Prev: error with newfftd
Next: add_block to copy a block
From: Steve Amphlett on 16 Mar 2010 08:00
64-bit C-MEX s-functions built using R2006b will not load into R2009a. Only 64-bit, 32-bit is still ok. I searched the cumulative release notes between these versions, but could not find anything that said these files need to be rebuilt. So somewhere between these two versions, compatibility got broken. Does anyone know when (which version) that happened?

My goal (as ever) is to use the oldest version of Matlab that still supports the most recent release (and all releases between).

Thanks in advance,

- Steve
From: Rune Allnor on 16 Mar 2010 09:04
On 16 Mar, 13:00, "Steve Amphlett" <Firstname.Lastn...(a)Where-I-
Work.com> wrote:
> 64-bit C-MEX s-functions built using R2006b will not load into R2009a.  Only 64-bit, 32-bit is still ok.  I searched the cumulative release notes between these versions, but could not find anything that said these files need to be rebuilt.  So somewhere between these two versions, compatibility got broken.  Does anyone know when (which version) that happened?

What compilers were involved? It seems that the MS Visual
Studio are evolving. The /Wp64 flag, that checks for 32bit/
64bit portability issues worked in VS2005 but is flagged as
'deprecated' in VS2008.

Seems to be the kind of thing you might want to look in to.

Rune
From: Steven Lord on 16 Mar 2010 10:06

"Steve Amphlett" <Firstname.Lastname(a)Where-I-Work.com> wrote in message
news:hnnrsl$3qd$1(a)fred.mathworks.com...
> 64-bit C-MEX s-functions built using R2006b will not load into R2009a.
> Only 64-bit, 32-bit is still ok. I searched the cumulative release notes
> between these versions, but could not find anything that said these files
> need to be rebuilt. So somewhere between these two versions,
> compatibility got broken. Does anyone know when (which version) that
> happened?

You said you searched the cumulative release notes ... did you include the
compatibility summaries when you performed that search?

http://www.mathworks.com/access/helpdesk/help/techdoc/rn/bqsrae0.html

There's an item in the R2009a compatibility summary that indicates that
"MEX-Files Calling BLAS or LAPACK Functions Must Be Updated On 64-Bit
Platforms" -- does your MEX-file use BLAS or LAPACK?

Does it use the version 5 API (also called out in the 9a compatibility
summary)?

Since you mentioned that you're using these as S-functions, there's a
similar table for Simulink:

http://www.mathworks.com/access/helpdesk/help/toolbox/simulink/rn/bqmg7b6-6.html

I don't see anything offhand that would only affect 64-bit platforms, but
then again I don't know what your S-functions do so perhaps you'll see
something that affects your S-functions that I didn't.

If you receive an error when loading these S-functions into release R2009a,
I also recommend you search for that message on the support website; if that
too doesn't reveal why they are not working in release R2009a, I suggest you
send the error (and the MEX-file, if possible) to Technical Support so the
support staff that specialize in working with Simulink or those who
specialize in MEX-files can investigate the cause of this problem.

--
Steve Lord
slord(a)mathworks.com
comp.soft-sys.matlab (CSSM) FAQ: http://matlabwiki.mathworks.com/MATLAB_FAQ


From: Steve Amphlett on 16 Mar 2010 12:06
"Steven Lord" <slord(a)mathworks.com> wrote in message <hno39d$feh$1(a)fred.mathworks.com>...
>
> "Steve Amphlett" <Firstname.Lastname(a)Where-I-Work.com> wrote in message
> news:hnnrsl$3qd$1(a)fred.mathworks.com...
> > 64-bit C-MEX s-functions built using R2006b will not load into R2009a.
> > Only 64-bit, 32-bit is still ok. I searched the cumulative release notes
> > between these versions, but could not find anything that said these files
> > need to be rebuilt. So somewhere between these two versions,
> > compatibility got broken. Does anyone know when (which version) that
> > happened?
>
> You said you searched the cumulative release notes ... did you include the
> compatibility summaries when you performed that search?
>
> http://www.mathworks.com/access/helpdesk/help/techdoc/rn/bqsrae0.html
>
> There's an item in the R2009a compatibility summary that indicates that
> "MEX-Files Calling BLAS or LAPACK Functions Must Be Updated On 64-Bit
> Platforms" -- does your MEX-file use BLAS or LAPACK?
>
> Does it use the version 5 API (also called out in the 9a compatibility
> summary)?
>
> Since you mentioned that you're using these as S-functions, there's a
> similar table for Simulink:
>
> http://www.mathworks.com/access/helpdesk/help/toolbox/simulink/rn/bqmg7b6-6.html
>
> I don't see anything offhand that would only affect 64-bit platforms, but
> then again I don't know what your S-functions do so perhaps you'll see
> something that affects your S-functions that I didn't.
>
> If you receive an error when loading these S-functions into release R2009a,
> I also recommend you search for that message on the support website; if that
> too doesn't reveal why they are not working in release R2009a, I suggest you
> send the error (and the MEX-file, if possible) to Technical Support so the
> support staff that specialize in working with Simulink or those who
> specialize in MEX-files can investigate the cause of this problem.


Thanks Rune & Steven.

My thinking that there was a compatibility break was based on a customer reporting failure to load the MEX file, combined with a colleague of mine who reported something similar a while back. So I installed R2009a on a Vista/XP64 dual-booter and rebuilt the MEX file.

I'm just done with some additional testing. It seems that my MEX file may be built using either R2006b or R2009a and it'll still load into R2009a. Under both XP64 and Vista. Grrrrr! I can't repeat the reported failure ... and it's on a customer computer in Japan, so it's not like I can wander across the site to fix it, or even take over his desktop remotely.

The compatibility tables are really nice. I'd not seen them before, ploughing through miles of release notes instead. I couldn't see anything in any of those tables to suggest a compatibility break.

I expect another "that didn't work" email from the customer tomorrow. Then it'll be a case of sending him source and build instructions.

Best regards,

- Steve
 | 
Pages: 1
Prev: error with newfftd
Next: add_block to copy a block