[Science]   [BAS home]   [Met home]   [Beowulf home]	Antarctic Meteorology

Platforms

Note: except if I qualify them, all run ties are on 2 proc, since this is most efficient under ethernet interconnect.

Athlon (32 bit native) fujitsu compiler

See misc for compile vars and config files for this.

Opteron (64 bit native) portland compiler

Built with gcom_2.9b8 and mpich-1.2.1 (probably).

Note: in effort to halt drift, tried compiling with -O1 and -O0. It runs slower (of course) but (remarkably) is bit comparable.

The compile vars are: here.

See yacje

Itanium (64 bit native) intel compiler

Don't have compile vars yet, using: "ifort -r8 -i8 -c -g -align -CB". (-g and -CB there to try to catch errors: doesn't help).

On compilation it complains about stuff in intfctl1.f (repeat dec of INTEGER J), meanctl1.f (repeat dec of tag), decmfltr.f (JMT_GLOBAL as int occurs after first use).

Itanium (64 bit native) g95 compiler

Don't have compile vars yet, using: "g95 -O2 -r8 -fno-second-underscore -c".

Notes:

1. g95 is a free compiler. I downloaded the itanium 64-bit-int's version.
2. Don't use -i8. This is unneeded, and (due to a bug) demotes logicals to 4 bytes.
3. At runtime it will die in meanctl with the unit 8 stuff. I don't use this, so its best just deleted. See the lux_open mod for what to kill.
4. The SW and LW (units 57 and 80) namelists don't work as written: g95 doesn't allow reads of the form a(1,1,1,1)=1,2,3, to mean a(1:3,1,1,1)=1,2,3,. This is a bit of a pain and needs the namelist files re-written.
5. A few other minor compilation issues, mostly bugs in the code it found.
6. Its slow (at a guess, it should be about 3 times faster. Also slow to compile), but the diags are good.
7. g95 appears to be incompatible with Electric Fence (-lefence) I think because of malloc/free stuff conflicts. But thats OK, because you don't need efence, since it works...