CVS log for tahoe/src/main/FEManagerT.cpp

need SendEqnsToSolver as public for constructing global matricies arbitrarily

add function to return cohesive strength

added Veronda Westmann potential
and made minor changes for debugging of Diffusion Element

merging changes from BRANCH_david_spatial_2

add default parameters to fields of solver_phase

corrected path to partition file

track global print step number

(1) partition information moved to CommManagerT
(2) external I/O constructed in TakeParameterList

write flag to mark changing geometry to to .io.ID file

better error messages

changes for revised CommManagerT

(1) InitStep now sets the system configuration with UpdateConfiguration before
    calling InitStep on rest of system.
(2) Relaxation now handled in SolveStep instead of by the solvers

merging changes from BRANCH_david_spatial_2

added note

In CloseStep, changed order so that elements are closed before nodes. This
keeps information from nodes (e.g. last displacements) around longer to
allow elements to use this information to update

revision to new flow chart for InitStep. Some 2D benchmarks pass, some don't,
but they all run.

work on meeting new flow chart, reconfiguration and renumbering happen in initstep
MD only work so far, not all benchmarks pass

GCC4 doesn't like variable name exception in the argument list to catch

(1) guess input file format from the file name if the type isn't specified
(2) select output file format based on the input file type if the type isn't specified

extract title from the input parameters

added method for sending only nodal data to output

set amount of runtime output based on logging flag

added calls to node manager to allow calculation of the total system energy

MD only dynamic bridging scale calculations
added ignore_continuum to top level bridging xml input file

remove unused input parameter

don't bother trying to renumber equations for groups without any equations.

Merge Changes from BRANCH_david_2
includes rework of MD parallel implementation

Merged changes from BRANCH_david_2 to BRANCH_david_2_merge to prepare to merge into main branch

slightly revised ordering of construction in TakeParameterList. The sequence
still does not allow CommManagerT::Configure to be called with spatial
decomposition b/c the node manager hasn't been constructed yet. This will
need to go on another branch.

clean up

updated to reflect change in PartitionT.h

removed debugging code

corrected definition of sub-lists in decomp_method_choice

reivsed handling of the decomposition method. The code now calculates a
spatial decomposition, though it cannot compute with it yet.

merging changes from BRANCH_david_1: revised handling of model decomposition
for parallel execution

clean-up before merge

cleanup before merge with main branch

use proper operator to compare to GlobalT::SystemTypeT values

look for decomposition files and generate them if needed in TakeParameterList

(1) removed InitCodeT
(2) corrections with task
(3) corrections reading partition information

continued debuging parallel standard tahoe

revised initialization process so pointer to dynamically allocated objects
are stored before initialization so memory is properly cleared if an exception
occurs during initialization. Old implementation leaked memory under these
conditions.

added some debugging code

debugging the parallel run

New class created to perform decomposition

changed occurrence of max_loops

finished move of classes from FEManagerT_mpi to FEManagerT, added task variable where needed

updated changes from integration of FEManagerT and FEManagerT_mpi

combined FEManager_mpi and FEManager

(1) removed unused functions/variables
(2) use SolverT factor method to construct solvers

merging changes from BRANCH_parameter_4. Tahoe is now XML-powered.

removed unneeded code

clean up

corrections for parallel execution

continued merge of changes from BRANCH_parameter_3

merging changes from BRANCH_parameter_3

changes to restore parallel capability. Decomposition working again.

corrected evaluation of pass condition for solver phases

merging changes from BRANCH_penetration_2: changes need to make command
line options accessible from within the FEManagerT

merging changes from BRANCH_bridging_6: changes to allow assignment of
equations numbers beginning at some shift.

1.  Modified files to account for ifstreamT.h and ofstreamT.h changes on main line

updates for changes to fstreamT/ifstreamT/ofstreamT

revised J2 materials

added error check

enabled Newton's method with line search

merging changes from BRANCH_bridging_5

merging changes from BRANCH_parameter_2. BRANCH_parameter_3 continues on
with revisions from the main line since BRANCH_parameter_2.

allow empty element list

make command line options accessible through FEManagerT and ElementSupportT

WriteOutput is const

corrected getting echo_input and compute_IC flags from the parameter list

(1) replaced ValueT::String parameters with ValueT::Word's
(2) added optional "title" and "author" fields to tahoe parameter list

corrected initialization of fReadRestart flag

FEManagerT::SetEquationSystem now takes an equation number shift parameter
to move assignment of equations away from the default origin

merging changes from BRANCH_bridging_4:
(1) corrected atomic volume in virial stress calculations and pair potential
    Cauchy-Born materials
(2) added flag to control whether image nodes are included in the list of
    non-ghost nodes
(3) corrected cubic spline window function with spherical support
(4) added selectable window functions and parameters for MeshfreeBridgingT

added wrappers for writing snap shot over a set of nodes

more error checking during construction

more revised construction patterns and removal of unneeded code. First test
of new code now runs!

set/get solver phase parameters

revised construction patterns

revised construction patterns

changes to construct and initialize problem using XML input

working toward running from XML input

changes for switch to XML input

merging changes from BRANCH_paklein_3:
(1) ResetStep now returns a GlobalT::RelaxCodeT to signal the global matrix
needs to be reset as well.
(2) write restart information from element groups to separate files

changes for revisions to integrators

use wrapper macro for dynamic_cast

write restart information from element groups to separate files

Restoring a line that should be in the repository version.

Changes to handle OutputSetT::ModeT kElementFromSideSet.

stricter enforcement of const-ness in accessing elements lists, arrays, and matricies

-j HEAD, altered slip vector and strain calculations for Periodic BC's

merging changes from BRANCH_bridging_3:
(1) added output of nodal residuals during iteration output if rank check
    code != 0
(2) added accessors to RHS and LHS by solver group

only write system output (displacement and residual) for nodes owned by
each processor

added code to write residual force during iteration output if solver
check code is different from 0 (zero).

(1) beginnings of writing nodal residuals during iteration output
(2) updates for revised TimeManagerT
(3) write ghost node information during echo of input

added capability to run FE/MD coupled problems with a monolithic solution
scheme using the MultiManagerT to drive two sub-FEManagerT's

added kRelaxSystem status flag

ResetStep now returns a GlobalT::RelaxCodeT to signal the global matrix
needs to be reset as well.

clean-up writing restart files

temporary extension to the Newton+LS solver

small changes to avoid UMR warnings during ParameterTree::BuildBranch

added PCG solver to parameter tree

derive from ParameterInterfaceT and define some parameters and sub-lists

changes to construct IOManager as early as possible, though configuration
is still the same as before.

changes for new analysis code for nonlinear heat conduction

(dummy check-in)
(1) moved ReadRestart into InitialCondition
(2) ReadRestart now returns bool if restart information is read
(3) modified SetTimeStep to change the time step system-wide, including
    in the TimeManagerT and the SolverT's.
(4) Modified InitialCondition so that if the initial condition of the
    system is not read from a restart file, it is calculated at t = 0
    with dt = 0.
(5) modified WriteRestart so output increment counters don't advance when
    it is called with dt = 0.

explicit calls to ReadRestart in RunBridging not needed anymore because
it has been incorporated into InitialCondition

skip call to renumbering equations if the group contains more than one
field. The RelabellerT is not implemented to handle this. It assumes
each row in the equations iArray2DT's passed in corresponds to a unique
tag in the connectivities.

Returning sets of  equivalent nodes in connectivies generated by KBC's and
FBC's thus informing mesh decomposition algorithms that some sets of nodes (only those
(only ones generated by TiedNodesT so far) should be treated as the same even
though they have differing node IDs.

merging changes from BRANCH_bridging_1: added new "multi-Tahoe" execution
modes where two instances of the Tahoe simulation object are created to
run separate calculations that are coupled.

changes to allow SolveStep to be called multiple times for the same
time increment, so multiple Tahoe's can be solved in a staggered
fashion.

changes to release locks on assemble. Also provide access to solvers

All the bridging code is in place, but hasn't been run through yet.

tying atom and continuum solvers. Put a time sequence solve loop on the
execution manager. Just need that projection.

revisions for new code layout. builds with CodeWarrior. not benchmark
tested yet.

merging changes and conflicts from BRANCH_atom_1

relax system after reading restart

first finished implementation of periodic boundary conditions for serial
and atom decomposition. Code has not been tested yet, and something isn't
right with restarts

moving MP tasks to CommManagerT

moving more responsibility and data structures to CommManagerT

moving more work for parallel execution into CommManagerT

setting up CommManagerT to handle different decomposition schemes

revised constructor for CommManagerT

start incorporation of CommManagerT

another mistake with the current solver group fixed

was tracking the current solver group number in the wrong place.

(1) current group flag is now correct during initialization of equation
    system and sending equations to solver. Both triggered by solver.
(2) Version() now returns static const char

replaced flag __MPI__ with flag __TAHOE_MPI__ because the old flag is pre-defined in some environments

changes argument in AssembleRHS from dArrayT to nArrayT<double> so sub-classes
can be assembled as flat arrays

(1) pass through new version of AssembleLHS
(2) use ExceptionT methods instead of throw

translated line endings

removed long unused argument from call to WriteOutput

(1) added a "current group" flag to indicate which group is currently being solved.
(2) also now use the InGroup method to determine which elements contribute to which equation group instead of Group since each element may contribute to multiple groups.

merge changes from BRANCH_paklein_1:
(1) update to use ExceptionT
(2) removed deprecated Allocate statements. Replaced with Dimension

replaced deprecated calls to Allocate with the new method Dimension in
classes from toolbox/abc

updated to ExceptionT

toolbox header Constants.h has been renamed toolboxConstants.h

added method NodeToProcessorMap which returns the home processor for
the nodes given in the list

(1) incorporating CommunicatorT to handle MP messages
(2) restored exception handling for MP jobs

when renumbering equations, disconnected graphs were halting
the code. How it just skips renumbering.

started replacing direct calls to the MPI library with the
CommunicatorT class (toolbox).

Merging BRANCH_name_1 onto main branch. The code compiles and
passes the benchmarks on SGI, linux, Sun, and CodeWarrior in Win2K.

Initial wrapping of namespace Tahoe around libtahoe. It currently compiles
on linux and passes the benchmarks. Other systems haven't been tried yet.

merge updates from BRANCH_field_1

revised error message

allow more than one solver phase per solver, but require that each
solver gets at least one phase.

added more solution control for multi-solver problems. User specifies the
number of iterations for each solver per loop, the number of interations
per solver to convergence, and the total number of loops through the
solvers per time step.

WriteSystemConfig now writes all time derivatives of the nodal field
variables

fixed error in step cut routine

clean up

WriteParameters updated for multi-field parameters

bug fixes

running level.0 benchmarks, except MPS (no active equations)

(1) NodeManagerPrimitive -> NodeManagerT
(2) revising FEManagerT::Solve

the whole business now compiles and links with CodeWarrior. Still need
to revise FEManagerT::Solve before we can see it the whole thing runs.

yet another day's good, good work:
(1) reworking the elements to use ElementSupportT instead of getting/putting
    info directly to the FEManagerT/NodeManagerT/etc.
(2) general trimming, cleaning, fixing...

another day's work:
(1) the node manager tree is revised and compiles
(2) the solver tree compiles but will need revision
(3) controller tree seems OK
(4) the primitives tree compiles
...still need to work on the elements and getting the whole business
up and running.

the day's check-in of multi-field development

construction of time integrators moved to TimeManagerT

changes to support functions needed by the NOX solver and to work with
revisions to the controllers

(1) the output set to processor map file, used during -join_io, is now
    associated with the parameter file name instead of the geometry file.
(2) corrected issue associated with setting the output time stamp when
    writing output data. Previously, the time was only set during WriteOutput.
    Now a default value is set during InitStep. The old was causing trouble
    with history node output between calls to WriteOutput.

revised IOManager::SetCoordinates

revised OutputBaseT::SetCoordinates

revised listing of block data per output set to handle free sets

added error checking for initialization of IOManager

all references to element/node/side/output sets have been updated to
use StringT id's instead of int's

added kPML analysis code which runs explicit dynamics with nsd*nsd degrees of freedom per node

added missing headers removed from FEManagerT.h

send model manager messages to cout instead of the .out file

added more errorchecking during model manager initialization

for decomposition (and joining?), the model manager cannot be initialized as readonly == true, because this means the input manager isn't constructed and no model data can be read.

merged BRANCH_model_1 and resolved conflicts

revised to work with revised console classes

(1) new version number -> v3.5
(2) removed VariFD nodes classes. Were not being used anymore

1. Modify ReadParameters and WriteParameters
2. Modify Initialize
3. Remove SetInput
4. Move fInputFormat and fModelFile to FEManagerT_mpi, could eventually
remove entirely when decomposition uses ModelManagerT.
5. Remove functions to return fInputFormat and fModelFile.

Comment out references to VariFDNodesT and VariFDDynNodesT from AddNode and
SetNodeManager.

1. Renamed SetIO function to be SetOutput.
2. Added call to SetInput from Initialize.
3. Created function SetInput to initialize ModelManagerT.
4. Created object ModelManagerT*.
5. Created function ModelManager to return ModelManagerT*.
6. Added ModelManagerT* to deconstruction list.

update from compatibility with new controller interface

added call to nodes for RegisterOutput after the elements register for output

moved call to WriteOutput from after to before CloseStep for the nodes and
elements. This simplifies handling of internal state variables at output.

enabled Abaqus output, but haven't verified the output format

(i) new input file version because of new ElasticT element output
(ii) corrected problem with relative paths to restart files

made changes to work with changes to IOManager to allow something other than
the true time set during WriteOutput for things like the iteration output, for which
the time doesn't change.

open external files with relative path from main input file

new assembly operators:
- matrix assembly with independent row and col destination arrays.
   not supported for all global matrix types. will get error message
   where not supported.
- equation numbers for assembly are now passed as nArrayT not
   iArrayT, for better flexibility in gathering equation numbers

version change to 3.4 and move to relative paths for external files

better exception messaging during WriteOutput

new version number with change to element output specifications
for cohesive surface elements.

Previously, a write restart increment of 1 meant just one restart
file for the whole simulation at the end. Now, 1 is treated like
all other restart increments.

Added DissambleLHSDiagonal to allow access to values on the diagonal
of the global LHS matrix.

To support joining element output variables, the ID's for all element
blocks in an output set are written to "[job name].io.ID", by rank 0,
immediately after the output set it registered.

Descending into sub-directories is no longer done by treating the sub-directories
as targets. This caused problems because some directories were missed and mult-
iple directories were being processed simulatenously. Directories are not hit one
at a time using "for" loops. Also, a common set of suffix rules in now in
    macros/suffix
and "make" instructions for the sub-directories are now read from
    macros/subdir.targets
NOTE: the make system in the new macros module is not compatible with the
old make system. To rebuild, you must update all modules.

The current input file version has changed to v3.02 because of
changes to the element input. For allowing writing in some "backward
compatibility", the input file version is how available from FEManagerT
using
	const StringT& Version(void) const
Version strings can be compared using StringT::versioncmp

initial check-in

Initial revision