- From: ak.miller at auckland.ac.nz (Andrew Miller)
- Subject: [cellml-dev] PCEnv under Vista 64-bit, XP 32-bit and XP 64-bit
- Date: Wed, 26 Sep 2007 17:40:10 +1200
Randall Britten wrote:
>
Hi
>
>
I wonder if we should be putting these kinds of discussions in the tracker
>
anyway. The Cmgui project use their tracker that way, and it allows them to
>
build up a knowledge base. What do you think?
>
Hi,
I think that it would be a good idea to do this, except that Alan
currently can't access the tracker and so we obviously can't do this
until that issue is resolved.
Best regards,
Andrew
>
Regards,
>
Randall
>
>
-----Original Message-----
>
From: Alan Garny [mailto:alan.garny at dpag.ox.ac.uk]
>
Sent: Wednesday, 26 September 2007 5:19 p.m.
>
To: 'Andrew Miller'
>
Cc: 'Randall Britten'
>
Subject: RE: PCEnv under Vista 64-bit, XP 32-bit and XP 64-bit
>
>
Hi Andrew,
>
>
>
>> Anyway, I have been doing some tests on different Windows platforms
>
>> and came across something rather worrying in my opinion. Basically, I
>
>> get different results using different operating systems.
>
>>
>
>> I ran the attached CellML model using the official version of PCEnv
>
>> (i.e. 0.2) under:
>
>>
>
>> - Vista Ultimate 64-bit (my main operating system),
>
>>
>
>> - XP Pro 32-bit (in a VM), and
>
>>
>
>> - XP Pro 64-bit (in a VM).
>
>>
>
>> I cannot recall for certain the parameters I used for the simulation,
>
>> but they were definitely the same in all cases. The simulation
>
>> finished, I exported the results and manually removed all the data,
>
>> but the ones about the transmembrane potential. As you can see from
>
>> the attachment, the results I got are different. I understand that
>
>> this might be the case for various reasons (e.g. gcc generates a
>
>> slightly different code on each operating system, the data are not
>
>> necessarily generated at the exact same time). Still, if you plot
>
>> those data, you cannot help but notice that the results are indeed
>
>> different.
>
>>
>
>> I have the feeling that this might because of gcc (or any compiler for
>
>> that matter), but I cannot currently test that hypothesis and was
>
>> wondering whether you could at least confirm to me that you are also
>
>> experiencing the same behaviour?
>
>>
>
>>
>
> I don't actually have a 64-bit Windows system available to me, so I
>
> can't confirm this. However, it is worth noting that different platforms
>
> have different FPU behaviour and differences in precision. If this is
>
> making a major difference to the end-point then this probably raises
>
> concerns about the stability of the model under the integrator used.
>
>
>
>
Fair point. I will try to run the same kind of test using different models
>
to see what we get. Note, though, that even between Vista Ultimate 64-bit
>
and XP 64-bit, I got different results...
>
>
Alan.
>
>
>
- [cellml-dev] PCEnv under Vista 64-bit, XP 32-bit and XP 64-bit, Andrew Miller, 09/26/2007
Archive powered by MHonArc 2.6.18.