NEW Variables

 

The NEW clause asserts that the named variables act as private variables to each iteration of the INDEPENDENT loop. That is, there would be no interfering assignments and uses in the loop, and thus no change in ghe behavior of the program, if new objects were created for the NEW variables for each iteration of the DO loop and those objects were destroyed at the end of each iteration. Thus, no values flow into NEW variables from execution before the loop, no values flow from NEW variables to execution after the loop, and (most importantly) no values flow from one iteration to another through NEW variables.

Advice to users.A pointer or allocatable variable may appear in a NEW clause. The interpretation of the paragraph above, in these cases, is that one should not rely on the value, the association status, or the allocation status of such a variable on entry to the loop; rather, such variables should be allocated or pointer assigned in the loop body before they are used. It would also be advisable to deallocate or nullify such a variable in the loop body after its last use as well. (End of advice to users.)

Rationale.NEW variables provide the means to declare temporaries in INDEPENDENT loops. Without this feature, many conceptually independent loops would need substantial rewriting (including expansion of scalars into arrays) to meet the rather strict requirements described above. Note that a temporary must be declared NEW only at the innermost lexical level at which it is assigned, since all enclosing INDEPENDENT assertions must take that NEW into account. Note also that index variables for nested DO loops must be declared NEW; the alternative was to limit the scope of an index variable to the loop itself, which changes Fortran semantics. FORALL indices, however, are restricted by the semantics of the FORALL; they require no NEW declarations. (End of rationale.)

 
• Examples of NEW