[cig-commits] r22700 - seismo/3D/SPECFEM3D/trunk

[dkomati1 at geodynamics.org]

Author: dkomati1
Date: 2013-08-05 17:18:52 -0700 (Mon, 05 Aug 2013)
New Revision: 22700

Modified:
   seismo/3D/SPECFEM3D/trunk/todo_list_please_dont_remove.txt
Log:
updated the to-do list


Modified: seismo/3D/SPECFEM3D/trunk/todo_list_please_dont_remove.txt
===================================================================
--- seismo/3D/SPECFEM3D/trunk/todo_list_please_dont_remove.txt	2013-08-05 19:48:03 UTC (rev 22699)
+++ seismo/3D/SPECFEM3D/trunk/todo_list_please_dont_remove.txt	2013-08-06 00:18:52 UTC (rev 22700)
@@ -4,7 +4,7 @@
 
 (items listed in no particular order)
 
-- suggestion 42:
+- suggestion 43:
 ----------------
 To make the three codes (2D, 3D_Cartesian, 3D_GLOBE) conform strictly to the Fortran standard,
 we should replace all "call system(...)" statements with "call execute_command_line(...)", which is the
@@ -23,19 +23,12 @@
 
 - suggestion 40:
 ----------------
-Currently, the way of PML implemented for acoustic/solid simulation is memory consuming.
+Currently, the way PML is implemented for a fluid/solid simulation is memory consuming.
 We should in the future reduce the memory usage in the following way:
 1: replace the rmass_acoustic_interface(nglob_acoustic)  with rmass_acoustic_interface(nglob_acoustic_solid_interface)
 2: replace the potential_dot_dot_acoustic_interface(nglob_acoustic)  with potential_dot_dot_acoustic_interface(nglob_acoustic_solid_interface)
 where nglob_acoustic denotes the whole point set inside acoustic region, nglob_acoustic_solid_interface denotes the point set on acoustic_solid_interface
 
-- suggestion 39:
-----------------
-Currently, the way of PML implemented for fluid/solid simualtion is memory consuming.
-We should in the future reduce the memory usage in the following way:
-1: replace the rmass_acoustic_interface(nglob_acoustic)  with rmass_acoustic_interface(nglob_acoustic_solid_interface)
-2: replace the potential_dot_dot_acoustic_interface(nglob_acoustic)  with potential_dot_dot_acoustic_interface(nglob_acoustic_solid_interface)
-
 - suggestion 35:
 ----------------
 
@@ -44,12 +37,9 @@
 - suggestion 38:
 ----------------
 
-Add full loop vectorization for critical calculation loops, to make sure we gain a factor of about 1.8 in terms of CPU time as suggested
-by INTEL and as done by Cray and NCSA in their local version of SPECFEM3D_GLOBE. Already done (independently of Cray and NCSA) by Dimitri
-in SPECFEM3D_Cartesian, but the Cray and NCSA improvements for SPECFEM3D_GLOBE have not been merged yet (we asked NCSA several times, but
-unsuccessfully for now). However it would not be too difficult to redo these improvements independently by studying what Dimitri has done
-in SPECFEM3D_Cartesian (grep for "FORCE_VECTORIZATION" in the whole source code) and doing the same things in SPECFEM3D_GLOBE.
-The changes to make are straightforward, and reduce total wall clock-time by about 1.8 times, i.e., a very significant amount.
+Add full loop vectorization for critical calculation loops, to make sure we gain a factor of about 1.8 in terms of CPU time.
+Already done by Dimitri in SPECFEM3D_GLOBE (fully, grep for "FORCE_VECTORIZATION" to see where and how),
+but only partially done for now in SPECFEM3D_Cartesian.
 
 - suggestion 39:
 ----------------