[ADMB Users] More -> Fwd: Summary -> ADMB V11.0 build (compile/link) time using MinGW or Visual C++ 2010 Express in Windows 7 (32 bit) - very slow

[Larry Jacobson (NOAA Federal)]

Sorry, my original note summarizing approaches to reducing compile time 
for ADMB using the MinGW or Visual C++ 2010 Express compilersin Windows 
left out 3 important items.  Unfortunately, one cannot recompile 
e-mails.  In any case:

1) Optimization is turned off using the MinGW compiler by changing line 
line sym=-O3 in adcomp.bat so that it reads sym=-O.

2) Optimization is turned off using the Microsoft compiler by omitting 
-Ox from the first call to the compiler.  That is, use:
> cl -c /EHsc -DUSE_LAPLACE -DWIN32 %opt% -D__MSVC32__=8 -I. 
> -I"%ADMB_HOME%"\include -I"%ADMB_HOME%"\contrib %1.cpp
instead of:
> cl -c /EHsc -DUSE_LAPLACE -DWIN32 %opt% /Ox -D__MSVC32__=8 -I. 
> -I"%ADMB_HOME%"\include -I"%ADMB_HOME%"\contrib %1.cpp

3) The popular IDE for Windows uses the MinGW compilerwith a slightly 
different adcomp.bat file that is probably in a directory named 
something like C:\admb\admb101-gcc452-win32\bin. To turn off 
optimization in the IDE, change theline sym=-O3 in adcomp.bat so that it 
reads sym=-O(same as #1).  See Arni's note belowand an additional one 
from John Sibert below Arni's.

Cheers!

> Hi Larry,
>
> ADMB-IDE does not reinvent any wheels, so to compile models it just 
> invokes the "adcomp" shell script that comes with ADMB. In the 
> compilation window, you see something like:
>
> *** adcomp simple
> g++ -c -O3 -Wno-deprecated -D__GNUDOS__ -Dlinux -DOPT_LIB 
> -DUSE_LAPLACE -fpermissive -I. 
> -I"c:/admb/admb101-gcc452-win32\include" -o simple.obj simple.cpp
>
> The easiest way to change the compiler optimization options is to edit 
> the adcomp.bat script directly, after making a backup copy of the 
> original. Then you will see something like this in the compilation window:
>
> *** adcomp simple
> g++ -whatever the adcomp.bat script does now
>
> Given the importance of optimization options (btw, thanks for 
> analyzing that), we might modify the "adcomp" script in the near 
> future to allow the user to set optimization options outside the 
> script, like "adcomp -O2 mymodel". ADMB-IDE will then introduce a new 
> menu entry to set those options within the IDE.
>
> Arni

 From John Sibert:
>
> Another way to greatly decrease compile time for applications is to 
> break the source code into  smaller pieces by changing ADMB FUNCTIONs 
> into real C++ functions. It is generally considered good practice for 
> maintaining large software projects. It is not difficult to do, but 
> does require extra work and and climbing a bit of a learning curve for 
> the "make". A simple Makefile for ADMB projects hidden on the ADMB 
> website 
> http://www.admb-project.org/developers/admb-developers-workshops/admb-developers-workshop-march-13-16-2012/simple-makefile-for-admb.

-------- Original Message --------
Subject: 	Summary -> ADMB V11.0 build (compile/link) time using MinGW or 
Visual C++ 2010 Express in Windows 7 (32 bit) - very slow
Date: 	Mon, 25 Mar 2013 07:56:56 -0400
From: 	Larry Jacobson (NOAA Federal) <larry.jacobson at noaa.gov>
Reply-To: 	Larry.Jacobson at noaa.gov
To: 	ADMB Users <Users at admb-project.org>



My original question concerned speeding up ADMB builds using the MingGW
and Microsoft 2010 Express compilers with ADMB V11.0 on a 32 bit Windows
computer.

Rick Methot, Jim Ianelli, Alen Hicks, Johnoel Ancheta and Dave Fournier
responded.  All suggested reducing the level of optimization with the
cpp program gets compiled.  This is done in the adcomp.bat file.

Turning off optimization works wonders for compile time and has modest
effects on run time. Build time with the Microsoft compiler dropped from
8.25 minutes to 0.13 minutes (not a typo!) and the run time increased by
only 0.17 minutes.  Build time with the MinGW compiler dropped from 1.75
minutes to 1 minute and the run time decreased by about 0.25 minutes
(surprising but the MinGW web site said this can happen and my results
varied a bit from trial to trial). Some results are summarized in a
bitmap table below.

I wouldn't say that there was a clear winner between Microsoft and
MinGW.  It depends on the optimization level and whether you are
interested in time to build or time to run.  Both are free...

Rick Methot got the same results building the Stock Synthesis model
using the Microsoft compiler. Getting rid of the optimization shortened
the build time a lot but had little effect on run times.

Johnoel Ancheta indicated that the ADMBTeam is currently working on this
issue also.

Dave Fournier recommended getting a new computer or running in Linux. I
use the Linux option pretty routinely for ADMB models (for speed) but
didn't test it here.  I have noticed that the compile times on Linux are
longish too and I bet they get much shorter when the optimization level
is reduced.  I agree with him about the new computer but with
sequestration etc. on the horizon it is not going to happen.

Alan Hicks recommended (several times) using reduced optimization for
debugging/programming and full optimization for releases. Sounds like a
good idea to me. Some care seems advisable because turning off the
optimization may have unintended side effects.  I shouldn't have any
effect on error checking during the compilation step but what do I know?


Thanks everyone and cheers!



-- 
**********************
Larry Jacobson
National Marine Fisheries Service
Northeast Fisheries Science Center
166 Water Street
Woods Hole, MA 02543-1026
Voice: 508-495-2317
Fax: 508-495-2393
E-mail: larry.jacobson at noaa.gov
**********************

-- 
**********************
Larry Jacobson
National Marine Fisheries Service
Northeast Fisheries Science Center
166 Water Street
Woods Hole, MA 02543-1026
Voice: 508-495-2317
Fax: 508-495-2393
E-mail: larry.jacobson at noaa.gov
**********************