Parallel Variables

A parallel variable (pvar) is similiar to a standard C variable except that it adds shape information to its type and storage class. While the shape defines rank, dimensions, and layout of parallel object, a parallel variable is the actual allocation of that space.

Declaring a Parallel Variable.

A shape must be first declared in order to declare a parallel variable:

shape [10][100]S;   /* fully defined, 2 dimensional, 10 by 100 shape */

float:S a;   /* parallel variable, is 2 dimensional, 10 by 100 elements */

Typically, any variable of Standard C type may be declared as a parallel variable:

shape [10]S;

typedef int:S Pi;   /* Type Pi is a parallel int of shape S */

extern int:S b;     /* b is an extern parallel int of shape S */

int:S y[5];         /* y is an array of 5 parallel ints of shape S */

struct date {
   int month;
   int day;
   int year;
 };

struct date:S hire_date;   /* hire_date is a parallel structure of shape S */

One can also declare pointers to parallel variables, such as:

float:S *pa;   /* pa is a pointer to a parallel float of shape S */

A parallel variable of shape "void" is a special type of parallel variable that can be used for declaring prototypes and actual arguments. A parallel variable of shape "void" is useful for functions which will accept a parallel variable of different types of shapes depending on the arguments that it is referenced with:

   shape [10]S;
   shape [10][10]T;
   int:S a;
   int:T b;
   ...
    func(a);
    func(b);
   ...

   func ( int:void pvar )
   ...
   pvar = 1;
   ...

When func is referenced with parallel variable "a", the assignment is a 1-d assignment of size 10. When referenced with "b", the assignment is a 2-d assignment of 10 x 10.

As noted earlier, the use of fully unspecified shapes may impede performance, this applies to void shapes. The temptation to use void shapes for flexibility in calling a function with varying types of shapes should be avoided if they flexibility is not needed. Remember, the trade-off with flexibility is often run-time performance.

Options:

Send comments or suggestions to dpce2 at crescentbaysoftware.com.