Annotation of /MITgcm/model/src/ini_forcing.F

Revision 1.16 - (hide annotations) (download)
Wed May 5 18:32:34 1999 UTC (25 years, 1 month ago) by adcroft
Branch: MAIN
CVS Tags: checkpoint21, checkpoint22, checkpoint23, checkpoint24, checkpoint25
Changes since 1.15: +27 -23 lines

Implemented new I/O package (mdsio.F). This package does parallel
I/O in much the same way as dfile.F used to except it uses "direct
access" rather than (f77) unformatted sequential access.

Problems with dfile.F package included:
  o unnecessary memory use (each process had two global sized buffers)
  o inability to read the files it had written without post-processing
  o "tiled" files were tiled by process/thread rather than actual tiles
  o created huge numbers of files with no alternatives

Features of the mdsio.F package:
  o direct-access binary writes
  o no excessive memory use
  o ability to read/write from multiple record files
  o "tiled" files are based on "WRAPPER" tiles so that the number
    and content of files is independent of the number of threads
    and/or processes
  o option to create single "global" files rather than "tiled" files
  o ability to read both "global" and "tiled" files
    [Caveat: the tiling of files must match the model tiles]
  o checkpoints now use a single file per model section
    ie.  one file for the hydrostatic model core, one file
    for the non-hydrostatic extensions and one file for the C-D
    extensions
  o the mid-level I/O routines now is broken into more source files
    read_write_fld.F supplies basic I/O routines with the same interface
                     as the original I/O package
    read_write_rec.F supplies I/O routines which allow multiple records
    write_state.F    writes the model state
    checkpoint.F     supplies the read/write checkpoint routines

All the example input data has had to be modified to be direct-access.
Otherwise only routines that used I/O have been affected and not
all of those have been due to the continuity of arguments in
the read_write_fld.F routines.

What needs to be done?  We have to create a suite of conversion
utilities for users with old-style data. Also supply the option
for using old-style I/O, not just for die-hards but for reading
data too extensive to be converted. And more...