Changeset 18888

Timestamp:

11/28/14 10:02:30 (10 years ago)

Author:

Eric.Larour

Message:

CHG: new control core that relies on m1qn3, but computes the gradients using the AD technology. The update of the control
inputs is also interesting in that we do not rely on a pid indexed ordering, but an sid indexed ordering. Using this
ordering, for each run of simulad, we feed the updated X vector of independents to a new FemModel (yes, we create a new
FemModel) initialized from the disk input file, except for the X vector itself. The reason we have to do this is to ensure
that the active AD variables are the same for each run, and because we don't know yet how to stop and restart a tape with
only a subset of the active variables updated -> we are not sure whether this would actually work!

Location:

issm/trunk-jpl/src/c/cores

Files:

• WrapperCorePointerFromSolutionEnum.cpp (modified) (1 diff)
• controlad_core.cpp (added)

issm/trunk-jpl/src/c/cores/WrapperCorePointerFromSolutionEnum.cpp

@@ -47 +47 @@
                         case 2: solutioncore=controlm1qn3_core; break;
                         case 3: solutioncore=controlvalidation_core; break;
+                        case 4: solutioncore=controlad_core; break;
                         default: _error_("control type not supported");
                 }