ver 2.5.1 -- 8/06/2010 --- 'The Obvious Child' * Printing updates: everyting is printed much more nicely now * Approximated degrees of literals in binary graph are not reset between calculations ver 2.5.0 -- 7/06/2010 (SAT Race'10 version) * A lot of performance bugs have been fixed. Activities of clauses were wrongly updated with their abstract representation during subsumtion for example. Also, we now use a well-tested set of magic constants instead of making them up using intuition. Apparently, intuition in the field of SAT leads to headaches and (in severe cases) to dementia. * A lot of code has been added regarding binary clause graphs. It is now regularly cleaned from useless binary clauses. Also, the useless binary clauses are regularly generated to subsume and strenghthen other clauses with them -- and once used, these useless binary clauses are thrown away. * Hyper-binary clauses are now generated using an algorithm relying purely on the datastrucures available in modern SAT solvers -- i.e. the fact that binary clauses have their own watchlists, and so allow for efficient propagation of the binary clauses separately from other clauses. ver 2.4.2 -- 9/05/2010 * Gaussian elimination has finally been fixed. It can now be tried out with the command-line switch "--gaussuntil=X", where X is the maximum depth. I usually set 100, but this is probably a wrong default. You should experiment with your own cipher. NOTE: Gauss is still experimental. If it segfaults, please file a bug. * The solver can now print out all the solutions to a problem. Simply use the "--maxsolutions=X" option, where X is the maximum number of solutions you need. You may use this option in conjunction with the very experimental "--greedyunbound" which unassigns some variables in such a way that the given solution is still correct, but some variables may not appear in it. * Command line switches have been corrected. They are now all lower-case and use the prefix "--" instead of "-" ver 2.4.1 -- 30/04/2010 Serious bug fixed that read data from change memory in subsume0, and hyper-binary resolution has been disabled, since it caused satisfiable instances to become unsatisfiable. ver 2.4.0 -- 26/04/2010 The first real release of CryptoMiniSat v2. It contains the following set of improvements: * XOR clauses are extracted at the beginning of the solving * Anti- or equivalent variables are detected at regular intervals and are replaced with one another, eliminating variables during solving * xor-clauses are regularly XOR-ed with one another such as to obtain binary XOR clauses. These binary xor-clauses are then treated as variable replacements instructions (i.e. "v1 XOR v2 = false" means that v1 is replaced with v2) * Phase calculation using Jeroslow and Wang, and phase saving with randomised search space exploration. The average branch depth is measured for each instance, and the solver makes a random phase flip with 1/avgBranchDepth probability * Random search burst are used to search unexplored areas of the search space at regular intervals * Automatic detection of cryptographic and industrial instances. Dynamic restart is used for industrial instances, and static restart for cryptographical instances. Detection is based on xor-clause percentage and variable activity stability. * Regular full restarts are performed to detect if the problem hasn't changed enough due to learnt clauses and assigned variables to behave more like a cryptographical instance than an industrial instance or vice-versa. * Both GLUCOSE-type learnt clause activity and MiniSat-type learnt clause activity heuristics are supported. During dynamic restarts, the GLUCOSE heuristic is used, while during static restarts, the MiniSat-type heuristic is used. * SatELite-type variable elimination, clause subsumption and clause strengthening is regularly performed. The occurrence lists are, however, not updated all the time such as the case is with PrecoSat. Instead, occurrences are calculated on per-use basis * On-the-fly subsumption is used to check whether the conflict clause automatically subsumes the clause that caused the conflict. * Binary clauses are propagated first before non-binary clauses are propagated. * 32-bit pointers are used for the watchlists on 64-bit architectures, using out the fact that most bits in the 64-bit pointer are actually fixed * Hyper-binary resolution is used when the hyper-binary clause subsumes any of the original clauses * Clauses are regularly scrubbed from variables that have been assigned * Preliminary blocked-clause elimination is used to remove pure literals * Distinct subproblems are regularly searched for and detected. These subproblems and solved with subsolvers. As a side-ntoe, this eliminates the original theoretical need for phase-saving (enabling the random flipping of phase, which is also used) * xor-clause subsumption is regularly performed * So-called dependent variables are removed along with their xor-clauses. This means that variables that only occur in one xor-clause and in no other clause are removed along with the XOR clause. Once the solving has finished, this xor-clause is re-introduced and a suitable value for the variable is found to satisfy the XOR. * Failed variable probing with both-propagated and binary XOR detection. All variables are successively propagated both to TRUE and FALSE. If one of these branches fails, the variable is assigned the other branch. If none fails, but the intersection of assignments is non-empty, those assignments are made. Essentially the same is done to non-binary XOR-s: if both v and !v propagate a given binary XOR, that XOR is learnt. * Designed to work as a library and as a drop-in replacement for MiniSat ver 2.3.2 -- 28/12/2009 * further ints have been replaced with uints * ZLIB can now be disabled * Visual C++ 2008 can now compile the sources * Statistics generation is much faster (thanks to Martin Maurer for spotting this) ver 2.3.0 -- 17/12/2009 * binary learnts are converted to 2-long xors if possible, and eliminated * lots of heuristics tuning * Cleanclauses is now default instead of removeSatisfied in simplify() * Stable in case used as a library * Lots of regression tests added * Cleaner logging * ints have been replaced with uints (less warnings with -Wall) * a lot of speedups for gauss -- packed, multi-matrix representation ver 2.2 -- 20/11/2009 * xor-clause finding * matrix finding * var-replacing * heuristics to disable gauss * much better+cleaner stats generation (e.g. fcopy.cpp removed) * lots of bug-fixing * satelite added, with cryptominisat_ext.sh as a wrapper script ver 2.1.1 -- 30/10/2009 * Learnt clause distribution stats * Added regression testing ver 2.1.0 -- 28/10/2009 * hand-made (non-GPL Bignum) packed representation of both matrix' rows * removed dependency on GPL Bignum library ver 2.0.1 -- 24/10/2009 * Added Gaussian elimination ver 1.2.6 -- 24/10/2009 * Corrected unitialised maxRestarts ver 1.2.6 -- 24/10/2009 * Corrected unitialised maxRestarts ver 1.2.5 -- 24/10/2009 * Maximum restarts can be configured * Better verbose debug printing ver 1.2.4 -- 22/10/2009 * CryptoMiniSat is printed as the first line of the program (instead of "This is MiniSat 2.0 beta") ver 1.2.3 -- 22/10/2009 * better README file * better use of the automake autoconf toolchain ver 1.2.2 -- 22/10/2009 * phase saving added (thank you, glucose solver team) * better printing of statistics * better explanation of statistics * accept 'v' and 'var', 'g' and 'group' * better parsing of 'v','var' and 'g','group' * don't allow too long group and variable names * branch length distribution added * better Makefile.cvs * cmake option added * updated INSTALL instructions * '-march=native' is default when using cmake ver 1.1 -- 29/04/2009 * Renamed to CryptoMiniSat ver 1.0 -- 15/04/2009 * Some updated statistics! Now average rank of guessed var is shown