[Dart-dev] DART/branches Revision: 10871

[dart at ucar.edu]

hendric at ucar.edu
2017-01-12 16:49:57 -0700 (Thu, 12 Jan 2017)
163
Updating rma documentation for release.  There is still some
additional information to be added about the stages to write
which will have to wait until tomorrow.




Modified: DART/branches/rma_fixed_filenames/rma_doc/bitwise_considerations.html
===================================================================
--- DART/branches/rma_fixed_filenames/rma_doc/bitwise_considerations.html	2017-01-10 20:53:40 UTC (rev 10870)
+++ DART/branches/rma_fixed_filenames/rma_doc/bitwise_considerations.html	2017-01-12 23:49:57 UTC (rev 10871)
@@ -25,15 +25,25 @@
 </tr>
 </table>
 
-By bitwise we mean bit for bit identical results in the output obs_sequence file and the restarts (netcdf-to-netcdf or dart format-to-dart format) when comparing one version of the code to another.
+By bitwise we mean bit for bit identical results in the output obs_sequence 
+file and the restarts (netcdf-to-netcdf or dart format-to-dart format) when 
+comparing one version of the code to another.
 
-For testing the code to be bitwise with Lanai there are several things to change/set in RMA_DART and your mkmf:
+For testing the code to be bitwise with Lanai there are several things to 
+change/set in RMA_DART and your mkmf:
 
 <ol>
 <li>assim_tools_mod.f90:<br>
-     <code>lanai_bitwise = .true.</code>  This is hard coded to false because it is very slow to be bitwise with Lanai.  <code>lanai_bitwise = .true.</code> causes the vertical conversion of observations to be done in get_close_obs inside the sequential obs do loop.  See <a href="vertical_conversion.html">vertical conversion</a> for details about this change.
+     <code>lanai_bitwise = .true.</code>  This is hard coded to false because 
+     it is very slow to be bitwise with Lanai.  
+     <code>lanai_bitwise = .true.</code> causes the vertical conversion of 
+     observations to be done in get_close_obs inside the sequential obs do 
+     loop.  See <a href="vertical_conversion.html">vertical conversion</a> for 
+     details about this change.
 <li>filter_nml:<br>
-    <code>output_forward_op_errors = .true.</code> This will cause the forward operator code to calculate all forward operators, even if some ensemble members fail. 
+    <code>output_forward_op_errors = .true.</code> This will cause the forward 
+    operator code to calculate all forward operators, even if some ensemble 
+    members fail. 
 <li>mkmf:<br>
     ifort fp-model precise<br>
     In general use the debugging FLAGS in the mkmfs provided with DART.
@@ -40,8 +50,11 @@
 </ol>
 
 <h3>Important</h3>
-The <em>CAM</em> and <em>bgrid_solo</em> model_mods have been altered so the state is in a different order inside filter. Thus dart format restarts will <b>not</b> be bitwise with Lanai dart format restarts, but netcdf files will be (after running dart_to_cam).
+The <em>CAM</em> and <em>bgrid_solo</em> model_mods have been altered so the 
+state is in a different order inside filter. Thus dart format restarts will 
+<b>not</b> be bitwise with Lanai dart format restarts, but netcdf files will 
+be (after running dart_to_cam).
 
 
 </BODY>
-</HTML>
\ No newline at end of file
+</HTML>

Modified: DART/branches/rma_fixed_filenames/rma_doc/distributed_state.html
===================================================================
--- DART/branches/rma_fixed_filenames/rma_doc/distributed_state.html	2017-01-10 20:53:40 UTC (rev 10870)
+++ DART/branches/rma_fixed_filenames/rma_doc/distributed_state.html	2017-01-12 23:49:57 UTC (rev 10871)
@@ -25,23 +25,26 @@
 </tr>
 </table>
 
-The key part of RMA DART is having a state that is physically distributed across processors.
-The location in memory of any part of the state vector (which processor and where in memory on that processor)
-is completely under the control of filter, not model_mod.
+The key part of RMA DART is having a state that is physically distributed 
+across processors.  The location in memory of any part of the state vector 
+(which processor and where in memory on that processor) is completely under 
+the control of filter, not model_mod.
 <p>
 Implications of this:
 <ul>
-<li> The model_mod never gets a whole state vector to use. So no whole vector for a forward operator, and
-no whole vector for the mean.
-<li> The model_mod can not make any assumptions about the order of elements in the state.
-<br> Currently, filter is ordering variables in the order they are listed in add_domain and with the
-dimenion order of the netcdf file. This is what is happening in most model_mod converters (model_to_dart,
-dart_to_model).  However CAM and bgrid_solo rearrange the state in Lanai. These model_mods (and converters)
-have been changed to not rearrage the state.
+<li> The model_mod never gets a whole state vector to use. So no whole vector 
+     for a forward operator, and no whole vector for the mean.
+<li> The model_mod can not make any assumptions about the order of elements in 
+     the state.
+<br> Currently, filter is ordering variables in the order they are listed in 
+     add_domain and with the dimenion order of the netcdf file. This is what 
+     is happening in most model_mod converters (model_to_dart, dart_to_model).  
+     However CAM and bgrid_solo rearrange the state in Lanai. These model_mods 
+     (and converters) have been changed to not rearrage the state.
 </ul>
 
-So, how does the model_mod access the state without having the vector and not knowing the state order?
-- state accessor routines.
+So, how does the model_mod access the state without having the vector and not 
+knowing the state order?  - state accessor routines.
 
 <h4> State Accessor Routines </h4>
 
@@ -50,26 +53,31 @@
 <code>function get_dart_vector_index(i, j, k, dom_id, var_id)</code>
 <p>
 get_dart_vector_index returns the dart index for a given i,j,k of a variable.
-Note if the variable is 1D j and k are ignored. If a variable is 2D k is ignored. Note only
-variables upto 3D are supported, but this could be extended to support upto 7 dimensional variables
-(or whatever fortran and netcdf will support).
+Note if the variable is 1D j and k are ignored. If a variable is 2D k is