Reduction Operations in Parallelized Loops (C Only)

The compiler can recognize and properly handle most reduction operations in a loop during both automatic and explicit parallelization. In particular, it can handle reduction statements that have either of the following forms:

var = var op expr;
var assign_op expr;

where:

var
 Is an identifier designating an automatic or register variable that does not have its address taken and is not referenced anywhere else in the loop, including all loops that are nested. For example, in the following code, only S in the nested loop is recognized as a reduction:
int i,j, S=0;
#pragma ibm parallel_loop 
for (i= 0 ;i < N; i++) {
    S = S+ i;
    #pragma ibm parallel_loop 
    for (j=0;j< M; j++) {
        S = S + j;
    }
}
 
op
 Is one of the following operators:
+   -   *   ^   |   & 
assign_op
 Is one of the following operators:
+=   -=   *=   ^=   |=   &= 
expr
 Is any valid expression.

Recognized reductions are listed by the -qinfo=reduction option. When using IBM directives, use critical sections to synchronize access to all reduction variables not recognized by the compiler. OpenMP directives provide you with mechanisms to specify reduction variables explictily.


Program Parallelization
Countable Loops
Shared and Private Variables in a Parallel Environment


Control Parallel Processing with Pragmas


#pragma Preprocessor Directives for Parallel Processing
#pragma ibm critical Preprocessor Directive
info Compiler Option