OMPIndependent¶
The iterations of this region are independent of one another.
Declaration¶
Syntax¶
OMPIndependent()
Arguments¶
None.
Return values¶
None.
Описание¶
The clause records an assertion the compiler is allowed to rely on: no iteration of the region depends on another, so they may run in any order, on any number of threads, or on a device, without a dependence analysis.
Fortran’s do concurrent carries the assertion in the language. An OpenMP
parallel do does not — its iterations may synchronize with each other
through critical, atomic or ordered — so the clause is only written
once the body has been checked to hold no such construct.
A region with this clause must contain no nested OMPRegion that
synchronizes its iterations, which is what the parallel_canonicalize pass
checks before writing it. Every region that pass produces carries the
clause, whichever of the three constructs the region was written as, and no
other region does: it is therefore also what tells the passes below that a
region is a parallel loop they may choose a lowering for, rather than a
construct only the OpenMP pass lowers.
Examples¶
(OMPIndependent)