Octave 3.0.1 Scientific Linux 5.1 (RHAT) build failure

[Labitt, Bruce]

Hi,

I am trying to get octave 3.0.1 to build on my machine.  I have
Scientific Linux 5.1 - 64 installed as the OS.  I'd like everything to
build to 64 bits so I can take advantage of the increased memory space.
I need to calculate out some large ffts (128M point).

I have gcc 4.1.2 on my machine.  I compiled and tested fftw3.2alpha3.
It is installed in /usr/local/bin

I ran the octave configure with the following options:

$ ./configure --with-f77=gfortran --with-fftw --enable-64
--enable-shared

Configure exited with an error code = 0.

I then ran make.  The last part of the message string was "Octave
successfully built".  Now choose from the following:

        ./run-octave
        make check
        make install

I executed ./run-octave.  I got ... : eror while loading shared
libraries: libfftw3.so.3: cannot open shared object file:  No such file
or directory.

If this is insufficient information to get help I can send config.log
and makeoctave.log to the list.  I did not know which sections to snip
and which to send.

If someone could provide some pointers and or assistance I'd greatly
appreciate it.

Thanks,
Bruce

_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Dmitri A. Sergatskov]

On Thu, May 8, 2008 at 9:27 AM, Labitt, Bruce
<labittb1@...> wrote:
>

>  I executed ./run-octave.  I got ... : eror while loading shared
>  libraries: libfftw3.so.3: cannot open shared object file:  No such file
>  or directory.
>

Most likely you did not install libfft correctly. The dynamic loader ld cannot
find it; /usr/local/lib is not in its search path.

Dmitri.
--
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Jordi Gutiérrez Hermoso-2]

On 08/05/2008, Labitt, Bruce <labittb1@...> wrote:
>  I'd like everything to
>  build to 64 bits so I can take advantage of the increased memory space.
>  I need to calculate out some large ffts (128M point).

I'm still not exactly clear on what the benefits of 64bit environments
are, and for us  "early adopters", it seems that there are are more
drawbacks. I've seen at least one software package segfault due to
assumptions of 32bit pointers, and I am currently suspecting that one
more package is suffering the same.

You do seem to get more precision on some data types (long doubles are
16 instead of 12 bytes), but I'm pretty sure you don't need 64bits for
doing 128-meg ffts. Only if you need to address more than four gigs of
memory do you really need the 64 bits.

>  I executed ./run-octave.  I got ... : eror while loading shared
>  libraries: libfftw3.so.3: cannot open shared object file:  No such file
>  or directory.

As Dima said, there probably is a minor error in the way that libfftw
is installed. Where is it installed? Can you link to it otherwise? Try
a simple C or C++ program that links to libfftw and see if you can do
it.

- Jordi G. H.
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Labitt, Bruce]

On Thu, May 8, 2008 at 9:27 AM, Labitt, Bruce
<labittb1@...> wrote:
>

>  I executed ./run-octave.  I got ... : eror while loading shared
>  libraries: libfftw3.so.3: cannot open shared object file:  No such
file
>  or directory.
>

Most likely you did not install libfft correctly. The dynamic loader ld
cannot
find it; /usr/local/lib is not in its search path.

Dmitri.
--
[Labitt, Bruce]

Is there a way to add it to the search path?

Bruce

_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Dmitri A. Sergatskov]

On Thu, May 8, 2008 at 2:13 PM, Labitt, Bruce
<labittb1@...> wrote:

> On Thu, May 8, 2008 at 9:27 AM, Labitt, Bruce
> <labittb1@...> wrote:
>>
>
>>  I executed ./run-octave.  I got ... : eror while loading shared
>>  libraries: libfftw3.so.3: cannot open shared object file:  No such
> file
>>  or directory.
>>
>
> Most likely you did not install libfft correctly. The dynamic loader ld
> cannot
> find it; /usr/local/lib is not in its search path.
>
> Dmitri.
> --
> [Labitt, Bruce]
>
> Is there a way to add it to the search path?

Create a file in /etc/ld.so.conf.d/

with name say "local.conf"
in this file type in:

/usr/local/lib

Save and exit.
Run (as root) "ldconfig -v"
Make sure you get libfft3 in the output of this command.

>
> Bruce
>

Regards,

Dmitri.
--
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[David Bateman-3]

Jordi Gutiérrez Hermoso wrote:
> I'm still not exactly clear on what the benefits of 64bit environments
> are, and for us  "early adopters", it seems that there are are more
> drawbacks. I've seen at least one software package segfault due to
> assumptions of 32bit pointers, and I am currently suspecting that one
> more package is suffering the same.

If no one runs on a 64-bit platform none of the bugs of 64-bit systems
will be found. That being said I have the misfortune of currently
running on a 64bit system and yes there are lots of bugs, and I only
just got a version of java that worked, etc etc etc. I haven't tried a
64-bit compile of Octave yet though :-)

D.
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Labitt, Bruce]

-----Original Message-----
From: Dmitri A. Sergatskov [mailto:dasergatskov@...]
Sent: Thursday, May 08, 2008 4:06 PM
To: Labitt, Bruce
Cc: help-octave@...
Subject: Re: Octave 3.0.1 Scientific Linux 5.1 (RHAT) build failure

On Thu, May 8, 2008 at 2:13 PM, Labitt, Bruce
<labittb1@...> wrote:

> On Thu, May 8, 2008 at 9:27 AM, Labitt, Bruce
> <labittb1@...> wrote:
>>
>
>>  I executed ./run-octave.  I got ... : eror while loading shared
>>  libraries: libfftw3.so.3: cannot open shared object file:  No such
> file
>>  or directory.
>>
>
> Most likely you did not install libfft correctly. The dynamic loader

ld
> cannot
> find it; /usr/local/lib is not in its search path.
>
> Dmitri.
> --
> [Labitt, Bruce]
>
> Is there a way to add it to the search path?

Create a file in /etc/ld.so.conf.d/

with name say "local.conf"
in this file type in:

/usr/local/lib

Save and exit.
Run (as root) "ldconfig -v"
Make sure you get libfft3 in the output of this command.

>
> Bruce
>

Regards,

Dmitri.
--
[Labitt, Bruce]

Thanks.  I found that fftw failed make check, so I recompiled.  Probably
had some incompatible options.  Now it passes make check.  I also added
the file local.conf as indicated above.  I now get libfft3 in the output
list.  I am recompiling octave now.  Hopefully everything will pass.

Bruce

_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Francesco Potorti`-2]

>If no one runs on a 64-bit platform none of the bugs of 64-bit systems
>will be found.

I am running a dual core amd64 with Debian testing, and I see no
problems with Octave.  Isn't that built with 64 bits?

--
Francesco Potortì (ricercatore)        Voice: [phone] (op.2111)
ISTI - Area della ricerca CNR          Fax:   [phone]
via G. Moruzzi 1, I-56124 Pisa         Email: Potorti@...
Web: http://fly.isti.cnr.it/           Key:   fly.isti.cnr.it/public.key
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[John W. Eaton]

On  9-May-2008, Francesco Potorti` wrote:

| >If no one runs on a 64-bit platform none of the bugs of 64-bit systems
| >will be found.
|
| I am running a dual core amd64 with Debian testing, and I see no
| problems with Octave.  Isn't that built with 64 bits?

Unless you built it with the --enable-64 configure option, then no,
not really.  With --enable-64, array sizes and indices are represented
as 64-bit integers.  Without it, they are still just 32-bit integers.

The hard part about making Octave work properly as a 64-bit
application is interfacing with other libraries and Fortran code.
Also, if you need the larger array size limits allowed by the 64-bit
version, you will likely need large file support, and so far no one
has done the work for that.

BTW, the limit is 2^31 (2GB, or approximately 268x10^6 double
precision elements, which is approximately a 16,000x16,000 matrix) not
2^32 (4GB) because we have to used signed integers for array sizes and
indices because of our dependence on Fortran code like the BLAS and
LAPACK.  However, on your 64-bit system with a "32-bit" Octave, the
limit of any individual array is 2GB, but you can have more than one.
So unless you really need to work with individual arrays that are
larger than 2GB, you probably don't need to compile Octave with
--enable-64.

jwe
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Jordi Gutiérrez Hermoso-2]

On 08/05/2008, David Bateman <adb014@...> wrote:
> If no one runs on a 64-bit platform none of the bugs of 64-bit systems
>  will be found. That being said I have the misfortune of currently
>  running on a 64bit system and yes there are lots of bugs, and I only
>  just got a version of java that worked, etc etc etc.

Right, right, I don't mind helping smoothing out bugs whenever I can.
I am just wondering what the benefit will be once this goal is
somewhat achieved. We'll be able to address over four gigs of memory.
What else?

- Jordi G. H.
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Jordi Gutiérrez Hermoso-2]

On 08/05/2008, Francesco Potorti` <Potorti@...> wrote:
> >If no one runs on a 64-bit platform none of the bugs of 64-bit systems
>  >will be found.
>
>
> I am running a dual core amd64 with Debian testing, and I see no
>  problems with Octave.  Isn't that built with 64 bits?

Did you compile with --enable-64? Octave has some experimental 64bit
features that aren't enabled by default.

The other segfaults I was mentioning weren't in regards to Octave, but
to other things (and I'm suspecting that Octaviz is having similar
problems, but I'm still waiting for confirmation that the segfault I'm
seeing is a 64bit problem).

- Jordi G. H.
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Sergei Steshenko-2]

--- "John W. Eaton" <jwe@...> wrote:

> On  9-May-2008, Francesco Potorti` wrote:
>
> | >If no one runs on a 64-bit platform none of the bugs of 64-bit systems
> | >will be found.
> |
> | I am running a dual core amd64 with Debian testing, and I see no
> | problems with Octave.  Isn't that built with 64 bits?
>
> Unless you built it with the --enable-64 configure option, then no,
> not really.  With --enable-64, array sizes and indices are represented
> as 64-bit integers.  Without it, they are still just 32-bit integers.
>

[snip]

Does --enable-64 also somehow enable long doubles ?

I.e. does it help to get better numeric accuracy ?

Thanks,
  Sergei.


Applications From Scratch: http://appsfromscratch.berlios.de/


      ____________________________________________________________________________________
Be a better friend, newshound, and
know-it-all with Yahoo! Mobile.  Try it now.  http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Przemek Klosowski]

Even if you're running on AMD64, you could still be running 32-bit kernel
and then all your binaries would have to be 32-bit as well, I believe.
'uname -m' will tell you about your kernel (x86_64 or i686).

On a 64-bit Linux, you could still run either 32 or 64bit octave;
'octave -v' will tell you which one you have.
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[John W. Eaton]

On  9-May-2008, Sergei Steshenko wrote:

| Does --enable-64 also somehow enable long doubles ?
|
| I.e. does it help to get better numeric accuracy ?

No.

jwe
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Sergei Steshenko-2]

--- "John W. Eaton" <jwe@...> wrote:

> On  9-May-2008, Sergei Steshenko wrote:
>
> | Does --enable-64 also somehow enable long doubles ?
> |
> | I.e. does it help to get better numeric accuracy ?
>
> No.
>
> jwe
>

Is there a way to enable long doubles ?

Thanks,
  Sergei.

Applications From Scratch: http://appsfromscratch.berlios.de/


      ____________________________________________________________________________________
Be a better friend, newshound, and
know-it-all with Yahoo! Mobile.  Try it now.  http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Labitt, Bruce]

> Most likely you did not install libfft correctly. The dynamic loader
ld
> cannot
> find it; /usr/local/lib is not in its search path.
>
> Dmitri.
> --
> [Labitt, Bruce]
>
> Is there a way to add it to the search path?

Create a file in /etc/ld.so.conf.d/

with name say "local.conf"
in this file type in:

/usr/local/lib

Save and exit.
Run (as root) "ldconfig -v"
Make sure you get libfft3 in the output of this command.

>
> Bruce
>

Regards,

Dmitri.
--
[Labitt, Bruce]

Thanks. That helped.  I'm getting a little closer.  When I do a make
check I pass all the tests except rand.cc where I fail 14 tests, plus a
few tests where I know I have not installed stuff like qhull.  However,
at the test linear-algebra/condest.m I get a panic: Segmentation fault.
Any clues as where to look?  Flush/repeat?

I tried running octave nonetheless and could not seem to plot anything
with more than 100 elements.  Whenever I did a plot or use grid I would
get 4 lines

line xxxxxxx: undefined variable: in
line xxxxxxx: undefined variable: in
line xxxxxxx: undefined variable: in
line xxxxxxx: undefined variable: lw

I'd be greatful for any help.

Thanks,
Bruce

_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Dmitri A. Sergatskov]

On Fri, May 9, 2008 at 4:58 PM, Labitt, Bruce
<labittb1@...> wrote:
> [Labitt, Bruce]
>
> Thanks. That helped.  I'm getting a little closer.  When I do a make
> check I pass all the tests except rand.cc where I fail 14 tests, plus a
> few tests where I know I have not installed stuff like qhull.  However,
> at the test linear-algebra/condest.m I get a panic: Segmentation fault.
> Any clues as where to look?  Flush/repeat?

condest code relies on blas/lapack libraries and those are probably not
compiled with 64-bit integers for matrix indexing...

>
> I tried running octave nonetheless and could not seem to plot anything
> with more than 100 elements.  Whenever I did a plot or use grid I would
> get 4 lines
>
> line xxxxxxx: undefined variable: in
> line xxxxxxx: undefined variable: in
> line xxxxxxx: undefined variable: in
> line xxxxxxx: undefined variable: lw
>
> I'd be greatful for any help.
>

what is your version of gnuplot?
Do you have mouse support enabled (i.e.
can you zoom with mouse when you do plot
something) ?

> Thanks,
> Bruce
>

Dmitri.
--
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Labitt, Bruce]

-----Original Message-----
From: Dmitri A. Sergatskov [mailto:dasergatskov@...]
Sent: Friday, May 09, 2008 7:04 PM
To: Labitt, Bruce
Cc: help-octave@...
Subject: Re: Octave 3.0.1 Scientific Linux 5.1 (RHAT) build failure

On Fri, May 9, 2008 at 4:58 PM, Labitt, Bruce
<labittb1@...> wrote:
> [Labitt, Bruce]
>
> Thanks. That helped.  I'm getting a little closer.  When I do a make
> check I pass all the tests except rand.cc where I fail 14 tests, plus
a
> few tests where I know I have not installed stuff like qhull.
However,
> at the test linear-algebra/condest.m I get a panic: Segmentation
fault.
> Any clues as where to look?  Flush/repeat?

condest code relies on blas/lapack libraries and those are probably not
compiled with 64-bit integers for matrix indexing...

>
> I tried running octave nonetheless and could not seem to plot anything
> with more than 100 elements.  Whenever I did a plot or use grid I
would
> get 4 lines
>
> line xxxxxxx: undefined variable: in
> line xxxxxxx: undefined variable: in
> line xxxxxxx: undefined variable: in
> line xxxxxxx: undefined variable: lw
>
> I'd be greatful for any help.
>

what is your version of gnuplot?
Do you have mouse support enabled (i.e.
can you zoom with mouse when you do plot
something) ?

> Thanks,
> Bruce
>

Dmitri.
--
[Labitt, Bruce]

I think I may not have compiled ATLAS and LAPACK with 64 bit.  I'll
check on that.  My version of gcc is 4.1.2 (Red Hat 4.1.2-14)

GNUPLOT Version 4.0 patchlevel 0 was the binary I installed.  I can
uninstall it and compile 4.2.3 if it will matter...

Thanks,
Bruce

_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave