| 4 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
| 5 |
============================== |
============================== |
| 6 |
|
|
| 7 |
|
o pkg/mdsio: |
| 8 |
|
- read/write tiled (local) files: read/write 1-level tile chunk at a time |
| 9 |
|
(instead of segment of length sNx); expected to speed up tiled IO. |
| 10 |
|
o pkg/exch2: |
| 11 |
|
- take buffer copy from/to array out of S/R exch2_send/recv into new |
| 12 |
|
S/R exch2_put/get ; Exch of local variable now works with MPI+MTH |
| 13 |
|
(tested by removing commom block statement in SOLVE_FOR_PRESSURE.h). |
| 14 |
|
- ad version of send/recv no longer needed (but needs exch2_ad_put/ad_get) |
| 15 |
|
- simplify argument list of S/R exch2_get_uv_bounds and exch2_get_scal_bounds |
| 16 |
|
(which replaces exch2_get_recv_bounds & exch2_get_send_bounds) by |
| 17 |
|
including Topology header file (and common blocks). |
| 18 |
|
- implement EXCH_IGNORE_CORNERS in scalar exchange (rx1); |
| 19 |
|
remove 2nd exch call in exch2_s3d_rx (no longer needed). |
| 20 |
|
- document S/R arg. list (+ update some comments) |
| 21 |
|
o pkg/seaice: |
| 22 |
|
- clean up computation of Hibler+Bryan (1987) stress coupling for the case |
| 23 |
|
of LSR (this change is expected to change the results slightly because now |
| 24 |
|
it uses slightly difference moduli from the second last LSR solution, but |
| 25 |
|
that is more consistent with the stress computations; this part of the code |
| 26 |
|
is not tested in the verification experiments) |
| 27 |
|
- fix diagnostics for flux of ice in case of B-grid |
| 28 |
|
- turn strain rates eij and press into global fields within global |
| 29 |
|
common block in SEAICE.h for more straighforward diagnostics |
| 30 |
|
- fix diagnostics for principal stress components, press, zeta, eta |
| 31 |
|
to give something meaning full |
| 32 |
|
o change tiling of global_ocean.cs32x15 (12 tiles, 32x16) and generate |
| 33 |
|
new output (all 4 tests) using default rotationPeriod. |
| 34 |
|
o pkg KPP: |
| 35 |
|
- add flag (LimitHblStable) to avoid limiting BL depth under stable conditions |
| 36 |
|
- fix vertical indices in RI_IWMIX (match between KPP grid from 0 to Nr+1 |
| 37 |
|
and normal grid from 1 to Nr) |
| 38 |
|
o Fix computations of L4rdt in mom_calc_visc.F |
| 39 |
|
o eesupp: |
| 40 |
|
new S/R ALL_PROC_DIE (like EEDIE, without irrelevant threads err msg): |
| 41 |
|
should be called before a "stop" if we know that all process are going |
| 42 |
|
to stop; if one Proc does call this S/R, will hang in MPI_FINALISE. |
| 43 |
|
|
| 44 |
|
checkpoint61o |
| 45 |
|
o pkg/exch2: change check_depths to issue just a warning (instead of error+stop) |
| 46 |
|
o model/src/ini_curvilinear_grid.F: |
| 47 |
|
compute angleSin/CosC when they are not read from a file |
| 48 |
|
o pkg/seaice: |
| 49 |
|
add pseudo time stepping for LSOR |
| 50 |
|
o pkg/exch2: |
| 51 |
|
temporary fix for exch2_s3d_rx; update results of exp. hs94.cs-32x32x5.impIGW |
| 52 |
|
o pkg/seaice: |
| 53 |
|
Adding separate sea ice albedo, drag coefficients, and HO in Southern |
| 54 |
|
Ocean from those in Arctic Ocean. See *_south in SEAICE_PARAMS.h |
| 55 |
|
o SingleCpuIO: |
| 56 |
|
fix bug put before chkp61n (May 11), when more than 2 MPI cpu are used. |
| 57 |
|
o pkg/seaice: |
| 58 |
|
put exchanged variable in (local) common block for MPI+Multi-Threads |
| 59 |
|
to work (necessary with exch1, and for now with exch2 too). |
| 60 |
|
o verification: to get more experiments tested with mpi, mth and mpi+mth, |
| 61 |
|
change tiling of: |
| 62 |
|
aim.5l_Equatorial_Channel, aim.5l_LatLon, dome, exp4, global_ocean.90x40x15, |
| 63 |
|
hs94.cs-32x32x5, MLAdjust, offline_exf_seaice & tutorial_plume_on_slope |
| 64 |
|
update output of exp: |
| 65 |
|
exp4, global_ocean.90x40x15(+.dwslp), MLAdjust(all 5), |
| 66 |
|
tutorial_plume_on_slope and hs94.cs-32x32x5 (use default rotationPeriod) |
| 67 |
|
+ ideal_2D_ocean: use default rotation period + test Global_Sum_SINGLECPU |
| 68 |
|
(note: output.txt generated on faulks using only 1 tile) |
| 69 |
|
o eesupp & mdsio (SingleCpuIO): |
| 70 |
|
- remove MISSING_TILE_IO (from MDSIO_OPTIONS.h), replaced by run-time |
| 71 |
|
parameter "W2_useE2ioLayOut" (read from file "data.exch2"). |
| 72 |
|
Note: was not effective for globalFile & useSingleCpuIO=F; now fixed. |
| 73 |
|
- move MDSIO_SCPU.h (pkg/mdsio) to EEBUFF_SCPU.h (eesupp/inc). |
| 74 |
|
- fix GLOBAL_SUM_SINGLECPU when using Exch2; re-use same buffers and same |
| 75 |
|
gather/scatter S/R as with SingleCpuIO (=> 1 less 2D global RL array). |
| 76 |
|
o pkg/exch2: |
| 77 |
|
- new code to set-up W2-Exch2 topology (replace matlab-topology-generator): |
| 78 |
|
read parameter file "data.exch2" if it exists; otherwise try default regular |
| 79 |
|
cube without blank-tile; new header file "W2_EXCH2_SIZE.h"; |
| 80 |
|
- update matlab-topology-generator (exch2_setup_cs6_print.m) to be compatible |
| 81 |
|
with new code (not to be used, just as a backup solution). |
| 82 |
|
- add examples (e.g., for lat-lon-cap_120) of parameter file "data.exch2" |
| 83 |
|
in utils/exch2/input. |
| 84 |
|
|
| 85 |
|
checkpoint61n |
| 86 |
|
o eesupp & mdsio (SingleCpuIO): |
| 87 |
|
- move mapping to global io-buffer inside gather_2d/scater_2d: this save |
| 88 |
|
memory (1 less 2D global RL array); skip 1 buffer copy; and only send/ |
| 89 |
|
receive real*4 array (instead of real*8) to read/write 32.bit files. |
| 90 |
|
(new template for gather_2d/scater_2d, new set of S/R MDS_PASS_R4,8toRL,S |
| 91 |
|
and remove new mapping S/R) |
| 92 |
|
o pkg/mdsio: in preparation for topology-generator within pkg/exch2: |
| 93 |
|
- Change declaration of SingleCpuIO buffer, reduced to only 1 dim.: |
| 94 |
|
xyBuffer_size, which is set to W2_ioBufferSize from W2_EXCH2_PARAMS.h |
| 95 |
|
when using exch2. Add S/R to map global io-buffer to global model array. |
| 96 |
|
- uses exch2_global_Nx,y instead of exch2_domain_nxt,nyt. |
| 97 |
|
|
| 98 |
|
checkpoint61m |
| 99 |
|
o write_utils : change WRITE_1D_R8 to WRITE_1D_RL |
| 100 |
|
and add 3 other S/R for other types (R4,R8,RS). |
| 101 |
|
Change calls to WRITE_0D/1D_R[4,8,S,L] to match the type of argument. |
| 102 |
|
o eesupp/inc/CPP_EEMACROS.h (EXCH & GLOBAL_SUM/MAX macros): |
| 103 |
|
- ensure suffix (_R4,_R8,_RS,_RL) correspondance between macro and S/R ; |
| 104 |
|
- reduce number of macro definitions (keep only those which are used); |
| 105 |
|
can always call directly the corrresponding S/R if needed. |
| 106 |
|
- everywhere in the code: change macros (EXCH & GLOBAL_SUM/MAX) suffix |
| 107 |
|
_R4/_R8 to _RS/_RL when applied to _RS/_RL variable. |
| 108 |
|
Fixed wrong type of arg in: |
| 109 |
|
atm2d/put_ocnvars.F cheapaml/cheapaml.F ocn_compon_interf/ocn_apply_import.F |
| 110 |
|
ctrl/ctrl_map_ini_ecco.F seaice/seaice_do_diags.F seaice/seaice_model.F |
| 111 |
|
To fix: global_sum of var in common block is wrong (multi-threaded) in: |
| 112 |
|
dic/dic_atmos.F ebm/ebm_area_t.F ebm/ebm_zonalmean.F sbo/sbo_calc.F |
| 113 |
|
To fix: aim_v23/aim_do_co2.F (BARRIER & GLOBAL_SUM within bi,bj loops) |
| 114 |
|
o adjustment.cs-32x32x1: test "blank tiles" when compiling with mpi; 2nd |
| 115 |
|
"nlfs" mpi test is going to stop when checking for unconnected tile egdes. |
| 116 |
|
o eesupp/src: new S/R to stop everyone when 1 (or more) Proc. find an error |
| 117 |
|
o pkg/exch2: new S/R EXCH2_CHECK_DEPTHS to check that unconnected tile |
| 118 |
|
edges are closed (zero depth); might happen with "blank tiles"; |
| 119 |
|
for now, disabled if using OBCS. |
| 120 |
|
o tools/gemake2: delete modifs from March 27-29 2009 (with option "-mpi": |
| 121 |
|
link *_mpi files from "MODS" dir) to return to a strait-forward behavior. |
| 122 |
|
o move read_write.F to write_utils.F ; new parameter (maxLengthPrt1D) |
| 123 |
|
to reduce length of 1.D array print in STDOUT. |
| 124 |
|
o adjustment.cs-32x32x1: changed to oceanic test with 1 continent |
| 125 |
|
+ use 48 tiles (16x8): this will allow to test "blank tiles". |
| 126 |
|
o start to test MPI + multi-threaded (change testreport to skip tests |
| 127 |
|
where number of tiles do not match). |
| 128 |
|
o genmake2: add option "-omp" to turn on Open-MP ; |
| 129 |
|
passed to testreport (if using -mth) ; merge few (*+mth) optfiles. |
| 130 |
|
o pkg/exch2 (+ utils/exch2): ignore blank tiles (instead of removing them) |
| 131 |
|
and keep the original tile numbering. Note: compatible with old |
| 132 |
|
W2_EXCH2_TOPOLOGY.h & w2_e2setup.F topology files. |
| 133 |
|
|
| 134 |
|
checkpoint61l |
| 135 |
|
o Overlaps had been forgotten in calculating ijk keys : fixed |
| 136 |
|
o fix cyrus makedepend for gcc 4.4 |
| 137 |
|
o experiment fizhi-cs-aqualev20: |
| 138 |
|
switching to new version of S/R GETPWHERE (#define TRY_NEW_GETPWHERE) |
| 139 |
|
and generate new output. Note: new output is much closer to the old |
| 140 |
|
one (v1.8) generated with pgf77 -O2 on Mar 14 2007 than it is from |
| 141 |
|
the previous one (v1.9) generated with pgf77 -O0 on Jun 13 2007. |
| 142 |
|
o pkg/fizhi: |
| 143 |
|
- change to avoid indices going deliberately over array-bounds |
| 144 |
|
- remove unnecessary MPI stuff |
| 145 |
|
- perpetual spring equinox with "#define FIZHI_USE_FIXED_DAY" |
| 146 |
|
- new version of S/R GETPWHERE available with #define TRY_NEW_GETPWHERE: |
| 147 |
|
does not go over array-bounds and therefore returns reliable output. |
| 148 |
|
- in fizhi_driver.F & fizhi_turb.F, fix "get_alarm" calls: |
| 149 |
|
this S/R has 4 output arguments, which are used inside "get_alarm"; |
| 150 |
|
when it is called with the same variable "ndum" in place of several |
| 151 |
|
output arguments, this produces weird side effect inside "get_alarm" |
| 152 |
|
(e.g., datein=dates(n) is overwritten by timein=times(n) since both |
| 153 |
|
datein & timein, resp. 2nd and 3rd arg., are pointed to the same "ndum" |
| 154 |
|
address) and produces error. |
| 155 |
|
o tools/gemake2: (March 27-29 2009) |
| 156 |
|
with option "-mpi": link *_mpi files from "MODS" dir (this is what |
| 157 |
|
testreport does). |
| 158 |
|
o lab_sea experiment: change from 2x1 tiles to 2x2 tiles and regenerate output |
| 159 |
|
o pkg/seaice: fix bug in new version of S/R SEAICE_LSR.F: |
| 160 |
|
update lab_sea.lsr & lab_sea.salt_plume results |
| 161 |
|
o pkg/seaice: |
| 162 |
|
Major overhaul of the LSOR solver: change discretization from original |
| 163 |
|
finite differences to finite volume. The old routine is still in the |
| 164 |
|
file seaice_lsr.F and can be turned on with the CPP-flag |
| 165 |
|
SEAICE_OLD_AND_BAD_DISCRETIZATION |
| 166 |
|
Along with this change, the discretization of the strain rates is changed. |
| 167 |
|
The same flag turns on the old discretization in seaice_calc_strainrates.F |
| 168 |
|
All verification experiments with seaice are affected (except for |
| 169 |
|
seaice_exf_offline); global_ocean.cs32x15/input.icedyn recovers the old |
| 170 |
|
results by setting the above flag. |
| 171 |
|
Further changes involve changes in defaults: |
| 172 |
|
- ALLOW_FLOODING is defined and turned on by default |
| 173 |
|
- SEAICEadvSnow = .true. is now the default |
| 174 |
|
- SEAICEadvSalt = .true. is now the default |
| 175 |
|
- SEAICEadvAge = .true. is now the default |
| 176 |
|
- SEAICE_clipVelocities = .false. is now the default |
| 177 |
|
(as per J. Zhang's recommendation) |
| 178 |
|
- B-grid, and thus not tested: SEAICE_TEST_ICE_STRESS_1/EXPLICIT_SSH_SLOPE |
| 179 |
|
is defined, SEAICE_TEST_ICE_STRESS_1 is renamed into SEAICE_BICE_STRESS |
| 180 |
|
- seaice_growth: replace computation of UG by a simple copy from wspeed |
| 181 |
|
|
| 182 |
|
checkpoint61k |
| 183 |
|
o pkg/seaice: a few bug fixes mostly concerning the metric terms: |
| 184 |
|
seaice_lsr: fixed a few metric terms, |
| 185 |
|
affects lab_sea.lsr, lab_sea.salt_plume, seaice_obcs |
| 186 |
|
add more debugging diagnostics |
| 187 |
|
seaice_evp: removed superfluous many metric terms of FV discretization |
| 188 |
|
of stress divergence |
| 189 |
|
affects lab_sea, lab_sea.hb87 |
| 190 |
|
seaice_ocean_stress: removed superfluous many metric terms of FV discretization |
| 191 |
|
of stress divergence for non-evp solver |
| 192 |
|
(this part is not tested anywhere) |
| 193 |
|
seaice_calc_strainrates: fixed a bug in no-slip boundary conditions, |
| 194 |
|
they were effectively only half slip |
| 195 |
|
affects lab_sea.hb87 |
| 196 |
|
updated all relevant verification experiments (except lab_sea//output_adm.txt |
| 197 |
|
which was left over, but fails at level 7 & 6) |
| 198 |
|
|
| 199 |
|
checkpoint61j |
| 200 |
|
o pkg/seaice: turn a few hard wired parameters into run time parameters |
| 201 |
|
o pkg/seaice/exf/monitor: turn netcdf coordinate variable T into model |
| 202 |
|
time in seconds as promised by the usints in seaice_monitor and exf_monitor |
| 203 |
o pkg/mnc: fix the coordinate variables for multi-tile curvi-linear grids |
o pkg/mnc: fix the coordinate variables for multi-tile curvi-linear grids |
| 204 |
that are not cubed-sphere grids |
that are not cubed-sphere grids |
| 205 |
o pkg/seaice: fix a bug for the rare case of SEAICE_VECTORIZE_LSR |
o pkg/seaice: fix a bug for the rare case of SEAICE_VECTORIZE_LSR |
Parent Directory
| 
Revision Log
| 
Revision Graph
| 
Patch