Changes between Version 259 and Version 260 of doc/tec/changelog_2018

Timestamp:

Sep 10, 2013 9:51:44 AM (11 years ago)

Author:

raasch

Comment:

--

doc/tec/changelog_2018

@@ -11 +11 @@
 
 ||='''Date'''  =||='''Author'''  =||='''svn\\Revision'''  =||='''Last\\Release'''  =||='''Type''' =||='''Description''' =||
+|----------------
+{{{#!td style="vertical-align:top;width: 50px"
+10/09/13
+}}}
+{{{#!td style="vertical-align:top;width: 50px"
+SR
+}}}
+{{{#!td style="vertical-align:top;width: 75px"
+r1221
+}}}
+{{{#!td style="vertical-align:top"
+3.9
+}}}
+{{{#!td style="vertical-align:top"
+N, C, B
+}}}
+{{{#!td style="vertical-align:top"
+'''New:'''\\
+openACC porting of reduction operations. An accelerator-version for {{{flow_statistics}}} with modified loop structure k,i,j has been implemented. It is activated with preprocessor flag {{{-D__openacc}}}. The extra accelerator version is required because so far, the openACC standard only allows reduction operations on simple scalars. Since 1D-vectors along k are used in flow_statistics, they had to be replaced by scalars and the k loop has now to be used as the outermost loop. Additional 3D-flag arrays have been introduced for replacing the 2D-index arrays {{{nzb_s_inner}}} and {{{nzb_diff_s_inner}}} in routines {{{pres}}} and {{{flow_statistics}}}. Respective "global-sum" loops  are running from {{{k = nzb}}}. Within the loops, values for grid points below the surface (topography) are multiplied by zero, all others by one, using the flag array {{{rflags_invers}}}. This array is dimensioned (j,i,k) to allow for better cache usage in the loops of the accelerator version of {{{flow_statistics}}}.
+(flow_statistics, init_grid, init_3d_model, modules, palm, pres, time_integration)
+
+
+'''Changed:'''\\
+For PGI/openACC performance reasons (PGI compiler version 13.6, CUDA 5.0) the default compile options have been set to "{{{-ta=nocache}}}", which gives a speed-up of about 10-20%. For the same reason, the environment variable {{{PGI_ACC_SYNCHRONOUS}}} is set to 1 in the simple run script, which significantly improves the performance about 80%.
+(MAKE.inc.pgi.openacc, palm_simple_run)
+
+The type of flag array {{{wall_flags_0}}}, used in the Wicker-Skamarock scheme for advection of the vertical wind component, has been changed to 32bit {{{INTEGER}}}. An additional array {{{wall_flags_00}}} has been introduced to hold flag bits 32-63. This is required because the former used {{{KIND = SELECTED_INT_KIND(11)}}} caused wrong results with openACC.
+(advec_ws, init_grid, modules, palm)
+
+'''Bugfix:'''\\
+Dummy argument {{{tri}}} in 1d-routines replaced by {{{tri_for_1d}}} because of name conflict with array {{{tri}}} in module {{{arrays_3d}}}. (tridia_solver)
+}}}
 |----------------
 {{{#!td style="vertical-align:top;width: 50px"