Panzer: Workset Refactor Phase 1 and 2

[seamill]

Enhancement

@trilinos/panzer

Associated with #4230

In this issue we tackle phase 1 and 2 of the workset restructuring. To reiterate the goals of this work include:

• [x] Merge Workset and WorksetDetails into a single class.
• [x] Remove workset size from the WorksetContainer (moved to WorksetDescriptor)
• [x] Move orientation storage to WorksetFactory
• [x] Clean up workset construction interface (i.e. give the user a single, simplified path to build worksets)
  • [x] Worksets should not construct based on STK, but through an intermediary (currently the LocalMeshInfo object)
• [x] Modify workset so that requesting things using Descriptors will spawn new BasisValues/PointValues/IntegrationRules/etc.
• [ ] Update tests
• [ ] Update Drekar
• [ ] Update Charon
• [ ] Update EMPIRE

This work ended up changing a lot more code than expected, including:

• Removal of WorksetNeeds object (no longer used to construct worksets)
  • This is useful in that we don't need PhysicsBlocks to construct a Workset, HOWEVER, we do need field patterns (GlobalIndexer) during workset construction for orientation calculations.
• IntegrationValues2 had to be refactored/simplified to allow side assembled worksets to line up their integration points
• Worksets are now able to store auxiliary data not defined in Panzer using the 'Details' interface
• We now have to pass 'workset_size' though the various model evaluator construction interfaces since it is no longer stored in the WorksetContainer
• Creation of generalized 'OrientationsInterface' to simplify how orientations are stored in the WorksetFactory
• The changes to workset construction means that local cell indices are ordered differently than before - doesn't affect results, but broke a lot of tests
• 'Side' workset construction (e.g. BCWorksets/EdgeWorksets) can now be broken into multiple worksets instead of just the one
  • A consequence of this is that we no longer pass back a map of worksets indexed by subcell index, but instead a vector from which we can query the workset subcell index.
• Fixed the issue where the ConnManager/STK had different local indices for ghost cells than the LocalMeshInfo object
• LocalMeshInfo construction using STK now uses only the STK interface and doesn't generate any linear algebra structures to build connectivity
• Potential problem: Workset construction is going to be slower than before due to the need to retain connectivity information in the LocalMeshInfo object instead of the STK_interface (i.e. each time we partition the mesh to construct a workset, we need to restructure the connectivity). This will be fixed later, but for now we're just trying to pull Worksets construction away from STK.

Follow-up work to save for later:

• LocalMeshInfo object needs to be broken down into a Connectivity object and a Geometry object.
  • The connectivity object needs to support partitioning, as well as node/edge/face connectivity instead of just 'side' connectivity. Initially this would just be a non-STK ConnManager.
• Need a set of utility functions to generate simple LocalMeshInfo/ConnManager objects for cartesian and unstructured meshes without going through STK or writing it by hand. This would help move a lot of the adapters-stk tests into the disc-fe code.

[github-actions[bot]]

This issue has had no activity for 365 days and is marked for closure. It will be closed after an additional 30 days of inactivity. If you would like to keep this issue open please add a comment and/or remove the MARKED_FOR_CLOSURE label. If this issue should be kept open even with no activity beyond the time limits you can add the label DO_NOT_AUTOCLOSE. If it is ok for this issue to be closed, feel free to go ahead and close it. Please do not add any comments or change any labels or otherwise touch this issue unless your intention is to reset the inactivity counter for an additional year.

[github-actions[bot]]

This issue was closed due to inactivity for 395 days.