7.0 complex type problem

7.0 complex type problem

While trying to compile lapack 3.0 (from netlib) with
newest ifc version I've encountered problems with all
complex type routines (both complex and complex*16) -
the library compiles OK but then all the tests give me
all sort of wrong output. With previous versions (5.0
and 6.0) I had only minor problems and the libraries are
usable quite OK - any ideas, suggestions? Is that bug
or feature? Maybe improving compatibility with F95 causes
some compatibility problems with F77 in which the library is written?

Romek

8 post / 0 nuovi
Ultimo contenuto
Per informazioni complete sulle ottimizzazioni del compilatore, consultare l'Avviso sull'ottimizzazione

There's not enough information here to go on. I doubt that the problem has anything to do with F95 support.

Please come up with a small but complete example (a sample netlib routine and your own sample code) and submit a problem report to Premier Support - we'll be glad to take a look.

Steve

Steve - Intel Developer Support

Yeah, I knew that it won't be enough info to answer but proper answering involves
exchange of some more data than just few lines of text and so I wanted to get in touch first. The story is as follows:
starting with ver. 5.0 on RedHat 6.2 we have switched here from Portland Group compiler environment to the unsupported Intel ifc compiler and due to the incompatibilities with lapack/blas compiled with gcc we have had to compile our own ifc versions of both (I still don't know how much fex MKL can replace blas so to be on the safe side we compiled both :) - anyway, at that time I only had to change few things in the makefiles and one in sources to compile it with almost all tests passed OK;
later on, on switching to the RH 7.3 we have also upgraded ifc to ver. 6.0 and recompiled lapack/blas with even less effort (no sources meddling at all, only few small changes in makefiles);
now, still on RH 7.3, I wanted to switch to the 7.0 ifc and wanted to recompile both libraries - as I said earlier they do compile OK but all the complex tests produce
much more diagnostic output than it has happened for earlier versions and that it's for the real/DP counterparts.
To be more informative I'd like to send samples of those files but copying them here won't do any good and so can You give me some normal e-mail address to send them as attachments? Or should I try to attach them here?

Romek

Premier Support is how you should report problems.

Steve

Steve - Intel Developer Support

There is a small problem with sending anything to premier :( it asks about user ID and advices to ask my local Intel reseller for one in case I don't have it but as we didn't purchase the compiler (we use Linux-unsupported version) we don't have anybody to ask about such thing! Is there any other way to contact them?

Romek

When you downloaded the compiler, it gave you the option of registering for support at that time. The e-mail you got for the download should have given you registration details, but you can go to https://shale.intel.com/registrationcenter/register.asp to register now. You'll need your serial number.

Note that, while you can submit problem reports, you may not receive a personal reply since you have the unsupported version.

Steve

Steve - Intel Developer Support

I believe that ifc 7.0 has made a choice favoring (by default) speed of complex division over correctness for the full range of operands. In case this choice may be found unsatisfactory for some customers, a change may be in the works. Likewise, better documentation should be on the way. If you aren't sure whether this is the issue which concerns you, setting up your intel premier support account and providing test cases would be a good idea.

Thanks for all the answers and help, indeed I did get some info about premiere support so I'll try it but I guess the last message might be an answer and most probably I should just wait for a while :)

Romek

Lascia un commento

Eseguire l'accesso per aggiungere un commento. Non siete membri? Iscriviti oggi