MeshInterface¶

The heart of $T^{\infty}$ is the MeshInterface that describes read-only access to an unstructured grid. MeshInterface objects are the primary output of the pre-processing step but are used nearly all $T^{\infty}$ plugins.

Only two mesh entities are described: nodes and cells. Additional entities (edges, faces, etc) must be derived from cells and nodes.

Partitioning¶

Each instance of a MeshInterface represents one partition of an unstructured grid. Mesh entities (nodes and cells) may be resident on more than one partition, however they will be owned by only one partition. If a cell is resident on a partition, all nodes within that cell must also be resident on the partition. Entities that are resident but not owned are typically referred to as ghost or halo entities.

Node and Cell IDs¶

Each node has a local node identifier (ID) and a global node ID. Likewise, each cell has a local cell ID and a global cell ID. Local IDs cannot be sparse. Local IDs begin at 0 and end total number of local entities - 1. Local IDs are only used by the local partition, no other partition should know a local ID for any entity. This allows a local partition to reorder entities (for cache efficiency, bandwidth reduction, or convienience) without communicating that reordering.

Unlike local IDs, the global ID of an entity is unique across all partitions that entity is resident. The set of global IDs across all partitions cannot be sparse, they must begin at 0 and end at total number of entities - 1. However, global IDs on a single partition will almost always be sparse.

Note

Local IDs are represented using a 32 bit integer, global ids are represented using a 64 bit integer. Complex simulations may require more than 2.2 billion total mesh entities; however, it is considered unlikely that any one partition will have over 2.2 billion resident entities.

Mesh Domains¶

$T^{\infty}$ supports multiple simultaneous mesh domains. Multiple domains are required for overset simulations. Each domain is partitioned across all the ranks of a specific MPI communicator. Global IDs are only valid within a single domain. There is no entity identifier valid across multiple domains.

Conventions¶

Cells are accessed by calling void MeshInterface::cell(int cell_id, int* cell_ptr), the same call is used for both volume and surface cells. This call writes the local node IDs for the cell into the passed pointer by using the CGNS convention for node ordering. (See https://cgns.github.io/CGNS_docs_current/sids/conv.html). It is the caller’s responsibility to ensure there is sufficient memory to fit all the nodes in the requested cell. The length of a cell is determined calling CellType MeshInterface::cellType(int cell_id) to get the cell’s type. Then calling int MeshInterface::cellTypeLength(CellType type) to get the length of that type. At this time only fixed width cell types are supported. (NGON and NFACE are not supported.)

Surface cells are listed using the same cell(int cell_id, int* cell_ptr) calls. $T^{\infty}$ does not differentiate between surface and volume elements in the cell accessors. Surface cells must be oriented so that the unit vector from a right handed walk of the vertices produces a unit normal pointing out of the neighboring volume element.