The ADPfusion package
ADPfusion combines stream-fusion (using the stream interface provided by the vector library) and type-level programming to provide highly efficient dynamic programming combinators.
From the programmers' viewpoint, ADPfusion behaves very much like the original ADP implementation <http://bibiserv.techfak.uni-bielefeld.de/adp/> developed by Robert Giegerich and colleagues, though both combinator semantics and backtracking are different.
The library internals, however, are designed not only to speed up ADP by a large margin (which this library does), but also to provide further runtime improvements by allowing the programmer to switch over to other kinds of data structures with better time and space behaviour. Most importantly, dynamic programming tables can be strict, removing indirections present in lazy, boxed tables.
As a simple benchmark, consider the Nussinov78 algorithm which translates to three nested for loops (for C). In the figure, four different approaches are compared using inputs with size 100 characters to 1000 characters in increments of 100 characters. "C" is an implementation ("./C/" directory) in "C" using "gcc -O3". "ADP" is the original ADP approach (see link above), while "GAPC" uses the "GAP" language (<http://gapc.eu/>). <<https://github.com/choener/ADPfusion/gaplike-performance.png>>
Please note that actual performance will depend much on table layout and data structures accessed during calculations, but in general performance is very good: close to C and better than other high-level approaches (that I know of).
Even complex ADP code tends to be completely optimized to loops that use only unboxed variables (Int# and others, indexIntArray# and others).
Completely novel (compared to ADP), is the idea of allowing efficient monadic combinators. This facilitates writing code that performs backtracking, or samples structures stochastically, among others things.
This version is still highly experimental and makes use of multiple recent improvements in GHC. This is particularly true for the monadic interface.
Newley added are the ADP.Fusion.GAPlike modules. These allow for writing grammars with only one (non)-terminal combinator. The logic for index manipulation is now moved into data types for terminals and non-terminals.
While this change leads to slightly more complicated instances for each new terminal or non-terminal, the overall code complexity is significantly lower. In addition, Constraint Kinds make complex interactions between (non)-terminals possible, while still managing to produce high-performance code.
The final goal would, of course, be to have no inter-terminal combinators anymore.
* GHC 7.6, LLVM, and -fnew-codegen recommended: gives a speedup of x2 for GAPcriterion
Long term goals: Outer indices with more than two dimensions, specialized table design, a combinator library, a library for computational biology.
Two algorithms from the realm of computational biology are provided as examples on how to write dynamic programming algorithms using this library: <http://hackage.haskell.org/package/Nussinov78> and <http://hackage.haskell.org/package/RNAFold>.
Changes since 0.0.1.2:
* require GHC 7.6
* ADP.Fusion.GAPlike module for (almost) combinator-less grammars
* ConstraintKinds for constrained parsers in GAPlike.
Changes since 0.0.1.0:
* compatibility with GHC 7.4
* note: still using fundeps & and TFs together. The TF-only version does not optimize as well (I know why but not yet how to fix it)
Using the new code generator?
The new code generator is not official yet, but I recommend trying it out: <<https://github.com/choener/ADPfusion/gaplike-newcodegen.png>>
|Versions||0.0.1.0, 0.0.1.1, 0.0.1.2, 0.1.0.0|
|Dependencies||base (4.*), ghc-prim, primitive (0.5.*), PrimitiveArray (0.4.*), vector (0.10.*)|
|Copyright||Christian Hoener zu Siederdissen, 2011-2012|
|Author||Christian Hoener zu Siederdissen, 2011-2012|
|Category||Algorithms, Data Structures, Bioinformatics|
|Source repository||git clone git://github.com/choener/ADPfusion|
|Upload date||Wed Nov 7 18:22:28 UTC 2012|