Let's kill c-iterate.c
N8TM@aol.com
N8TM@aol.com
Thu Sep 30 23:58:00 GMT 1999
In a message dated 9/8/99 4:36:35 PM EST, craig@jcb-sc.com writes:
> do these iterators represent a programmer's
> promise to not care *one whit* about the order or parallel-ability
> of iterated computations? I don't think that's *quite* the case for
> Fortran's vector notation, but am not knowledgeable enough about F90
> to say.
Fortran array assignments require that the entire calculation should be
carried out using the data values existing immediately prior to the
assignment. A compiler has the choice of figuring out what order of
operations is required to make this happen, or simply to place the results in
a temporary vector until finished, if there is a possibility of aliasing
which is not ruled out by the standard. If there's a clearer or more correct
way to state this, please let me know. FORALL opens more cans of worms but
asserts parallelizability.
Tim
tprince@computer.org
More information about the Gcc-patches
mailing list