The New Sudoku Players' Forum

[Moschopulus]

Hi gsf,
how fast is your solver?
On average, would it solve 2000 puzzles per second, 4000, or more?

[Sue De Coq]

It's much faster to find multiple solutions to a single puzzle, which typically requires Araidne's Thread to be unwound just a few steps, than to solve puzzles from scratch, which requires Ariadne's Thread to be built up over the entire grid. Even so, the gsf solver is still seriously fast. My solver uses JDK 1.5, which one would expect to run slower than the optimized native code used by gsf, but not by the observed factor. Back to the profiler ...

[gsf]

how fast is your solver?
On average, would it solve 2000 puzzles per second, 4000, or more?

for the 100K 1-constrained puzzles in
http://www.research.att.com/~gsf/sudoku/FNBTXYW-1-con.dat.gz
about 19K/s assuming valid (1 solution) puzzles, about 13K/s with the uniqeness check, on a 2.8Ghz P4

solver command lines to reproduce the results:

Code: Select all

sudoku -d -qFN -f- -Fpuzzles=%n%,guesses=%Q%,iterations=%I%,seconds=%t FNBTXYW-1-con.dat

sudoku -u -d -qFN -f- -Fpuzzles=%n%,guesses=%Q%,iterations=%I%,seconds=%t FNBTXYW-1-con.dat


-qFN enables the simplest F and N constraints
-d enables depth first search
-u verifies validity (exactly one solution)

[gsf]

It's much faster to find multiple solutions to a single puzzle, which typically requires Araidne's Thread to be unwound just a few steps, than to solve puzzles from scratch, which requires Ariadne's Thread to be built up over the entire grid. Even so, the gsf solver is still seriously fast. My solver uses JDK 1.5, which one would expect to run slower than the optimized native code used by gsf, but not by the observed factor. Back to the profiler ...

right, ~300K/s for multiple solutions to one puzzle, vs ~19K/s from scratch

[Lardarse]

It's something to do with chickens.

That wasn't what I was thinking, lol