Re: Is --enable-64 still experimental?

[Chechu Garguez]

Good morning to everyone

During the last two years we have been trying to prepare octave for 64bit addressing. Our goal is to be able of processing 3d images of cancer and porifera structures, with them 2GBytes array limitation is a big drawback.
With the last version of image package in OF we are still in need of "modifying" the includes from octave for compiling it.

There are any "make check" tests that could be done for verify the integrity of the built image package? It seems to work ,thought.
On previous octave editions; like v3.2; there was also possible to run #make static-binary-dist or  #make dynamic-binary-dist for been able of creating distributable packages with the libraries embedded on it.  There is any documentation or workaround that someone could point us for been able of creating it? GNU/Debian and Red Hat Enterprise Linux are our distributions.

Thankyou so much in advance,
Yours sincerely, Jose

2015-04-16 8:35 GMT+00:00 Chechu Garguez <address@hidden>:

So, for sure it's not a clean way to accomplish it, but for been able of compiling image pkg on 64bit systems we have commented out two of the three alternatives that give ambiguous resolution for gcc:


#octave/oct-inttypes.h


//operator T (void) const { return value (); }

operator double (void) const { return double_value (); }

//operator float (void) const { return float_value (); }

With that, mkoctave is cable of compiling the image 2.4.0 package and we can use its functions.

We had left out "operator double" because it seems to be the option that could generate double memory variables and could contain more data on it.

Are we doing right? Should we use "operator T" insted? Is there any better way to accomplish it? We had read ancient bug reporting and OF-packages status but we had not been able of finding other solutions.

Thankyou so much in advance,

Your sincerly , Jose

2015-03-19 23:15 GMT+01:00 Philip Nienhuis <address@hidden>:

John W. Eaton wrote
> On 03/19/2015 11:34 AM, Orion Poplawski wrote:
>> There's been a request in Fedora
>> (https://bugzilla.redhat.com/show_bug.cgi?id=1082507) to enable 64bit
>> indexing
>> in octave.  I see in 4.0.0-rc1 this is still marked as experimental,
>> which is
>> one of the reasons we're not enabling it.  Is this indeed still the case?
>> It
>> seems like it would be a good thing to enable otherwise.
>
> I think it's still incomplete.  For example, saving and loading data may
> not work properly with 64-bit indexing.

The test suite contains tests for load/save, I presume?

I've been using 64-bit indexing builds for several months now on Windows and
I saw no issues with load/save, neither with Octave-generated nor
Matlab-generated .mat files. I can't vouch for other file types.
Apart from that, __run_test_suite__.m shows a number of failed tests that
pass on 32-bit and --enable-windows-64 MinGW builds.

On Linux I've got a 64-bit idx build (4.1.0+) made with MXE along the lines
you've sketched in a bug report discussion some months ago; I do not use
Octave that much on Linux so I didn't test thoroughly but overall that build
works fine as well. Some tests did fail when run manually.

There are however still issues with several OF packages. See the task
tracker task #13313, the OF package issues mentioned there are equally valid
for Linux.

Philip




--
View this message in context: http://octave.1599824.n4.nabble.com/Is-enable-64-still-experimental-tp4669274p4669280.html
Sent from the Octave - Maintainers mailing list archive at Nabble.com.