Hi,
The following correspondence may be of use for those investigating
different optimization options.
Steve Griffies
-----------------------------------------------------------------------------
Hi,
Has anyone tested O3 optimization? I know that there is some
loss of accuracy with this level of optimization, but on the SGI that I
used to use for shallow water modeling, we decided that we were happy
with it. We also use 03 on the IBM at UVic.
I think that it makes a noticeable difference in execution speed, but I
havent tested it.
Harper Simmons
---------------------------------------------------------------------
Harper,
I have compiled using -O3 on the IBM. I have not made long enough nor
detailed enough measurements to know if or how performance might be
affected.
The particular machine on which I happen to be running had some resource
limitation which I have not yet identified. This caused the compilation
of
mpp_domains.F90 to run out of resources at -O3 even with -qmaxmem=-1
(which
implies use of all allowed resources). Jon Wolfe has reported that he
had no
such problem on his IBM system at -O3.
regards,
jeff durachta ([EMAIL PROTECTED])
---------------------------------------------------------------
Hi,
The reason that I am not using -O3 on the SGI is that it didn't help
the
execution time with HIM. Some subroutines were faster with -O3, while
others were actually slower. If someone were so inclined, they could
probable set different subroutines to use -O2 or -O3, although the
preferred choice for some subroutines would probably be a function of
local domain size.
When this lack of benefit is combined with the potential to change
answers and an unambiguously longer compile time, it didn't seem to make
much sense to use -O3 everywhere.
- Bob Hallberg ([EMAIL PROTECTED])
---------------------------------------------------------
Hi,
Indeed, our experience on the SGI with MOM4 is completely consistent
with Bob
H. Note that there is no standard. "-O3" is different for different
compilers. Indeed even between releases of a compiler, one can find
changes
to what is attempted at any given level of optimization.
regards,
Jeff Durachta
begin:vcard
n:Griffies;Stephen
x-mozilla-html:FALSE
org:DOC/NOAA/OAR/GFDL
adr:;;;;;;
version:2.1
email;internet:[EMAIL PROTECTED]
title:Physical Scientist
x-mozilla-cpt:;-11168
fn:Stephen Griffies
end:vcard
