Skip to content

Latest commit

 

History

History
121 lines (85 loc) · 3.79 KB

data.md

File metadata and controls

121 lines (85 loc) · 3.79 KB

Data Model

From the user's point of view, the FleCSI data model is extremely easy to use. Users can register data of any normal C++ type. This includes P.O.D. (plain-old-data) types and user-defined types.

Data Clients

Field Data

Handles

Permissions

In order for the FleCSI runtime to infer distributed-memory data dependencies that must be updated during task execution, the user must specify what permissions are required for each client or field handle type for each task. Currently, permissions are specified as template-style arguments to the parameters in the task signature:

  void task(mesh<ro> m, field<rw, rw, ro> f) { // ...

The permissions specifiers have the following meanings:

  • ro [read-only]

    Read-only access implies that ghost values will be updated if necessary before the user task is executed.

  • wo [write-only]

    Write-only access implies that shared values will be updated if necessary after the user task is executed.

  • rw [read-write]

    Read-write access implies that ghost values will be updated if necessary before the user task is executed, and that the shared values will be updated if necessary after the user tash is executed.

Storage Classes

The FleCSI data model provides an intuitive high-level user interface that can be specialized using different low-level storage types. Each storage type provides data registration, data handles, and data mutators (when appropriate) that allow the user to modify data values and structure. The currently supported storage types are:

  • dense

    This storage type provides a one-dimensional dense array suitable for storing structured or unstructured data.

  • sparse

    This storage type provides compressed storage for a logically dense index space.

  • global

    This storage type is suitable for storing data that are non-enumerable, i.e., data that are not logically stored as an array.

  • color

    This storage type is suitable for storing non-enumerable data on a per-color basis.

  • tuple

    This storage type provides c-struct-like support so that task arguments can be associated without forcing a particular data layout.

Example

flecsi_register_data_client(mesh_t, hydro, m);
flecsi_register_field(mesh_t, hydro, pressure, double, dense,
  1, cells);

void update_pressure(mesh<ro> m, field<rw, rw, ro> p) {
  
  for(auto c: m.cells(owned)) {
    p(c) = 2.0*p(c);
  } // for

} // update_pressure

Data Model

The FleCSI data model provides an intuitive high-level user interface that can be specialized using different low-level storage types. Each storage type provides data registration, data handles, and data mutators (when appropriate) that allow the user to modify data values and structure. The currently supported storage types are:

  • dense

    This storage type provides a one-dimensional dense array suitable for storing structured or unstructured data.

  • sparse

    This storage type provides compressed storage for a logically dense index space.

  • global

    This storage type is suitable for storing data that are non-enumerable, i.e., data that are not logically stored as an array.

  • color

    This storage type is suitable for storing non-enumerable data on a per-color basis.

  • tuple

    This storage type provides c-struct-like support so that task arguments can be associated without forcing a particular data layout.