On Fri, 20 Apr 2007, Peter W. Draper wrote:
> On Thu, 19 Apr 2007, Tim Jenness wrote:
>
>> On Thu, 19 Apr 2007, Peter W. Draper wrote:
>>
>> > On Wed, 18 Apr 2007, Tim Jenness wrote:
>> >
>> > > It's an optimization problem. Removing -O2 and it works. (I've
>> > > upgraded your version).
>> >
>> > Tim,
>> >
>> > what compiler and OS was this?
>> >
>>
>> g95 from September 2 2005 (so ancient). Fedora Core 4 64-bit.
>
> Thanks, so that's one of the previous sightings, not a new one.
>
>> > > I'm half inclined to replace pda_d1mach.f with the simple and
>> > > portable C wrapper.
>> >
>> > Yes, the Fortran version does seem to be cropping up as an issue a bit
>> > too often.
>>
>> and PDA even includes the C code....
>
> Yes, I've extracted that and created a C version with a Fortran wrapper and
> updated PDA.
>
> This isn't activated yet as I'd like to check this out on as many
> platforms as possible first, just to make sure all is well (it seems
> quite fundamental). Could anyone who has a moment update and
> build/install PDA and then run:
>
> make d1machtest; ./d1machtest
>
> and report any failures/successes?
>
> I've checked Solaris 8&10/f77, Tru64 UNIX/f90, Ubuntu6.10/g95, SL4.2/g77 and
> FC6/gfortran, so OS X needs looking at (pluto seems to be bouncing me today).
OS X seems OK too (at least 10.3), so I've made the switch to the C
version of PDA_D1MACH.
I've also removed d1machtest, so there's no way to check this now.
Malcolm, I noticed that your new PDA_LMERR routine used MSG_ and ERR_
calls, I've replaced these with the EMS_ equivalents as PDA should only
depend on EMS.
Peter.
|