--- MITgcm/doc/tag-index 2010/09/24 23:25:45 1.1303 +++ MITgcm/doc/tag-index 2011/01/27 20:52:10 1.1370 @@ -1,30 +1,379 @@ -$Header: /home/ubuntu/mnt/e9_copy/MITgcm/doc/tag-index,v 1.1303 2010/09/24 23:25:45 gforget Exp $ +$Header: /home/ubuntu/mnt/e9_copy/MITgcm/doc/tag-index,v 1.1370 2011/01/27 20:52:10 jmc Exp $ $Name: $ Notes on tags used in MITgcmUV ============================== +o pkg/gmredi: + - apply exchange to GM_PsiX,GM_PsiY to get valid values near CS-corners + (needed for multi-dim advection). +o model/src: + - fix wrong re-setting of addMass array (ALLOW_ADDFLUID) for atmospheric run. + +checkpoint62r (2011/01/24) +o testreport: + - add option -deloutp (-do) to remove output file after successful run + - add option -mfile (-mf) for MPI machile-file and for each test-exp, + extract exact list of procs from this file ; adjust COMMAND by + replacing TR_MFILE & TR_NPROC with local machile-file & Nb of MPI procs. +o pkg/ecco: replace old ineffective way of penalizing deviations from + first guess (of obcs) with code by Matt Mazloff +o pkg/ctrl: add code that allows the use of ALLOW_OBCS?_CONTROL without + exf/cal + - ctrl_init: fix initialisation (computation of diffrec) for obcs-ctrl + and handle case of obcs?period == 0 + Not affected by these changes: There is still an inconsistency in + the computation of start/endrec for the obcs variables for ALLOW_CAL + defined (see comment by Menemenlis, 7-Mar-2003) + - make S/R ctrl_get_gen_rec return something meaningful even if + ALLOW_CAL is not defined + - ctrl_getobcs?: simplify code so that it does not need to use + S/R exf_swapfields +o testreport: + allow to test MPI with different number of processors for each experiment: + create file "SIZE.h.mpi" in each build dir. + Note: update of "-command" argument not yet implemented. +o model/src, pkg/autodiff, pkg/ecco, pkg/seaice: + alleviate the need for additional 3D tapes when using + the non-inear free surface in the adjoint. This is done + by adding an 'update' of hfacc etc. to their current + value at the beginning of forward_step.F. +o pkg/seaice + seaice_lsr: move stress-divergence computations out of outer pseudo- + timestepping loop +o pkg/obcs/obcs_prescribe_read.F: move 'CALL OBCS_EXTERNAL_FIELDS_LOAD' + up, so that the calls of the ctrl_* subroutines are really at the end + of the routine. +o pkg/gmredi: + allow to specify a 2-D horizontal map and a 1-D vertical profil + as scaling factor for Isopycnal diffusivity (Redi) and GM diffusivity. +o pkg/seaice: + move lsr iterations from seaice_dynsolver to seaice_lsr. +o pkg/seaice: + reduce memory footprint of seaice mutldim advection tapes, + by using a 2D tape independent of the generic_advdiff ones, + and setting GAD_AREA=1 etc. rather than GAD_AREA=101 etc. + +checkpoint62q (2010/12/29) +o model/src: + - check horizontal grid-spacing for unset or zero/negative values. + - change AD test hs94.1x64x5 to be truly a 2-D set-up. +o pkg/flt: + - use (recent) mdsio S/R MDS_READVEC_LOC instead of local FLT_MDSREADVECTOR + - precision of input and output files set according to readBinaryPrec + and writeBinaryPrec (previously always 64.b), pickup remains at 64 bits. + - initial float position from tiled file was not converted to index space + (looks like a bug): fixed to make it similar to initial pos. global file. +o global_ocean.cs32x15: + change SIZE.h of AD set-up to get same tiling as with MPI. + Update output_adm, output_adm.thsice (pass@14) & output_adm.seaice (fail@11) +o model/src: + add flag to control printing of domain fields (bathymetry, hFac ...) in + ini_mask_etc (call to plot_fieds_xy*). +o pkg/thsice: + if useEXF, some forcing fields are not valid in overlap: fixed by computing + over the interior only + apply EXCH at the end (like with SEAICE advection). +o pkg/mdsio: + - add 2 arguments to S/R MDS_PASS_R4/8toRL/S for overlap size of buffer + to copy to/from, so that it can be used to replace MDS_PASS_WH_R8toRL/RS. + - change barrier call for safe multi-threads access to 3-D shared buffer. + - MDS_READ/WRITEVEC_LOC use buffers from MDSIO_BUFF_3D.h (instead of local ones) + - replace call to MDS_PASS_WH_R4/8toRL with call to MDS_PASS_R4/8toRL and + remove the 4 S/R MDS_PASS_WH_R4/8toRL/S. +o pkg/flt: + - add fourth-order Runge-Kutta integration + - add basic exch2 support (only 1 facet supported so far!) +o pkg/mdsio+rw: + new S/R to read local (or tile independent) vector (read counterpart of + mds_writevec_loc, and front-end write_glvec_rl/rs). +o global_ocean.cs32x15: additional input_ad.seaice experiment + that uses pkg/seaice rather than pkg/thsice. + (using multidim advection for seaice) +o towards more F95 syntax for array specification in namelist: + - change "data.diagnostics" (data.ptracers was changed before checkpoint62p) + of all verification experiments. + e.g.: line: fields(1,3)='ETAN ','ETANSQ ', + was changed to: fields(1:2,3)='ETAN ','ETANSQ ', + - finish adding "-DNML_EXTENDED_F77" to DEFINES list in pgf77 and g77 optfiles +o global_ocean.cs32x15: additional input.seaice experiment + that uses pkg/seaice rather than pkg/thsice. + +checkpoint62p (2010/12/17) +o pkg/thsice: rename (ad tape index) iicekey as ticekey, and move it + to THSICE_SIZE.h to avoid conflict with pkg/seaice. +o pkg/seaice: add seaice_diffusion.f seaice_map_thsice.f to adjoint. + allow for use of multim. advection in seaice adjoint. +o pkg/seaice: add diagnostics for snow precipitation and sublimation flux + - first attempt to close budgets by taking into account the freshwater + flux implied by the sublimation of snow and ice. So far only the + volume budget is closed, turn on with soon-to-be-retired-again CPP-Flag + SEAICE_ADD_SUBLIMATION_TO_FWBUDGET +o eesupp/src: + new S/R with new option (NML_EXTENDED_F77) to change array specification + from F95 standard to commonly accepted extended F77 form. + This would allow to use F95 syntax in namelist (needed with gfortran 4.4.5) + and to revert back to the original syntax (with -DNML_EXTENDED_F77) + for less tolerant f77 compiler (e.g., pgf77, g77). +o globalFiles: + commented out in parameter file "data" since it's not safe with MPI + (e.g., wrong on aces); keep it if useSingleCpuIO=T (but not doing much + in this case); no longer commented out in restart test (tst_2+2). +o add adTapeDir to specify a special directory where adjoint tapes are + written/read (adread_adwrite.F). adTapeDir is specified in data/PARM05 + and cannot be used together with mdsioLocalDir. +o nesting pkgs: + - packages_init_variables : put call to OBCS_INIT_VARIABLES in last position + - update Gianmaria's Nesting pkgs (from Contrib/nesting_sannino) + - add parameter files + utils dir + output from short run on beagle. +o pkg/seaice: + fix seaice_growth.F store directives; update results of exp. lab_sea.ad +o pkg/seaice: explicitly store tsurfloc in seaice_solve4temp.F to avoid + unnecessary copying in adjoint code, requires new parameter NMAX_ITER + +checkpoint62o (2010/11/25) +o pkg/exf: + - change CPP option EXF_VERBOSE to run-time logical param exf_verbose. +o pkg/seaice: + - seaice_ocean_stress: split a loop for better vectorization of the + adjoint code + - seaice_growth.F: fix bug in conservation of freshwater connected to + capping of heff, put contribution into d_HEFFbyNEG + - replace irritating parameters SEAICE_latentWater/Ice and SEAICE_sensHeat + by something more sensible (parameters that are what their name implies) + - change some defaults, so that by default exf-parameters are used for + things like rhoAir, cpAir,latent/sensible heat parameters, in order + to reproduce the old results, many data.seaice now contain funny values + for cpAir and latent heat of sublimation +o packages_readparms.F: move exf_readparms up in line so that + exf-parameters are visible for other packages (e.g. seaice) +o debugMode & debugLevel=2: + - differentiate between debugMode & debugLevel=2: + debugMode: controls printing of debug messages (sequence of S/R calls). + debugLevel: controls printing of algorithm intermediate results and stats. + - remove "if debugLevel=2 => debugMode" but keep "debugMode => debugLevel=2"; + => allows to print debug-stats without all sequence of call messages. + - switch control of debug print from debugLevel to debugMode in many S/R. +o pkg/obcs (obcs-sponge): + fix bug that was introduced on Oct 25 (in dynamics.F v1.151): put back the + call to OBCS_PRESCRIBE_EXCHANGES (need to exch gU,gV when using obcs-sponge) +o pkg/generic_advdiff: + - record overlap minimum size for each advection scheme; + - refine criteria on overlap minimum size; + - check and stop (in gad_check) if overlap size is too small. +o pkg/obcs: + - with ptracers: remove multiple calls to obcs_apply_ptracer (only done + once, in ptracers_implicit.F) + - with seaice: remove test on empty file name in obcs_apply_seaice (always + apply OB values when tile has OB) +o model/src, coriolis setting: + - integer parameter "selectCoriMap" (=0,1,2) replaces 3 logical + useConstantF, useBetaPlaneF, useSphereF ; can now read coriolis from + files (selectCoriMap=3) with any grid. + - add parameter "fPrime" for constant second Coriolis coeff (default=0). +o pkg/seaice/seaice_growth.F: + a few cosmetic changes that reduce the number of lines of code +o pkg/obcs: add boundary conditions following Stevens (1990) + - the code follows old MOM/NAOSIM code + - so far it is only implemented for the eastern and western boundaries + and NOT for pTracers +o pkg/rbcs: + - replace rbcsInIter by rbcsForcingOffset, the time in seconds at the + beginning of the first rbcs forcing period. + - add option rbcsSingleTimeFiles that makes rbcs read forcing data like the + offline package, i.e., one file per time, labeled by iteration number; + deltaTrbcs sets the timestep with which the data was generated; + rbcsIter0 is the forcing iteration corresponding to rbcsForcingOffset. + - rbcsForcingCycle=0 makes forcing non-cyclic. + - make arithmetic safe for negative times and odd forcing periods. + +checkpoint62n (2010/11/09) +o pkg/seaice: + fix masking in free-drift S/R ; update results of exp. lab_sea.fd +o pkg/obcs: + New S/R to copy tracer value from the OB location to narrow band region + (width = same as overlap) beyond the OB: preserve OB value (needed for + high-order scheme) which were overwritten by EXCH call if in overlap + region. Applied to theta, salt, ptracer and advected seaice variables. + With seaice: use maskInC in similar way as in gad_advection.F, + in seaice_advection.F and advect.F. + Update output of experiment seaice_obcs and change exp4 to test OBCS with + 4th order Adv.Scheme. for pasive salt and ptracer. +o tools/genmake2: + - when looking for an optfile (not specified), only search for Fortran compiler + (remove sym-link test + setting of CPP + makedepend check + C-compiler check) + - in get_fortran_c_namemangling: use "$CC $CFLAG" to compile C test programm + (instead of "$MAKE genmake_test.o" since an existing Makefile can be wrong) + - set default for CPP to "cpp -traditional -P" (was the case without specifying + an optfile, but otherwise default was just "cpp". + - comment out PLATFORM=$p_PLATFORM (no idea where p_PLATFORM comes from) + - when "CC" is not set in the optfile, look for possible C compiler + (instead of just setting CC=cc) + - put back version v.169 modif (from 2008/02/15), was dropped in v.173 changes: + => add a ".F" to ".o" rule for source files with no "#include" (such as + port_rand.F); otherwise, default ignores FOPTIM and just uses FFLAGS. + - move gfortran before g77 in list of possible Fortran compiler (consistent + with reference compiler switch from g77 to gfortran) +o pkg/seaice: + o remove SEAICE_OLD_AND_BAD_DISCRETIZATION-code, add a test to + seaice_check that stops the model, when the CPP flag is used. + o seaice_growth.F: make capping of ice volume conserving by returning the + capped ice as freshwater to EmPmR. This does not change the verification + experiments but will change runs where the capping is acutally active +o change reference platform+compiler from faulks+g77 to baudelaire+gfortran + (except the 2 fizhi test exp.); update output of exp.: bottom_ctrl_5x5, + global_ocean.cs32x15 (icedyn, thsice, AD thsice), global_ocean_ebm (AD), + lab_sea (fd, hb87, salt_plume), MLAdjust (3/5) and vermix (dd) +o tools/genmake2: + - merge genmake_warnings & genmake_errors into 1 log file: "genmake.log") + - change DUMPSTATE default to false (can use "-ds" option to switch it to T) + - clean previous genmake log files before anything else. + - change Makefile to remove testreport log files (*.tr_log) and genmake log + files only with "CLEAN" target (instead of "Clean"); +o pkg/generic_advdiff: + Change OBCS in multiDim advection (gad_advection.F): + Use maskInC to prevent intermediate update of tracer-field in OB regions, + thus keeping current-time-step OB values; and remove the call to OBCS_APPLY + which were applying future time-step OB values. + This way, all explicit calculations use current-time-step OB values + (simpler and more consistent). + Affects results when set-up uses time-varying OBC and multi-dim advection; + update output of experiment seaice_obcs. +o pkg/seaice: + move OPTIONS (formerly locally defined) to SEAICE_OPTIONS.h: + this includes SEAICE_GROWTH_LEGACY (+ SEAICE_DO_OPEN_WATER_GROWTH & + SEAICE_OCN_MELT_ACT_ON_AREA ) and SEAICE_SOLVE4TEMP_LEGACY. + Default (pkg/seaice/SEAICE_OPTIONS.h) is to define SEAICE_SOLVE4TEMP_LEGACY + but leave SEAICE_GROWTH_LEGACY undef ; For now, keep verification exp + unchanged (SEAICE_GROWTH_LEGACY defined in */code*/SEAICE_OPTIONS.h) +o pkg aim (2010/10/26): + - Introduce ALLOW_INSOLATION, moving toward a more general incoming solar radiation + formulation with obliquity, eccentricity, ... as run-time parameters, + - Introduce diagnostics for top-of-the-atmaophere incoming solar radiation and + upward shortwave radition at the surface (all-sky and clear-sky). +o pkg/obcs: + - writing sections with globalFiles=T does not work: switch to tiled-files + when writing sections for OBCS (change pkg/rw/write_rec.F accordingly). + - fix index bug (OB_Ie -> OB_Iw_) in obcs_apply_uvice.F + - apply OBCS_SEAICE_COMPUTE_UVICE to both uice and vice at all obcs + +checkpoint62m (2010/10/21) +o pkg/seaice: vectorize seaice_solve4temp.F + - move iterations loops outside of i/j-loops + - requires 2D arrays for TB and effConduct + - add 2D iceOrNot to replace HICE_ACTUAL.GT.0, as HICE_ACTUAL is + modified in this routine + - a little bit of clean up, adjusting indentation, etc. + - replaced a few TEN**a by exp(a*log(10.)) in new code for better + performance and '**TWO' by '**2' (because TWO is a float), + replaced ONE, ZERO +o eesupp/src: + fix flow dir for exch1_z_RX_cube, exch1_uv_RX_cube & exch1_bg_RX_cube +o pkg/seaice/seaice_growth.F + Revision 1.89 of seaice_growth.F completes the overall october 2010 revision and + merging process. The merged routine has two branches, and one activates the 'evolution' + branch by setting SEAICE_GROWTH_LEGACY to undef (at the top of the routine for now). + In summary, here is what the '10/10 revision' did consist of + part 1: seaice_growth.F revisions 1.71 to 1.75 + - Preliminary seaice_growth.F clean-up, keeping test results strictly unchanged. + The emphasis here was on improving readability (e.g. by trying to use meaningful + variable names) and modularity (e.g. by adding arrays to store heff etc. updates). + part 1.5: + - revising the lab_sea experiment to make it a more useful (i.e. stable + and representative of an actual model run). This was done by generating a + (winter) pickup to start from, switching back the main experiment dynamic + solver from evp to lsr, having one of the experiments use the free drift solution. + part 2: seaice_growth.F revisions 1.76 to 1.85 + - Sorting out hard-coded constants that prevented heat and fresh water conservation. + - Sorting out the various fields units and sign (set early on and reset at the end) + to avoid numerous scalings/re-scalings. By v1.85 the basic unit for ice, heat and + fresh water stock is 'effective ice meters', and >0 implies more ice/snow. + - Using tendency terms arrays (d_HEFF* etc.) to alleviate remaining code locks, + and simplifying logical expressions to further improve readibility. + part 3: seaice_growth.F revisions 1.86 to 1.89 + - Re-organize the blocks of code more logically. + - Merging the '_if.F' pieces that add something new. +o pkg/thsice: + - vectorize thsice_solve4temp and thsice_get_exf. + Problem: thsice_solve4temp has an unvectorizable statement that + might also be problematic for TAF + - thsice_get_exf: add the ALLOW_BULK_LARGEYEAGER04 CPP-flag. This + does not change the verifcation experiment, but will change all + experiments that have this flag undefined (not tested in verification); + a little reorganization to save a few 2D fields and recomputations + - 1st step of revisiting thsice adjoint (major recomp. sorted out) + - 2nd step: fix some store dirs + o seems to work somewhat, i.e. using ifort but not g77 + o likely cause are incomplete init. or recomputation + o start verification/global_ocean.cs32x15/input_ad.thsice +o pkg/obcs: add support for exch2 + - the position of the boundary and prescribed values are specified using + a global domain with exch2 facets stacked + - in x for N,S boundaries (like W2_mapIO=-1) + - in y for E,W boundaries (so E,W boundaries do not overlap) +o pkg/mdsio: add support for writing global sections with exch2 + file layout is as for obcs described above: + - for x-z sections, facets are stacked in x (like W2_mapIO=-1) + - for y-z sections, facets are stacked in y +o pkg/thsice: + - change a few routines for better vectorization: thsice_albedo.F, + thsice_step_temp.F, and in thsice_calc_thickn.F. Results are + not affected, but in particular the latter routine has changed + dramatically (not necessarily easier to read now). + - re-arrange calculation in THSICE_SOLVE4TEMP: move outside of i,j loops + iteration loop and call to S/R THSICE_GET_BULKF & THSICE_GET_EXF. +o verification/lab_sea: + - overhaul of the forward model experiments. The main motivation + for this revision is that lab_sea has proven quite fragile (when + changing the sea-ice code, and switching platforms) to a point + when failures were very hard to attribute to specific code problems. + - the suspected reason for this fragility is noise amplification by evp so, + to start with, we decided to replace evp with lsr in the main experiment. + - the input.lsr experiment thus became obsolete and was removed. + - it was replaced with the input.fd experiment that uses an + analytical "free drift" solution as the ice velocity field. + - Taken together, input (lsr), input.fd (free drift) and input.hb87 (evp) + will hopefully allow us to discriminate problems due to dynamic solvers + from problems due to the thermodynamic code. input.salt_plume (lsr) allows + us to test additional options, along with the salt plume package. + - Finally a pickup was generated (at the end of a one year run) to + make the tests more representative of an actual model run. Previously + the tests starting point was a 1m thickness of ice everywhere. +o pkg/seaice: + - introducing S\R seaice_freedrift.F that compute the + analytical solution to an approximate ice momentum + equation (i.e. the simple stress/coriolis balance). + - define SEAICE_ALLOW_FREEDRIFT to trigger the computation + of the free drift velocity field, and enable its use. + To actually use the free drift solution as the model ice + velocity field set SEAICEuseFREEDRIFT to .TRUE. in data.seaice. +o move "useOASIS" from PARAMS.h to EEPARAMS.h and read it from "eedata" + +checkpoint62l (2010/10/04) +o pkg/seaice: + -fix the USE_ORIGINAL_SBI version (for variable freezing point); + -use "celsius2K" in SEAICE_VARIABLE_FREEZING_POINT and in new + version (with USE_ORIGINAL_SBI undef); +o add hooks for new packages OASIS, the package will follow + (hooks exclude the seaice pkg for now) o pkg/autodiff: ALLOW_AUTODIFF_WHTAPEIO - Option to use mdsio_read_whalos.F/mdsio_write_whalos.F, + Option to use mdsio_read_whalos.F/mdsio_write_whalos.F, rather than mdsio_readvector.F/mdsio_writevector.F, within adread_adwrite.F to read/write adjoint-related tapes to disk. -o eesup and pkg/mdsio: ALLOW_WHIO - - Option to write/read 2D fields to files including tiles halos. - The main purpose is for adjoint related "tape I/O". +o eesupp and pkg/mdsio: ALLOW_WHIO + Option to write/read 2D fields to files including tiles halos. + The main purpose is for adjoint related "tape I/O". The secondary purpose is for debugging phases. - A word on those rather specifically intended purposes. The code was meant to be stripped down to the bare minimum. - Typically, there is a unique and basic mapping to the global + Typically, there is a unique and basic mapping to the global buffer (tile by tile, then proc by proc, then level by level), and byteswaps and meta-files are omitted. Indeed the (*_WH*) code is not - meant to generate user friendly outputs of various shapes and forms. + meant to generate user friendly outputs of various shapes and forms. Rather it is intended, for the knowledgeabale developers, to be largely independent of the many mdsio options, to allow - outputs of tiles+halos in debugging phases and in adjoint runs, + outputs of tiles+halos in debugging phases and in adjoint runs, and to allow some flexibility in tuning (adoint) I/O depending on each - file system behavior. With regard to tuning, most of the potential knobs + file system behavior. With regard to tuning, most of the potential knobs (to leave files open, to allocate the I/O indep. of usesinglecpuio, to specify directories, to specify the typical size of I/O calls, etc.) are not yet included, and will be added if they prove useful. @@ -317,8 +666,8 @@ but keep the former optfile for few weeks for smooth transition o pkg/ecco - allow user defined cost function contributions - CPP: ALLOW_USERCOST_CONTRIBUTION - USAGE: compute contributions in cost_usercost_all.F + CPP: ALLOW_USERCOST_CONTRIBUTION + USAGE: compute contributions in cost_usercost_all.F - note: this is identical to what ALLOW_GENCOST_CONTRIBUTION once was. the two options are complementary. o pkg/smooth