I've used the intel compilers on various clusters, always with qt
disabled. it's always been fast enough that I've not given it a second
thought. Have you watched top during the build? This may give some insight.
On 09/18/2012 12:01 PM, David E DeMarle wrote:
Last time I used icc (two years ago) I vaguely recall that compile
time was slower but it couldn't have been that much slower. I would
have remembered that. The paraview-manta render speed tests I was
doing back then showed that icc's compiled result was a percent or two
faster. This was on an intel based linux cluster.
About the same time frame I did have trouble with the pgi compilers on
crays. Burlen likely had recent knowledge about the differences.
On Sep 18, 2012 1:18 PM, "Hodge, Neil E." <[email protected]
<mailto:[email protected]>> wrote:
David:
-----Original Message-----
From: David E DeMarle <[email protected]
<mailto:[email protected]>>
Date: Monday, September 17, 2012 2:15 PM
To: Neil Hodge <[email protected] <mailto:[email protected]>>
Cc: "[email protected] <mailto:[email protected]>"
<[email protected] <mailto:[email protected]>>
Subject: Re: [Paraview] Building Paraview from source
>Are you compiling on a lustre file system? That can significantly
slow
>the compilation since the parallel filesystem is optimized for write
>throughput of large files. On ORNL jaguar in the NFS mounted
>directories it takes 10 minutes or so to compile. Compiling on the
>lustre mounted directories takes an hour or more. (10% per hour is a
>bit worse than what I've ever seen on the older trilab computers
so it
>may be something else). I get around this by compiling from the fast
>filesystem and copying the result to the lustre partition where the
>back end nodes can see it.
>
>Try "module unload altd". This is a profiling library that some
of the
>big machines use recenty that does not like "make -j anything". Once
>that is removed you can make -j to your hearts content.
>
No, I was not on the Lustre filesystem. Regardless, I moved
everything to
my home directory. This did not help the speed issue much.
Removing the kernel module did reduce the dependency problems
somewhat,
but not completely.
FYI, the essence of my build script follows:
cmake ../ParaView-3.14.1-Source \
-DCMAKE_C_COMPILER=icc -DCMAKE_CXX_COMPILER=icpc \
-DCMAKE_Fortran_COMPILER=ifort \
-DCMAKE_INSTALL_PREFIX:PATH=$HOME/PARAVIEW \
-DHDF5_INCLUDE_DIR=$HOME/HDF5/include \
-DHDF5_LIBRARY=$HOME/HDF5/lib/libhdf5.so \
-DPARAVIEW_ENABLE_PYTHON=ON
make -j 4
make install
However, after a bit of experimentation, I see that when I remove
the line
-DCMAKE_C_COMPILER=icc -DCMAKE_CXX_COMPILER=icpc
i.e., when I use the gnu compilers, the build seems to proceed at
a rate
that indicates a total build time of maybe 30 minutes or so.
So, the question is, are you aware of any problems with using the
intel
compilers to build paraview??? Thanks.
Cheers,
Neil
======================================================================
Neil Hodge, Ph.D.
Methods Development Group
Lawrence Livermore National Laboratory
_______________________________________________
Powered by www.kitware.com <http://www.kitware.com>
Visit other Kitware open-source projects at
http://www.kitware.com/opensource/opensource.html
Please keep messages on-topic and check the ParaView Wiki at:
http://paraview.org/Wiki/ParaView
Follow this link to subscribe/unsubscribe:
http://www.paraview.org/mailman/listinfo/paraview
_______________________________________________
Powered by www.kitware.com
Visit other Kitware open-source projects at
http://www.kitware.com/opensource/opensource.html
Please keep messages on-topic and check the ParaView Wiki at:
http://paraview.org/Wiki/ParaView
Follow this link to subscribe/unsubscribe:
http://www.paraview.org/mailman/listinfo/paraview
