Re: mesa-users] MESA install is taking forever to complete

[Bill P. <pa...@ki...>]

Hi Aaron,

On Jan 28, 2009, at 9:37 AM, Aaron Dotter wrote:

> Just a quick question: I'm trying to install mesa on my desktop  
> machine, compiled
> with gfortran because I don't have ifort, and 'tester' is taking a  
> VERY long time to
> complete.  Do you have any sense of how long it should take?

Minutes on my Intel mac powerbook.

> Right now it is over
> 24 hours on an AMD 2.4 GHz processor.  What is tester doing?

Each package in mesa has the same layout and the same installation  
steps.
And part of the installation is running a test to see if things look  
okay.
That test is done in the 'test' directory of the package; you'll find  
a test/make/makefile too.
The makefile for test creates a little test program called 'tester'.
So each each package, during installation a new 'tester' is created  
and run.
The output from tester is compared (using ndiff) to the expected  
output (in 'test_output').

> Obviously for real use of the code I will run it on a machine with  
> ifort but for
> testing and exploring the code I want to be able to run it on my  
> machine.

My guess is that you have an old gfortran.   There was a bug that  
caused the thing to hang like that.

You might want to update from http://gcc.gnu.org/wiki/GFortran

I'm currently using the following gfortran without problems:

/Users/billpaxton/mesa/star/test: gfortran -v
Using built-in specs.
Target: i386-apple-darwin8.10.1
Configured with: /tmp/gfortran-20081006/ibin/../gcc/configure -- 
prefix=/usr/local/gfortran --enable-languages=c,fortran --with-gmp=/ 
tmp/gfortran-20081006/gfortran_libs --enable-bootstrap
Thread model: posix
gcc version 4.4.0 20081006 (experimental) [trunk revision 140904] (GCC)


Please let me know if updating fixed things. I'll add a warning to  
the installation documentation about old gfortrans.

BTW: at one time I ran into trouble with the gfortran implementation  
of OpenMP.
I haven't gone back to check for a long time to see if things are ok  
now, but
in the meantime I'm suggesting that people using gfortran disable  
multi-threading.

setenv OMP_NUM_THREADS 1

Hopefully that problem will go away since it's great fun to use all  
the cores,
and mesa does a pretty good job of that when given a chance.

Thanks,
Bill