Changeset 493 for palm/trunk/SOURCE/user_statistics.f90

Timestamp:

Mar 1, 2010 8:30:24 AM (14 years ago)

Author:

raasch

Message:

New:
---
Output in NetCDF4-format. New d3par-parameter netcdf_data_format.

(check_open, check_parameters, close_file, data_output_2d, data_output_3d, header, modules, netcdf, parin)

Modules to be loaded for compilation (mbuild) or job execution (mrun)
can be given in the configuration file using variable modules. Example:

%modules ifort/11.0.069:netcdf lcsgih parallel

This method replaces the (undocumented) mpilib-variable.

WARNING: All fixed settings of modules in the scripts mbuild, mrun, and subjob
have been removed! Please set the modules variable appropriately in your
configuration file. (mbuild, mrun, subjob)

Changed:

---

Parameters netcdf_64bit and netcdf_64bit_3d have been removed. Use
netcdf_data_format = 2 for choosing the classic 64bit-offset format (this is
the default). The offset-format can not be set independently for the
3d-output-data any more.

Parameters netcdf_format_mask, netcdf_format_mask_av, and variables
nc_format_mask, format_parallel_io removed. They are replaced by the new
parameter netcdf_data_format. (check_open, close_file,
data_output_mask, header, init_masks, modules, parin)

Errors:

---

bugfix in trunk/UTIL/Makefile: forgot to compile for interpret_config

Bugfix: timeseries data have to be collected by PE0 (user_statistics)

File:

• palm/trunk/SOURCE/user_statistics.f90 (modified) (4 diffs)

palm/trunk/SOURCE/user_statistics.f90

@@ -4 +4 @@
 ! Current revisions:
 ! -----------------
-! 
+! Bugfix: timeseries data have to be collected by PE0
 !
 ! Former revisions:
@@ -25 +25 @@
     USE arrays_3d
     USE indices
+    USE netcdf_control
     USE statistics
     USE user
@@ -33 +34 @@
 
     INTEGER ::  i, j, k, sr, tn
+
+    REAL, DIMENSION(dots_num_palm+1:dots_max) ::  ts_value_l
 
 
@@ -70 +73 @@
 !--    creates two time series for the absolut values of the horizontal
 !--    velocities u and v.
-!       ts_value(dots_num_palm+1,sr) = ABS( u_max )
-!       ts_value(dots_num_palm+2,sr) = ABS( v_max )
+!       ts_value_l = 0.0
+!       ts_value_l(dots_num_palm+1,sr) = ABS( u_max )
+!       ts_value_l(dots_num_palm+2,sr) = ABS( v_max )
+!
+!--     Collect / send values to PE0, because only PE0 outputs the time series.
+!--     CAUTION: Collection is done by taking the sum over all processors.
+!--              You may have to normalize this sum, depending on the quantity
+!--              that you like to calculate. For serial runs, nothing has to be
+!--              done.
+!--     HINT: If the time series value that you are calculating has the same
+!--           value on all PEs, you can omit the MPI_ALLREDUCE call and
+!--           assign ts_value(dots_num_palm+1:,sr) = ts_value_l directly.
+!#if defined( __parallel )
+!       CALL MPI_ALLREDUCE( ts_value_l(dots_num_palm+1),                       &
+!                           ts_value(dots_num_palm+1,sr),                      &
+!                           dots_max-dots_num_palm, MPI_REAL, MPI_SUM, comm2d, &
+!                           ierr )
+!#else
+!       ts_value(dots_num_palm+1:,sr) = ts_value_l
+!#endif
 
     ENDIF