info
Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
Re: Floating point exception when taking a reciprocal
21 views
Skip to first unread message
Niels Walet
Jun 21, 2010, 2:15:13 AM6/21/10
to
Leo Alekseyev wrote:
> I am running into a strange bug where taking reciprocals of a list of
> floating point numbers crashes the kernel with a floating point
> exception. This is on a Linux x86_64 system, Ubuntu 9.10, running
> Mathematica 7.0.1.0
>
> Something like
> foo=Array[1.0&,1000]; 1/foo
> or
> foo=RandomReal/@Range[100]; 1/foo
> gives me the crash
> Replacing floats with integers makes things work fine, also, for small
> enough lists (order of 10 elements) everything works.
>
> Same version of Mathematica on a 32-bit linux system or on Windows
> doesn't exhibit this behavior.
>
> I am rather mystified by what's going on; my hardware/OS is pretty
> standard, so if the bug were this obvious it would have long been
> known by now... But if anyone else has run into similar behavior
> and/or knows of some sort of a fix, please let me know.
>
> Thanks,
>
> --Leo
>
>
I have seen similar problems (with Scientific linux 5 64 bits on an
intel i5 processor). The same code works on the identical OS with an
older Core2 processor. I have the feeling it is due to the issue with
the intel math library for i7
(http://support.wolfram.com/technotes/IntelCorei7.html), but am not
sure, since I can't directly talk to wolfram (site license).
> I am running into a strange bug where taking reciprocals of a list of
> floating point numbers crashes the kernel with a floating point
> exception. This is on a Linux x86_64 system, Ubuntu 9.10, running
> Mathematica 7.0.1.0
>
> Something like
> foo=Array[1.0&,1000]; 1/foo
> or
> foo=RandomReal/@Range[100]; 1/foo
> gives me the crash
> Replacing floats with integers makes things work fine, also, for small
> enough lists (order of 10 elements) everything works.
>
> Same version of Mathematica on a 32-bit linux system or on Windows
> doesn't exhibit this behavior.
>
> I am rather mystified by what's going on; my hardware/OS is pretty
> standard, so if the bug were this obvious it would have long been
> known by now... But if anyone else has run into similar behavior
> and/or knows of some sort of a fix, please let me know.
>
> Thanks,
>
> --Leo
>
>
I have seen similar problems (with Scientific linux 5 64 bits on an
intel i5 processor). The same code works on the identical OS with an
older Core2 processor. I have the feeling it is due to the issue with
the intel math library for i7
(http://support.wolfram.com/technotes/IntelCorei7.html), but am not
sure, since I can't directly talk to wolfram (site license).
--
Prof. Niels R. Walet Phone: +44(0)1613063693
School of Physics and Astronomy Fax: +44(0)1613064303
The University of Manchester Mobile: +44(0)7905438934
Manchester, M60 1QD, UK
Niels...@manchester.ac.uk http://theory.phy.umist.ac.uk/~mccsnrw
David Reiss
Jun 22, 2010, 7:03:16 AM6/22/10
to
For Mathematica version 7 and 7.01 there is a bug in the Intel Core i7
(Nehalem) based CPU processor architecture as it is used on Linux
which requires a special patched version of Mathematica to be
installed. you should contact Woflram support for the patched
version.
(Nehalem) based CPU processor architecture as it is used on Linux
which requires a special patched version of Mathematica to be
installed. you should contact Woflram support for the patched
version.
--David
> Prof. Niels R. Walet Phone: +44=
(0)1613063693
> School of Physics and Astronomy Fax: +44(0)1613064=
303
> The University of Manchester Mobile: +44(0)7905438=
934
> Manchester, M60 1QD, UK
> Niels.Wa...@manchester.ac.ukhttp://theory.phy.umist.ac.uk/~mccsnrw
0 new messages