Hi Theodore,
On 15 Aug 2016 05:27:18 -0700
"theodorepringle1_at_gmail.com [fc-solve-discuss]"
<fc-solve-discuss_at_yahoogroups.com> wrote:
> Patsolve 3.0.3 with 6.3 GB of memory resolves the following deals as
> impossible:
>
> 1107600547
I've tried to solve this deal ( 1107600547 ) on my x86-64 8 GB machine. I
first tried with fc-solve with «../source/Tatzer -l ci7m » (which implies --rcs
and other settings), but it ran out of RAM - so I guess patsolve consumes less
RAM than it. Then I tried with the depth_dbm_fc_solver (which should be
even more RAM-conserving than --rcs after applying a local patch to make it
multi-threaded and use 4 freecells instead of 2), but I noticed that I left it
running overnight and it still consumed only 0.1% of RAM. A bug!
Today I finally got to fixing it and did a git bisect (see
https://twitter.com/shlomif/status/774259153883791360 for a related tip that
proved useful), and found the problematic commit and then applied a fix and
wrote a regression automated testcase to fix it.
After some time, I applied the local patch again and reran the
depth_dbm_fc_solve command line:
./depth_dbm_fc_solver --num-threads 4 \
--offload-dir-path /home/shlomif/tmp/depth-dbm/ 1107600547.board | \
tee 1107600547.depth_dbm.dump
it finished in well under my available RAM, and confirmed the deal was indeed
impossible. So I have to thank Theodore for having me discover (and later
fix) this bug.
==========
A direction we may wish to investigate is to run the depth_dbm_fc_solver
with the reported intractable deals on some HPC machines I have indirect access
to, and which have up to 512 GB of RAM to see if it can find if they are
solvable or not.
Regards,
Shlomi Fish
--
-----------------------------------------------------------------
Shlomi Fish http://www.shlomifish.org/
The conversation about how someone shouldn’t do something in an IRC channel is
always at least twice as long as the text the accused person created in the
first place — Chris62vw’s Rule
Received on Fri Sep 09 2016 - 16:16:35 IDT