April 2014
-------------------
GENERAL INFORMATION
-------------------
This is version 3.02 of the search-based QBF solver DepQBF. Compared to
version 3.0, the semantics of the API for incremental use of the solver has
been adapted. Please see the header file 'qdpll.h' for comments and the code
examples in the subdirectory "examples" of the release.
The example './examples/basic-api-example2.c' is the most comprehensive one
and demonstrates the use of the API and, in particular, the 'qdpll_gc'
function.
Version 3.02 fixes a segmentation fault in the QDIMACS output generation by
the API functions 'qdpll_get_value' and 'qdpll_print_qdimacs_output'.
Version 3.02 supports shared library generation (contributed by Thomas
Krennwallner).
Many thanks to Robert Koenighofer, Thomas Krennwallner, Martin Kronegger, and
Andreas Pfandler for valuable feedback.
Compared to the previously released version 2.0, DepQBF 3.02 includes the
following major changes:
- Incremental solving: the API of the solver allows for incremental solving
based on a clause stack. Clauses can be added to and removed from the
current formula by push and pop operations. Incremental solving can be
beneficial in applications where a sequence of closely related formulae must
be solved. This way, the solver does not have to solve each formula from
scratch but tries to reuse information learned from previously solved
formulae.
- Solving under assumptions: assumptions are fixed variable assignments from
the leftmost quantifier block of a QBF. Assumptions can be added through the
solver API. In forthcoming calls, the solver tries to solve the formula
under the assignments given by the added assumptions.
- Long-distance resolution for clause and cube learning: traditional
Q-resolution explicitly rules out the generation of tautological
resolvents. In contrast to that, long-distance resolution admits certain
tautological resolvents. It was first implemented in the QBF solver
'quaffle' and is now also available in DepQBF.
- Some code maintenance and bug fixes: many thanks to Adria Gascon for
valuable feedback.
PLEASE SEE the header file 'qdpll.h', the examples in the subdirectory
'examples', and the command line documentation (call './depqbf -h') for
further information on the use of the solver and its library.
General features of DepQBF:
- The solver can be used as a library. The API is declared in file 'qdpll.h'
and the examples in the subdirectory 'examples' demonstrate the basic use.
- Advanced clause and cube learning based on QBF Pseudo Unit Propagation as
presented in the following paper: "Florian Lonsing, Uwe Egly, Allen Van
Gelder: Efficient Clause Learning for Quantified Boolean Formulas via QBF
Pseudo Unit Propagation. In Proc. SAT 2013."
NOTE: by default, this version of DepQBF applies a lazy variant of
QPUP-based QCDCL where no resolution steps are carried out. The traditional
approach to QCDCL which was implemented in earlier versions of DepQBF is
still available by command line option '--traditional-qcdcl'. Please see
also the command line documentation by calling './depqbf -h'.
- Generation of QDIMACS output (partial certificate): if the outermost
(i.e. leftmost) quantifier block of a satisfiable QBF is existentially
quantified, then DepQBF can print an assignment to the variables of this
block (and dually for unsatisfiable QBFs and universal variables from the
outermost block, if that block is universally quantified). To enable QDIMACS
output generation, run DepQBF with parameter '--qdo'. Note that the
assignment printed by DepQBF can be partial, i.e. not all variables are
necessarily assigned. In this case, the variables for which no value was
printed can be assigned arbitrarily.
- Trace generation (contributed by Aina Niemetz): DepQBF can produce traces in
QRP format (ASCII and binary version of the QRP format are supported; see
also the command line documentation). If called with the '--trace' option,
the solver prints *every* resolution step during clause and cube learning to
<stdout>. The output format is QRP ("Q-Resolution Proof"). For example, the
call './depqbf --trace input-formula.qdimacs > trace.qrp' dumps the trace
for the QBF 'input-formula.qdimacs' to the file 'trace.qrp'. The generated
trace file can be used to extract a certificate of (un)satisfiability of the
given formula using additional tools. See also the website
'http://fmv.jku.at/qbfcert/' and the related tool paper published at SAT'12.
NOTE: tracing must be combined with the trivial dependency scheme (i.e. the
linear quantifier prefix ordering) by option '--dep-man=simple'. Further, to
enable tracing for QPUP-based QCDCL, '--no-lazy-qpup' must be specified.
DepQBF consists of a dependency manager (file 'qdpll_dep_man_qdag.c') and a
core QDPLL solver (file 'qdpll.c'). During a run the solver queries the
dependency manager to find out if there is a dependency between two variables,
say 'x' and 'y'. Given the original quantifier prefix of a QBF, there is such
dependency if 'x' is quantified to the left of 'y' and 'x' and 'y' are
quantified differently. In contrast to that simple approach, DepQBF (in
general) is able to extract more sophisticated dependency information from the
given QBF. It computes the so-called 'standard dependency scheme' which is
represented as a compact graph by the dependency manager.
If you are interested only in the core solver based on QDPLL then it is
probably best not to look at the code of the dependency manager in file
'qdpll_dep_man_qdag.c' at all but only consider file 'qdpll.c'.
-------
LICENSE
-------
DepQBF is free software released under GPLv3:
https://www.gnu.org/copyleft/gpl.html
See also the file COPYING.
------------
INSTALLATION
------------
The latest release is available from http://lonsing.github.io/depqbf/
Unpack the sources into a directory and call 'make'. This produces optimized
code without assertions (default).
If you want to use the solver as a library in your own applications, then link
against 'libqdpll.a'.
Note: set the flag 'FULL_ASSERT' in file 'qdpll_config.h' from 0 to 1 to
switch on *expensive* assertions (recommended only for debugging). The solver
will run *substantially* slower in this case. As usual, using the compiler
flag 'DNDEBUG' removes all assertions from the code, regardless from the value
of 'FULL_ASSERT'.
-----------------------
CONFIGURATION AND USAGE
-----------------------
Call './depqbf -h' to display usage information. Further, undocumented command
line parameters can be found in function 'qdpll_configure(...)' in file
'qdpll.c'. These parameters are mostly experimental.
The solver returns exit code 10 if the given instance was found satisfiable
and exit code 20 if the instance was found unsatisfiable. Any other exit code
indicates that the instance was not solved.
Parameter '-v' enables basic verbose mode where the solver prints information
on restarts and backtracks to <stderr>. More occurrences of '-v' result in
heavy verbose mode where information on individual assignments is
printed. This can slow down the solver considerably and should be used for
debugging only.
Trace generation can be enabled by parameter '--trace'. Note that printing the
tracing information causes I/O overhead and might slow down the
solver. Writing traces in binary QRP format (enabled by parameter
'--trace=bqrp') usually produces smaller traces, as far as byte size is
concerned.
Calling DepQBF without command line parameters results in default behaviour
which was tuned on instances from QBFLIB. For performance comparisons with
other solvers it is recommended not to pass any command line parameters to
DepQBF.
By default, statistical output is disabled. To enable statistics, set the flag
'COMPUTE_STATS' in file 'qdpll_config.h' from 0 to 1. Similarly, time
statistics can be enabled by setting flag 'COMPUTE_STATS'.
--------------------------------------
IMPORTANT NOTES ON INCREMENTAL SOLVING
--------------------------------------
Please see the header file 'qdpll.h' for some documentation of the API
functions.
In applications which involve a very large number of incremental calls, the
overhead of maintaining the internal data structures in this release of DepQBF
might become non-negligible. In this case, please contact Florian Lonsing;
this is the first version of DepQBF which supports incremental solving and
hence your feedback is highly appreciated.
Incremental solving must be enabled by calling the API function
'qdpll_configure' with the parameters '--dep-man=simple' and
'--incremental-use', respectively. Please see also the example programs in the
subdirectory 'examples'.
The push and pop operations provided by the API allow to add and remove
clauses in a stack-based way. Therefore, clauses which are shared between many
formulas to be solved should be pushed onto the stack first. Clauses which
have to be removed soon should be added last so that they can be popped from
the stack easily.
In general, it is beneficial for the performance of the solver to avoid
needless push operations. For example, if you know that certain clauses will
never be removed from the formula then it is not necessary to call
'qdpll_push' before adding these clauses.
If assumptions are passed to the solver using 'qdpll_assume' AND clauses are
added later to the formula, then the API function 'qdpll_configure' must be
called with the parameters '--dep-man=simple' and '--incremental-use' after
the solver object has been created by 'qdpll_create'. Otherwise, if no clauses
are added, then the aforementioned calls of the API function 'qdpll_configure'
can be omitted.
----------
REFERENCES
----------
Florian Lonsing and Armin Biere. DepQBF: A Dependency-Aware QBF Solver. JSAT,
2010.
Aina Niemetz and Mathias Preiner and Florian Lonsing and Martina Seidl and
Armin Biere. Resolution-Based Certificate Extraction for QBF - (Tool
Presentation). In Proc. SAT 2012.
Florian Lonsing and Uwe Egly and Allen Van Gelder. Efficient Clause Learning
for Quantified Boolean Formulas via QBF Pseudo Unit Propagation. In Proc. SAT
2013.
Uwe Egly and Florian Lonsing and Magdalena Widl. Long-Distance Resolution:
Proof Generation and Strategy Extraction in Search-Based QBF Solving. In
Proc.LPAR 2013.
Florian Lonsing and Uwe Egly. Incremental QBF Solving. Technical report
submitted to arXiv Computing Research Repository (CoRR), February 2014.
-------
CONTACT
-------
For comments, questions, bug reports etc. related to DepQBF, please do not
hesitate to contact Florian Lonsing:
http://www.kr.tuwien.ac.at/staff/lonsing/
http://lonsing.github.io/depqbf/