Intel® Trace Analyzer and Collector User and Reference Guide
Visible to Intel only — GUID: GUID-48E24C20-9E5A-4524-9471-DC80ABAFDEB4
Visible to Intel only — GUID: GUID-48E24C20-9E5A-4524-9471-DC80ABAFDEB4
Defining and Recording Functions or Regions
Intel® Trace Analyzer and Collector can display and analyze general (properly nested) state changes, relating to function calls, entry/exit to/from code regions and other events occurring in a process. Intel Trace Analyzer and Collector implements a two-level model of states: a state is referred to by an activity name that identifies a group of states, and the state (or symbol) name that references a particular state in that group. For instance, all MPI functions are part of the activity MPI, and each one is identified by its function name, for instance MPI_Send for C and for Fortran.
The Intel Trace Collector API allows the user to define arbitrary activities and symbols and to record entry and exit to/from them. In order to reduce the instrumentation overhead, symbols are referred to by integer handles that can be managed automatically (using the VT_funcdef() interface) or assigned by the user (using the old VT_symdef() function). All activities and symbols are defined by each process that uses them, but it is not necessary to define them consistently on all processes (see UNIFY-SYMBOLS).
Optionally, information about source locations can be recorded for state enter and exit events by passing a non-null location handle to the VT_enter()/VT_leave() or VT_beginl()/VT_endl() functions.