Simon Cruanes
5d18086e53
fix(sat): resolution at level 0 is not recursive
...
recursion is implicit in the structure of the proof of the clause
that is unit at level 0, and thus responsible for propagating the
level-0 atom in the first place.
2021-11-14 22:50:12 -05:00
Simon Cruanes
f500704905
fix(sat): level-0 resolution needs to be recursive
2021-11-10 13:20:00 -05:00
Simon Cruanes
0abe4b7020
wip: decode more proof steps to quip
2021-10-27 21:50:28 -04:00
Simon Cruanes
992b93abcf
fix(sat): clause in a unsat result must not contain 0-level literals
2021-10-27 20:19:01 -04:00
Simon Cruanes
597a6c378e
wip: split VecI32 and VecSmallInt
...
- use VecSmallInt for small integers of type `int`
- use VecI32 to store actual int32 (in particular for proof steps)
2021-10-16 20:31:56 -04:00
Simon Cruanes
fd1d068997
proof stubs and sat proof
2021-10-12 22:13:28 -04:00
Simon Cruanes
d3537f2c3f
wip: refactor proof
2021-10-07 20:49:39 -04:00
Simon Cruanes
bbb995b0d5
refactor some names related to proofs; wip add unit paramod
2021-10-03 20:32:37 -04:00
Simon Cruanes
df40b5a5c1
wip: refactor(sat): generate detailed proofs again
...
because proofs now require hypotheses but not in a resolution order, we
can still do conflict minimization.
2021-09-29 22:18:36 -04:00
Simon Cruanes
bbe366989c
perf(sat): use Atom.Vec for temporary atom array
2021-09-27 19:28:45 -04:00
Simon Cruanes
c9e257d40b
cleanup
2021-09-27 12:14:04 -04:00
Simon Cruanes
5bed2d1c5f
detail
2021-09-26 23:56:40 -04:00
Simon Cruanes
a22bfe06c1
remove debug msgs
2021-08-31 23:19:06 -04:00
Simon Cruanes
debd8bcaf8
fix warning
2021-08-31 23:06:48 -04:00
Simon Cruanes
350a23d99e
feat: minimize conflicts
...
similar to minisat's level 2 of minimization.
2021-08-31 23:04:32 -04:00
Simon Cruanes
5080195c5b
feat: conflict minimization à la minisat
2021-08-31 22:59:38 -04:00
Simon Cruanes
521340a23f
feat: first full implem of clause pools
2021-08-31 22:56:42 -04:00
Simon Cruanes
10dca21f59
refactor: remove history in conflict resolution; remove simpls
...
no need to simplify reasons anymore, we rely on DRUP for that.
2021-08-31 22:56:11 -04:00
Simon Cruanes
f86498b386
feat: make it compile
2021-08-31 18:59:44 -04:00
Simon Cruanes
16bb65ebfa
wip: clause pools
2021-08-31 09:30:28 -04:00
Simon Cruanes
4a2367b1bd
refactor: use Atom.Vec (a VecI32) for atom vectors
2021-08-31 09:30:05 -04:00
Simon Cruanes
1877c00c02
wip: clauses pools
2021-08-30 09:32:32 -04:00
Simon Cruanes
73b39fe075
fix more warnings
2021-08-27 21:34:26 -04:00
Simon Cruanes
782afa4415
feat: move Int_id into its own module
2021-08-25 23:52:08 -04:00
Simon Cruanes
bb682b8080
fix(sat): emit proofs where needed
2021-08-23 00:08:09 -04:00
Simon Cruanes
baecce0946
feat: use Stat in SAT solver
2021-08-22 01:56:54 -04:00
Simon Cruanes
672f828c82
perf: return to default-pol=true
...
previous commit fixed sign error (move from `negated` to `same_sign`
so restore good behavior
2021-08-22 01:44:37 -04:00
Simon Cruanes
27796da5a8
fix preprocessing in th-bool
2021-08-22 01:29:01 -04:00
Simon Cruanes
e93e084eac
refactor: eager proofs; stronger preprocessing
...
proofs are now directly emitted (almost) everywhere, which simplifies
a lot of things. preprocessing is more recursive (a bit too much
really).
2021-08-22 01:13:41 -04:00
Simon Cruanes
075e251aed
fix: bad sign in SAT solver
2021-08-21 13:41:30 -04:00
Simon Cruanes
0e77c9ef33
perf(solver): default-pol=false
2021-08-20 18:49:15 -04:00
Simon Cruanes
1ab7d34a7d
refactor: make it compile again
2021-08-20 18:18:30 -04:00
Simon Cruanes
1d3867acb5
perf: use VecI32 for clause vector
2021-08-20 18:18:26 -04:00
Simon Cruanes
3fbb9af664
refactor(sat): hide atoms, API now talks only about literals
2021-08-19 09:35:54 -04:00
Simon Cruanes
8bc1f1c864
feat: inner DRUP proof checking for pure-sat-solver
2021-08-19 00:15:00 -04:00
Simon Cruanes
9f01b98cde
wip: imperative proofs
...
- getting closer to having the SMT solver compile again
- dummy proof implementation
- DRUP proof implementation for pure SAT solver
2021-08-18 23:59:39 -04:00
Simon Cruanes
7bead748a6
refactor: move SAT_PROOF into sidekick.core
...
SAT proofs are part of bigger proofs, now. And we expect them to work on
the literals of CDCL(T) directly, bypassing the low level boolean atoms
2021-08-17 23:59:02 -04:00
Simon Cruanes
27e775ee04
wip: refactor proofs for SMT
2021-08-12 22:05:49 -04:00
Simon Cruanes
b9800023ec
feat(sat): missing call to on_learnt; better API
2021-08-02 23:48:52 -04:00
Simon Cruanes
233df98229
perf(sat): proper GC for clauses
...
GC is now eager in the cleanup of watchlists and more elaborate.
It starts by marking all clauses justifying literals on the trail,
because we cannot remove them yet (they might be used in resolution
steps during clause analysis).
Then, we sort set of learnt clauses by decreasing activity, and pop them
one by one until the target size is reached. Clauses that are marked,
because they are on the trail, are kept on the side so we can push them
back at the end.
For each clause we want to remove, we start by marking it "dead".
Then we mark its two watch literals are "dirty", which means we will
need to purge their watchlists from dead clauses.
Then we purge watchlists (remove deadclauses entirely). At this point
the dead clauses are unreachable from both trail and watchlists.
We can remove all data for each such clause, and put their index for
recycling so another new clause can reuse their slot.
Finally we unmark dirty literals and saved-by-trail-explanation
explanation clauses.
2021-07-21 20:29:27 -04:00
Simon Cruanes
77c61b536e
refactor(sat): rename some bitfields, leaner clause creation
2021-07-21 20:24:03 -04:00
Simon Cruanes
3aa25cb2a2
sat: use an atom array for clauses again
...
allocator is not worth the complexity, and has a lot of double
indirections anyway.
2021-07-21 10:02:56 -04:00
Simon Cruanes
8b94e8404f
wip: data-oriented clauses
2021-07-20 23:34:58 -04:00
Simon Cruanes
97d87cc58f
perf: tiny detail in int indices in solver
2021-07-20 23:34:38 -04:00
Simon Cruanes
a03d7f6bef
fix
2021-07-20 23:31:21 -04:00
Simon Cruanes
89051fd3ad
perf(solver): use VecI32 in the heap
...
this reduces the size of the heap (4 bits per element, not 8), and
reduces GC work by not scanning the bigarray
2021-07-20 10:07:43 -04:00
Simon Cruanes
347e098fc2
misc
2021-07-20 09:20:14 -04:00
Simon Cruanes
f2b2bbb973
some stats for the SAT solver
2021-07-19 21:44:58 -04:00
Simon Cruanes
b624a1ca5d
cleanup: remove commented code
2021-07-19 21:15:16 -04:00
Simon Cruanes
e30cf9fdbf
perf: allocate less in conflict analysis
...
use preallocated vectors
2021-07-19 21:13:00 -04:00
