80 lines
3.3 KiB
Plaintext
80 lines
3.3 KiB
Plaintext
|
Release Notes for MiniSat 2.2.0
|
||
|
|
===============================
|
||
|
|
|
||
|
|
Changes since version 2.0:
|
||
|
|
|
||
|
|
* Started using a more standard release numbering.
|
||
|
|
|
||
|
|
* Includes some now well-known heuristics: phase-saving and luby
|
||
|
|
restarts. The old heuristics are still present and can be activated
|
||
|
|
if needed.
|
||
|
|
|
||
|
|
* Detection/Handling of out-of-memory and vector capacity
|
||
|
|
overflow. This is fairly new and relatively untested.
|
||
|
|
|
||
|
|
* Simple resource controls: CPU-time, memory, number of
|
||
|
|
conflicts/decisions.
|
||
|
|
|
||
|
|
* CPU-time limiting is implemented by a more general, but simple,
|
||
|
|
asynchronous interruption feature. This means that the solving
|
||
|
|
procedure can be interrupted from another thread or in a signal
|
||
|
|
handler.
|
||
|
|
|
||
|
|
* Improved portability with respect to building on Solaris and with
|
||
|
|
Visual Studio. This is not regularly tested and chances are that
|
||
|
|
this have been broken since, but should be fairly easy to fix if
|
||
|
|
so.
|
||
|
|
|
||
|
|
* Changed C++ file-extention to the less problematic ".cc".
|
||
|
|
|
||
|
|
* Source code is now namespace-protected
|
||
|
|
|
||
|
|
* Introducing a new Clause Memory Allocator that brings reduced
|
||
|
|
memory consumption on 64-bit architechtures and improved
|
||
|
|
performance (to some extent). The allocator uses a region-based
|
||
|
|
approach were all references to clauses are represented as a 32-bit
|
||
|
|
index into a global memory region that contains all clauses. To
|
||
|
|
free up and compact memory it uses a simple copying garbage
|
||
|
|
collector.
|
||
|
|
|
||
|
|
* Improved unit-propagation by Blocking Literals. For each entry in
|
||
|
|
the watcher lists, pair the pointer to a clause with some
|
||
|
|
(arbitrary) literal from the clause. The idea is that if the
|
||
|
|
literal is currently true (i.e. the clause is satisfied) the
|
||
|
|
watchers of the clause does not need to be altered. This can thus
|
||
|
|
be detected without touching the clause's memory at all. As often
|
||
|
|
as can be done cheaply, the blocking literal for entries to the
|
||
|
|
watcher list of a literal 'p' is set to the other literal watched
|
||
|
|
in the corresponding clause.
|
||
|
|
|
||
|
|
* Basic command-line/option handling system. Makes it easy to specify
|
||
|
|
options in the class that they affect, and whenever that class is
|
||
|
|
used in an executable, parsing of options and help messages are
|
||
|
|
brought in automatically.
|
||
|
|
|
||
|
|
* General clean-up and various minor bug-fixes.
|
||
|
|
|
||
|
|
* Changed implementation of variable-elimination/model-extension:
|
||
|
|
|
||
|
|
- The interface is changed so that arbitrary remembering is no longer
|
||
|
|
possible. If you need to mention some variable again in the future,
|
||
|
|
this variable has to be frozen.
|
||
|
|
|
||
|
|
- When eliminating a variable, only clauses that contain the variable
|
||
|
|
with one sign is necessary to store. Thereby making the other sign
|
||
|
|
a "default" value when extending models.
|
||
|
|
|
||
|
|
- The memory consumption for eliminated clauses is further improved
|
||
|
|
by storing all eliminated clauses in a single contiguous vector.
|
||
|
|
|
||
|
|
* Some common utility code (I/O, Parsing, CPU-time, etc) is ripped
|
||
|
|
out and placed in a separate "utils" directory.
|
||
|
|
|
||
|
|
* The DIMACS parse is refactored so that it can be reused in other
|
||
|
|
applications (not very elegant, but at least possible).
|
||
|
|
|
||
|
|
* Some simple improvements to scalability of preprocessing, using
|
||
|
|
more lazy clause removal from data-structures and a couple of
|
||
|
|
ad-hoc limits (the longest clause that can be produced in variable
|
||
|
|
elimination, and the longest clause used in backward subsumption).
|