Greg,
This will definitely give an improvement over the behavior I've been
seeing, but it is slightly less efficient than having flopsie exist as a
.so file, but probably not significant to warrant concern.
Where did everyone find out about the other versions of the Intel
compiler?
-Jon
Greg Daues wrote:
On Wed, 4 Oct 2006, Jonathan Myers wrote:
All,
Today we finally got to the bottom of the confusion regarding
flopsie, the CPU resource consumer. A while ago I found that it was
not behaving nearly as well as when Tim had originally tested it, and
that it was much closer to the speed he had seen when it was compiled
as a command-line application than when it was a shared object file
attached to Python.
Now we have discovered that it will behave the way Tim had seen in
the past, as a Python extension, if you are using icc v. 8.1 to
compile it, not icc v. 8.0, which is the default for TeraGrid. To
get yourself using icc v. 8.1, add
Hi Jon,
For quite a while now (weeks really) we have been using this
setting in the ~/.soft (as per Robyn's spec) and thus compiling
flopsie with the 8.1 icc. We were then placing the flopsie binary in
the PATH. (Thus, it
was not compiled as a Python extension.)
Is this consistent with the theory on the performance results ?
Greg
+intel-c-8.1.035-f-8.1.031-r1
to your ~/.soft file, then log back in. Now you can update
apps/Consumer to r. 560 and then run
./setup build
./setup install --lsst-devel
and you won't have to bother with make or copying anything to your
bin dir.
I'm testing it now to make sure everything works and hopefully
establish whether or not the speedup is appreciable enough to make a
difference for DC1.
-Jon
_______________________________________________
LSST-data mailing list
[email protected]
http://www.lsstmail.org/mailman/listinfo/lsst-data
_______________________________________________
LSST-data mailing list
[email protected]
http://www.lsstmail.org/mailman/listinfo/lsst-data
_______________________________________________
LSST-data mailing list
[email protected]
http://www.lsstmail.org/mailman/listinfo/lsst-data
