high_res_cube/utils/README.txt


Averaging MITgcm output:

The program and scripts here are a few quick hacks that efficiently
average binary files as output by MITgcm.  The following assumptions
apply:

  - all files are SINGLE-precision IEEE-754 stored in BIG-endian byte
    order (which is the standard for MITgcm single-precision MDSIO
    output)

  - the machine running the "average_fields" program is a
    LITTLE-endian architecture such as:

    - IA32 (Intel or AMD x86-compatible)
    - Linux on IA64 (eg. Altixes)


The program syntax is:

  $  make average_fields
  $  ./average_fields field_1 field_3 ... field_n output_field


Two additional scripts ("do_the_averages" and "job_ave") demonstrate
how to easily run N simultaneous average_fields programs within either
interative (the former) or queue-driven (the latter) environments.


NOTE: 

Outputs from this program can differ slightly from machine to machine
due to rounding errors.  The intermediate sums are computed internally
at double precision and converted to single precision immediately
before being written.  Both IA32 and IA64 machines have been compared
and the differences (when there are any!) all appear to be acceptable
within the slightly different rounding inherent with these two
architectures.

1
2	Averaging MITgcm output:
3
4	The program and scripts here are a few quick hacks that efficiently
5	average binary files as output by MITgcm. The following assumptions
6	apply:
7
8	- all files are SINGLE-precision IEEE-754 stored in BIG-endian byte
9	order (which is the standard for MITgcm single-precision MDSIO
10	output)
11
12	- the machine running the "average_fields" program is a
13	LITTLE-endian architecture such as:
14
15	- IA32 (Intel or AMD x86-compatible)
16	- Linux on IA64 (eg. Altixes)
17
18
19	The program syntax is:
20
21	$ make average_fields
22	$ ./average_fields field_1 field_3 ... field_n output_field
23
24
25	Two additional scripts ("do_the_averages" and "job_ave") demonstrate
26	how to easily run N simultaneous average_fields programs within either
27	interative (the former) or queue-driven (the latter) environments.
28
29
30	NOTE:
31
32	Outputs from this program can differ slightly from machine to machine
33	due to rounding errors. The intermediate sums are computed internally
34	at double precision and converted to single precision immediately
35	before being written. Both IA32 and IA64 machines have been compared
36	and the differences (when there are any!) all appear to be acceptable
37	within the slightly different rounding inherent with these two
38	architectures.
39