$ solvers with $F \wedge \ell$, where $\ell$ is a unit literal determined by lookahead function in each thread.
8. Perform a similar check as in 2. 8. Perform a similar check as in 2. But note that a thread can be UNSAT because the unit cube $\ell$ contradicted $F$. In this case learn the unit literal $\neg \ell$.
9. Shared unit literals learned by each thread, increase the maximal number of conflicts, go to 3. 9. Shared unit literals learned by each thread, increase the maximal number of conflicts, go to 3.
### Algorithm Variants ### Algorithm Variants