The New Sudoku Players' Forum

[blue]

Hi champagne,

This an amazing performance. Far from the pass1 (may be 30 times faster) run on my side. I did not reach your performance so far for the 18 enumeration, but we are in the same order of magnitude.

It was ~80 times faster than the pass1 code on m side, so definitely "same order of magnitude".
You've done well, congratulations !

Blue.

[champagne]

Hi Blue,
More to explain why I asked these questions. Here is the start

That took (43 days) x (12 cores), and it produced 323,897,702 ED puzzles..

As agreed later, this is a variant of the process applied to scan solution grids as in the pass 1 of the 17 clues scan.
with such results, this leads to more than 120 solution grids per second and per core.

My best results in the current test of a the band index 32 is around 20 solution grids per second. The code is available here

This is a pass1 in the 17 clues search with a code not too far in performance from your DLL when extended to a 18 search.

The hardware can not explain this huge difference, so there is likely a new breakthrough hidden in your results.

EDIT: the direct ratio 20 to 120 omits the fact that blue's run checks all bands and stacks. In my run, only one band is checked for >=7 clues

[JPF]

Your list includes 853357 non-minimal puzzles (out of 3142658), and mine only includes minimals.
After that, there are only 28777 that aren't in my list.

Thanks.
Indeed, my list includes some non-minimal puzzles but not all, which is inconsistent.
I have modified my file and the link accordingly.
What is striking, moreover, is the low contribution of my list to yours, which can be explained by the similarity of my method with yours (for the neighborhood search part).

JPF