(j3.2006) F03/0084, was Re: Interp letter ballot #25

Bill Long longb
Fri May 11 23:09:46 EDT 2012



On 5/11/12 5:56 PM, Dan Nagle wrote:
> Hello,
>
> On May 11, 2012, at 14:48 , Bill Long wrote:
>>
>> When you get to GPU-land, you end up returning  .false. for many of the "ieee_support..." inquiries in Clause 14.  Particularly, IEEE_SUPPORT_HALTING which is the one most people are interested in. (Among the small fraction of users who use any feature in Clause 14.)
>
>
> Most applications I've seen are coded for a combination of host
> and device.  Exactly what IEEE modes mean in a heterogeneous environment

Right.  And it is not always obvious whether a particular operation is 
happening on one or the other.

> is unclear to me.  I suspect that a great deal of work is required
> to make the contrary plainly appear.  Indeed, in an environment
> supporting asynchronous execution, the precise meaning of an exception

An exception occurs at a particular instruction in the code, not at a 
particular time.  So asynchronous, by itself, is not a killer.  (People 
have been using exceptions in MPI codes for years.)   However, with SIMD 
instructions and multiple threads, things get even less precise. 
Assuming exceptions are registered at all.

> is likewise unclear (at least to me).  Perhaps the issue can be resolved,
> but portably?

The easy solution is to just give up and say the processor does not 
support the feature.   Fortunately, the standard effectively allows that 
option.

Cheers,
Bill


>

-- 
Bill Long                                           longb at cray.com
Fortran Technical Support    &                 voice: 651-605-9024
Bioinformatics Software Development            fax:   651-605-9142
Cray Inc./Cray Plaza, Suite 210/380 Jackson St./St. Paul, MN 55101





More information about the J3 mailing list