Re: [lmi] census benchmark test

[Vadim Zeitlin]

On Sun, 23 Nov 2014 20:32:12 +0000 Greg Chicares <address@hidden> wrote:

GC> On 2014-11-23 18:53Z, Vadim Zeitlin wrote:
GC> [...patches...]
GC> >  Anyhow, I hope this finally clears this particular issue, please let me
GC> > know if you have any questions and/or whether I should commit the patches
GC> > above or you prefer to do it yourself.
GC> 
GC> I committed both within the last few minutes.

 Yes, I've seen these commits, thanks!

GC> Here are the only interesting lines in the output:
GC> 
GC> 20:25:46: Run case for census_1: 419ms elapsed (expected -3.46%)
GC> 20:26:01:     benchmark_census: ERROR (Assertion 
'std::fabs(diff_in_percents) < 10' failed (expected 11000ms, but actually took 
15266ms, i.e. +38.78%). [file /lmi/src/lmi/wx_test_benchmark_census.cpp, line 
104] )
GC> 
GC> Those will go away when we stop comparing to config-file timings.

 On a related note, a potentially interesting bit of information: when I've
recently switched from using MSVS 2009 to using MSVS 2013 for compiling LMI
here, I had to change the timings used in the config file because it turns
out that the binary built with the latter is much (i.e. more than 10%)
faster than the one built with the former (~35% difference in the case
running time!). If the difference between the currently used g++ 3.4
(~2004) and 4.9 (2014) are of the same order of magnitude -- and knowing
that the gcc optimizer went into at least one profound architectural change
during this time, this is not at all impossible -- switching to a new
compiler could bring even more benefits than I thought it might.

GC> I know I haven't yet answered all your messages on that subject,
GC> but I will after the US holiday.

 TIA!
VZ