[MITgcm-support] ifort optimization puzzle
Martin Losch
mlosch at awi-bremerhaven.de
Tue Mar 8 02:47:38 EST 2005
Hi,
together with Michael Schodlok I have set up a 1D experiment: 1 column
of water (nx=ny=1, overlap=2, nr=30, dz=30*10). We specify zero net
surface heat flux (qnet = 0), but -50W/m^2 shortwave heat flux (warming
due to the sun), in order to drive kpp. This works all very nicely and
we were happy, the mean temperature was constant, as it should (why
didn't we stop there?). Then I suggested to move to a faster machine
(from a $%^&* SUN to a nice linux-box with ifort). We used the standard
ifort build-options file with paths to netcdf appended. But suddenly
the system was losing heat at 1200W/m^2. Turning off optimization (-O0)
or using g77 gave the old results (with constant mean temperature). We
clearly have an optimization problem here. How serious do you think
this is (I admit that we are using a somewhat pathological
configuration)?
Martin
PS. in sea.mit.edu:/data4/mlosch/test1d.tgz I have a case similar to
our original one, but with different initial conditions for T and S,
here the mean temperature decreases by 0.5degC in 10 days,
corresponding to approximately 690W/m^2 heat loss
build_g77 has the Makefile for g77 (result in run00/mnc_0001)
build_o0 for ifort -O0 (result in run00/mnc_0002)
build_o1 for ifort -O1 (gives almost the same results as -O3, result in
run00/mnc_0003)
data files etc. in run00
I included the build option files I used, but I manipulated the
makefiles afterwards to change the optimization level for ifort.
More information about the MITgcm-support
mailing list