Changeset 16229

Timestamp:

09/23/13 10:13:00 (11 years ago)

Author:

utke

Message:

CHG revert for this file only the change done with 16201 - as discussed with Eric - it is likely inappropriate to use the counts of indepedents and dependents to control the logic when e.g. in IoModel.cpp there is

this->FetchData(&autodiff,AutodiffIsautodiffEnum);
if(autodiff)

etc, Also, removing the guard right now causes the logic to fail for the parallel adjoint because in fact only on rank 0 have explicit dependent and independent designation been made and that needs to be reflected exactly on the falls to the Adol-C tape interpreters. Without the guard these interpreters complain (legitimatly) like this:

ADOL-C error: Reverse sweep on tape 2 aborted!

Number of dependents(1) and/or independents(91)
variables passed to reverse is inconsistent
with number recorded on tape(0/0)!

ADOL-C error: Reverse sweep on tape 1 aborted!

Number of dependents(1) and/or independents(91)
variables passed to reverse is inconsistent
with number recorded on tape(0/0)!

File:

• issm/trunk-jpl/src/c/analyses/ad_core.cpp (modified) (1 diff)

issm/trunk-jpl/src/c/analyses/ad_core.cpp

@@ -46 +46 @@
 
                         /*preliminary checks: */
-                        femmodel->parameters->FindParam(&num_dependents,AutodiffNumDependentsEnum);
-                        femmodel->parameters->FindParam(&num_independents,AutodiffNumIndependentsEnum);
-                        if(!(num_dependents*num_independents)) return;
+                        if (my_rank==0) {
+                                femmodel->parameters->FindParam(&num_dependents,AutodiffNumDependentsEnum);
+                                femmodel->parameters->FindParam(&num_independents,AutodiffNumIndependentsEnum);
+                                if(!(num_dependents*num_independents)) return;
+                        }
 
                         if(VerboseAutodiff())_printf0_("   start ad core\n");