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

Modsets

Caution: some of these may be compiler-dependent, or machine dependent. You get the idea: use with caution. My compiler is the fujistsu one.

I have indicated which modsets I wrote and which are not mine. Dates on mine are when they appeared here.

Alan Iwi is starting a "Modcenter" at http://www.amdi.rl.ac.uk/homepages/iwi/um/modcentre/. Some of my mods are there.

If you don't know how to write or use modsets, see here.

General

1. lux_open.mod [not mine] - fixes nasty bug with obscure effects about re-opening files. If unfixed leads to bizarre crash on closing some files (unit 8?).
2. wmc-set-decomp.upd [2002/03/07] - sets the decomposition based on the number of processors given
3. linux_ocean.mod - misnamed mod that might make the timer work under MPP - I haven't checked.
4. linuxf_mpp.mod - not vital but probably helps the model avoid unimportant but irritatig segfault at end of run.
5. filter1a.mod - modifies the fourier filtering to effectively weaken it by a factor of 4 to how it should be. See yabce for details.
6. turing-type-mods.tar.gz - tarball of a whole pile of "generic" hadcm3 vn4.5 mods that I found on turing. Err, hope that OK with everyone. Many of them are probably unnecessary for most purposes. Here is a contents list.
7. setfil-write.upd - diagnostic to make SETFIL write out how much of the globe it is filtering.
8. ftj1f405 - allows ozone to be a diag (you also need ozone_4.5 user stash file).

Utilities mods

1. adh2f405 - from DH/RJ at the Hadley Center. Allows recon of 3-d ozone fields.
2. abort1a-ncpr.upd - [2002/07/12] - makes having a non-constant polar row non-fatal. I use it for building convpp, since the problem is clearly only meant to be a warning, in the case of diags. Necesary for EVP diag 32317.
3. compare1-32.upd - [29/05/2002; only for utility cumf, not the model] - upgrades some local variables to 64-bit to prevent cumf falling over on the energy diag calc.
4. convppfix.upd - [22/07/2002; conf utility convpp] - remove problem leading to ocean diags losing 2 points per row (related to haloes)

MPP modsets

1. ibuflen.upd - [2002/02/11; revised 2002/03/04] - fixes bug in MPP code when calculating max length of field in D1 array to write. If you have climate meaning turned on, you will be overwriting memory without this mod. May not be needed without climate meaning. [nb: will not compile into non-MPP code in its current form]. [2002/03/25: nb ibuflen.upd is subsumed within gsm3f406, a standard mod, if you use that].
2. wmc-fix-inittrac.upd - [2002/02/11] - fixes problem in INITTRAC which caused hanging under MPP.
3. wmc-set-env1.upd and wmc-set-env-c.upd - [2002/04/19] - These mods are my way of making the code work under MPP. The problem is to propagate the environment to all processors. I write it to file (before starting the job) then all procs read it, using those mods. Andy Heaps has a neater way of doing it (but I couldn't make his way work. If you can't either, read this).
4. qtpos1a.upd - [2002/02/25 upgraded 2003/03/21] - under my compiler and mpich, GC_ISUM *doesn't work* (at 64 bit). Argh... This modset makes qtpos1a use GC_RSUM (which does work) instead. Note that this mod has recently been upgraded to remove a flaw (which didn't stop it working but was dodgy). See qtpos1a.upd,v its CVS file. Also see gcom version 2.9b5 which is probably a better fix to this problem.
5. timer3a-return.upd - trivial mod to make the timer null. May fix hanging under MPP - probably not required.
6. timer3a-trace.upd (and timer1a-trace.upd for non-MPP) - another trivial mod, that makes the timer print out what routines have been called. Useful as a (partial) tracer if the code is crashing. Note that not all routines call timer.
7. blokinit.upd - see below under "seaice"

Bit Comparison modsets

1. bitcf-uvadv1c.upd [2002/02/18] - changes UVADV1C.dk to revert to the behaviour of UVADV1B.dk. Scientifically null, but changes the bits. With this, runs should be compare between single-proc and MPP-1-proc code (at least for a few days).

32-bit modsets

1. lux_32bit.mod [not mine]
2. omit-tracer-calc.upd - removes a diagnostic (I think) calc in tracer which appears to generate FPE errors at 32-bit.

Seaice modsets

1. fixdrift.upd - fix (minor?) bug in ICEDRIFT.
2. cnvstop.upd - [29/05/2002] - provide missing CNVSTOP deck.
3. bench_EVP.mf77.vn4.5 - [1/7/2002; 29/05/2002] - from Doug Cresswell - add EVP to UM. [This needs a bug fix (if you use SSTILT), edited in. Read the rcs file if you like: bench_EVP.mf77.vn4.5,v]
4. blokinit.upd - [19/06/2002; update 2003/04/16: fix 32->23; thanks AI] - fixes JMT that should be JMTM1 in blokinit. Crashes the 32-bit coupled model if not present (symptom (under fujitsu with 2 procs): dies in ocn_ctl or end of ice_ctl with SIGSEG11). The bug doesn't seem to show up in the 64-bit model, but it may just be doing evil behind the scenes.

Compiler specific modsets

Portland

1. fixmeanctl.mod [from Alan Iwi] - prevents FPE crash in meanctl - apparently works around a bug in the compiler. In fact, I find that compiling meanctl with debug (ie, sans optimisation: remove -fast too) fixes the problem too.

Misc notes

1. gsm3f406 is *only* suitable for MPP use (a corrected one, probably suitable for non-MPP, is here. The changes are small).
2. volvar_right_new is *only* suitable for MPP use (ditto volvar, volvar_right...). BUT you probably don't want to be using these anyway: they are for climate change runs only.