(j3.2006) co-array variable
Thu May 31 13:42:25 EDT 2007
Craig Rasmussen wrote:
>Could you please tell me what you mean by atomic in terms of the
>standard. I searched for atomic and it didn't come up. I worry
>about trying to equate scalar with atomic operations, but perhaps
>that wasn't your point.
It is probably better to avoid the use of "atomic" here since there is a
widely used concept of "atomic memory operations" which is different.
AMO's have been discussed as an addition to Fortran, but not accepted,
If you have a volatile variable, it is understood that it might be
changed by some agent not visible to the local execution context. Thus,
any reference needs to effectively involve a reload from memory, as
opposed to using a copy held in a register. The underlying assumption
for volatile is that the memory system operations are sufficiently
serialized that the load from memory will always get either the old
value or the new value, and not some bits from one and the remaining
bits from the other, at least on the granularity of a numeric storage
unit. The only thing new with co-arrays is that the "agent not
visible to the local execution context" could be a definition of the
variable by a different image.
Bill Long longb at cray.com
Fortran Technical Support & voice: 651-605-9024
Bioinformatics Software Development fax: 651-605-9142
Cray Inc., 1340 Mendota Heights Rd., Mendota Heights, MN, 55120
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the J3