Message1607

Author malte
Recipients erez, malte, silvia
Date 2011-08-13.19:12:29
Content
For future reference, here's the relevant excerpt of the original email that
discussed -m32 vs. -m64 memory usage:

===========================================================================
as a sanity test for our packaging, I did some experiments with the
final seq-opt-bjolp package this night, and noticed that it lost 17
(IIRC) problems compared to my previous tests, which is rather a lot.

Looking into this a bit more deeply, this seems to be due to -m64 using
a lot more memory than -m32. My tests were run with 2 GB (so this would
not be such an issue with the 6 GB in the competition), but still the
memory usage difference is rather large, and maybe we should do
something about this in the long run. I don't know if this is typical
behaviour or specific to LM configurations.

Example for the four toughest solved tasks in woodworking:

 #04: uses  737 MB with -m32, 1192 MB with -m64
 #14: uses 1277 MB with -m32, runs out of memory with -m64
 #15: uses 1583 MB with -m32, runs out of memory with -m64
 #24: uses  322 MB with -m32,  514 MB with -m64

I think there's nothing we could or should do about this for the
competition, but maybe it's worth opening an issue on this for later.
===========================================================================
History
Date User Action Args
2011-08-13 19:12:29maltesetmessageid: <1313255549.92.0.913480067209.issue213@gmail.com>
2011-08-13 19:12:29maltesetrecipients: + malte, erez, silvia
2011-08-13 19:12:29maltelinkissue213 messages
2011-08-13 19:12:29maltecreate