Technical data

#pragma Directives [3]
For all work distribution arguments except single, each chunk of iterations can
be vectorized, subject to the normal rules for vectorization.
A
rgument
D
escription
single The single argument specifies that the iterations
are distributed one at a time to available
processors. If no work distribution arguments are
specified, the default is single.
chunksize (
exp) The chunksize argument specifies that the
iteration space is broken into chunks of size exp,
with a possibly smaller residual chunk. exp is
an integral expression enclosed in parentheses
following the chunksize argument. This
expression is evaluated at run time, just prior to
execution of the loop. If chunksize is less than or
equal to 0, the chunksize is assumed to be 1.
Otherwise, if the chunksize value is greater than
the number of iterations in the loop, the value is
set to the number of iterations. The chunksize
expression cannot contain side effects.
numchunks (exp) The numchunks argument specifies that the
iteration space is broken into exp chunks of
equal size, with a possible smaller residual
chunk. exp is an integral expression enclosed in
parentheses following the numchunks argument.
This expression is evaluated at run time, just
prior to execution of the loop. If the numchunks
expression value is less than or equal to 0, the
value is set to 1. If the numchunks expression
value is greater than or equal to the number
of iterations, the value is set to the number of
iterations. The numchunks expression cannot
contain side effects.
guided When guided is specified, each of the iterations
are handed out in pieces of exponentially
decreasing size. Iterations are dispatched in an
attempt to balance the load on each processor
so that all processors finish their work at
approximately the same time. When only 1
iteration remains, it is dispatched to the next
processor, and the distribution is complete. The
S–2179–36 83