[Coin-discuss] clp performance much worse with g++4.1 vs. g++4.0

[Ted Ralphs]

I just wanted to clarify for the list that although there seem to be
minor differences in the actual binary produced (Osi produces a
different binary than the other two), the performance for me is exactly
the same with all three versions.

Cheers,

Ted

Ted Ralphs wrote:
> Each of the three projects you mentioned includes Clp 1.6.0 as part of
> its distribution, so all three of them are building an executable called
> "clp" from the Clp 1.6.0 source code and installing that executable in
> the same directory (the one you specified), each one overwriting the
> last. You can tell specifically what other releases a given project
> release depends on by looking at the subversion externals of the project
> in the source browser on the project's Trac page. For instance, in the
> case of Cbc, check out the very bottom of this page:
> 
> https://projects.coin-or.org/Cbc/browser/releases/2.0.0
> 
> So the "clp" you are ending up with is the Clp executable built as part
> of the Cbc distribution. Although all the executables use the same
> source code in this case, it is possible that the build is configured
> differently in each case because the presence or absence of other
> projects can trigger automatic changes in the configuration options. It
> seems to be the case that, in fact,  the executables built in each case
> are slightly different (I just checked this). I'm sure John can tell you
> exactly what the differences are.
> 
> Cheers,
> 
> Ted
> 
> Stuart A Siegel wrote:
>>
>> After John mentioned that he doesn't see a difference I dug a bit
>> deeper.  It turns out that the compiler issue is a red herring, my
>> aplogies.  What's causing the performance difference is my
>> (apparently) incorrect build procedure.
>>
>> When I build just Clp (using ../Clp-1.6.0 prefix=[somedirectory] &&
>> make install) then performance is good from somedirectory/bin/clp
>>
>> If I build Clp, Osi-0.97.0, and then Cbc-2.0.0 in that order using the
>> **same** prefix then somedirectory/bin/clp performs three times worse
>> than when I only build Clp.  This result is independent of the g++
>> version being used.  I chose to build this way (all three into the
>> same install directory) because I had assumed that each provides a
>> unique set of shared objects, but my experience suggests that  this is
>> not the case. 
>> John, I'll be at Watson tomorrow (in fact, I think I'm across the hall
>> from you) so I can show you what I mean.  I will post the mps file on
>> my gsa account and send you a link in separate message.
>>
>> -----------------------------------------------------------------------------------------
>>
>> Stuart Siegel - IBM TJ Watson Laboratory 33-238
>> (M) 914.396.9506 (H) 973.379.1810 (O) 914.945.1674
>>
>>
>> ------------------------------------------------------------------------
>>
>> _______________________________________________
>> Coin-discuss mailing list
>> Coin-discuss at list.coin-or.org
>> http://list.coin-or.org/mailman/listinfo/coin-discuss
> 
> 


-- 
Dr. Ted Ralphs
Associate Professor
Industrial and Systems Engineering
Lehigh University
(610)758-4784
tkralphs at lehigh.edu
www.lehigh.edu/~tkr2