Quantcast

Octave on Windows with large arrays ?

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
19 messages Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Octave on Windows with large arrays ?

Julien Bect-2
Hi all,

Is there a place where one can find 64 bits binaries for Octave 4.2.x on
Windows with support for large arrays ?

@++
Julien


Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

PhilipNienhuis
Julien Bect-2 wrote
Hi all,

Is there a place where one can find 64 bits binaries for Octave 4.2.x on
Windows with support for large arrays ?
How "large" would you want to have them?

With the 64-bit binaries on ftp.gnu.org/pub/octave/windows it is easy to create double arrays that are far larger than the 16 GB RAM in my machine and make it start trashing.

IIRC for sparse arrays there was (or is) an old bug about the default (integer) index pointer bit width; 64-bit indexing would be handy there.

Philip
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

Julien Bect-2
Le 15/03/2017 à 16:40, PhilipNienhuis a écrit :
Julien Bect-2 wrote
Hi all,

Is there a place where one can find 64 bits binaries for Octave 4.2.x on 
Windows with support for large arrays ?
How "large" would you want to have them?

With the 64-bit binaries on ftp.gnu.org/pub/octave/windows it is easy to
create double arrays that are far larger than the 16 GB RAM in my machine
and make it start trashing.


You mean that they already have 64 bits indexing ?

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

PhilipNienhuis
Julien Bect-2 wrote
Le 15/03/2017 à 16:40, PhilipNienhuis a écrit :
> Julien Bect-2 wrote
>> Hi all,
>>
>> Is there a place where one can find 64 bits binaries for Octave 4.2.x on
>> Windows with support for large arrays ?
> How "large" would you want to have them?
>
> With the 64-bit binaries onftp.gnu.org/pub/octave/windows  it is easy to
> create double arrays that are far larger than the 16 GB RAM in my machine
> and make it start trashing.


You mean that they already have 64 bits indexing ?
No (AFAIK), but they've been built for 64-bit Windows to take advantage of the 64-bit architecture. That means that the max *process limit* of 2 or 3 GB for all of Octave + its data is gone. IOW Octave is a real 64-bit program.
See here for max array sizes:

https://www.gnu.org/software/octave/doc/interpreter/Compiling-Octave-with-64_002dbit-Indexing.html#Compiling-Octave-with-64_002dbit-Indexing

In some places the information there seems a bit contradictionary to me (again I see 2GB metntioned but that doesn't hold, limits are larger depending on class of data); the section could use some cleaning up.
E.g., in a 64-bit (not 64-bit indexing!) Octave-4.3.0+ I can do this:

==================================================
>> a = zeros (30000);
>> whos
Variables in the current scope:

   Attr Name        Size                     Bytes  Class
   ==== ====        ====                     =====  =====
        a       30000x30000             7200000000  double
        ans         1x1                          8  double

Total is 900000001 elements using 7200000008 bytes
==================================================

Anyway it isn't hard to cross-build 64-bit indexing Octave with mxe. Actually all that is needed is to configure mxe-octave with at least --enable-64 (plus other flags you need).

(JWE made a few changes so that configure flag may have another name now but "configure --help" will show you the right options.)

Philip
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

tmacchant
In reply to this post by Julien Bect-2
----- Original Message -----

>From: Julien Bect 
>To: octave-maintainers 
>Date: 2017/3/16, Thu 01:01
>Subject: Re: Octave on Windows with large arrays ?
>
>
>Le 15/03/2017 à 16:40, PhilipNienhuis a écrit :
>
>Julien Bect-2 wrote
>>Hi all, Is there a place where one can find 64 bits binaries for Octave 4.2.x on
Windows with support for large arrays ?
>>How "large" would you want to have them? With the 64-bit binaries on ftp.gnu.org/pub/octave/windows it is easy to
create double arrays that are far larger than the 16 GB RAM in my machine
and make it start trashing.
>
>You mean that they already have 64 bits indexing ?
>

@Julien

I have built octave-4.2.1 for windows with 64 bits indexing.

http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1-LA-0.7z
size  256,085,305 byte

sha1sum 62e50e31a0b3c2fe9e570958accd7be91e9f9529

Source 
http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1.tar.lz
size 15,826,565
sha1sum 057dbaa30d0ef75e84db30aeda623a2561b0d547

Note that this build does not include pre-build octave-forge packages.

To install octave-forge packages

From octave prompt, execute
cd <install dir>/src
build_packages

you can install octave-forge packages but I have not tested yet.

At this moment, this is temporal and not described in my web site.
http://www.tatsuromatsuoka.com/octave/Eng/Win/


Regards

Tatsuro

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

Julien Bect-2
Le 16/03/2017 à 03:53, Tatsuro MATSUOKA a écrit :

> ----- Original Message -----
>> From: Julien Bect
>> To: octave-maintainers
>> Date: 2017/3/16, Thu 01:01
>> Subject: Re: Octave on Windows with large arrays ?
>>
>>
>> Le 15/03/2017 à 16:40, PhilipNienhuis a écrit :
>>
>> Julien Bect-2 wrote
>>> Hi all, Is there a place where one can find 64 bits binaries for Octave 4.2.x on
> Windows with support for large arrays ?
>>> How "large" would you want to have them? With the 64-bit binaries on ftp.gnu.org/pub/octave/windows it is easy to
> create double arrays that are far larger than the 16 GB RAM in my machine
> and make it start trashing.
>> You mean that they already have 64 bits indexing ?
>>
> @Julien
>
> I have built octave-4.2.1 for windows with 64 bits indexing.
>
> http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1-LA-0.7z
> size  256,085,305 byte
>
> sha1sum 62e50e31a0b3c2fe9e570958accd7be91e9f9529
>
> Source
> http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1.tar.lz
> size 15,826,565
> sha1sum 057dbaa30d0ef75e84db30aeda623a2561b0d547
>
> Note that this build does not include pre-build octave-forge packages.
>
> To install octave-forge packages
>
>  From octave prompt, execute
> cd <install dir>/src
> build_packages
>
> you can install octave-forge packages but I have not tested yet.
>
> At this moment, this is temporal and not described in my web site.
> http://www.tatsuromatsuoka.com/octave/Eng/Win/


Thanks a lot Tatsuro.

I will forward your email to my colleague who needed this.

@++
Julien


Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

tmacchant
----- Original Message -----

>From: Julien Bect 
>To: octave-maintainers
>Date: 2017/3/16, Thu 14:51
>Subject: Re: Octave on Windows with large arrays ?
>
>Le 16/03/2017 à 03:53, Tatsuro MATSUOKA a écrit :
>> @Julien
>>
>> I have built octave-4.2.1 for windows with 64 bits indexing.
>>
>> http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1-LA-0.7z
>> size  256,085,305 byte
>>
>> sha1sum 62e50e31a0b3c2fe9e570958accd7be91e9f9529
>>
>> Source
>> http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1.tar.lz
>> size 15,826,565
>> sha1sum 057dbaa30d0ef75e84db30aeda623a2561b0d547
>>
>> Note that this build does not include pre-build octave-forge packages.
>>
>> To install octave-forge packages
>>
>>  From octave prompt, execute
>> cd <install dir>/src
>> build_packages
>>
>> you can install octave-forge packages but I have not tested yet.
>>
>> At this moment, this is temporal and not described in my web site.
>> http://www.tatsuromatsuoka.com/octave/Eng/Win/
>
>
>Thanks a lot Tatsuro.
>
>I will forward your email to my colleague who needed this.
>
>@++
>Julien
>


@Julien



Excuse me I have forgotten to clean up the directory so that some files and directory for 4.2.0 remains.
Perhaps this will have side effect other than sizes
But I will build again after clean up.

Tatsuro

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

tmacchant




----- Original Message -----

> From: Tatsuro MATSUOKA
> To: Julien Bect  "octave-maintainers
> Cc:
> Date: 2017/3/16, Thu 15:37
> Subject: Re: Octave on Windows with large arrays ?
>
> ----- Original Message -----
>> From: Julien Bect 
>> To: octave-maintainers
>> Date: 2017/3/16, Thu 14:51
>> Subject: Re: Octave on Windows with large arrays ?
>>
>> Le 16/03/2017 à 03:53, Tatsuro MATSUOKA a écrit :
>>>  @Julien
>>>
>>>  I have built octave-4.2.1 for windows with 64 bits indexing.
>>>
>>>  http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1-LA-0.7z
>>>  size  256,085,305 byte
>>>
>>>  sha1sum 62e50e31a0b3c2fe9e570958accd7be91e9f9529
>>>
>>>  Source
>>>  http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1.tar.lz
>>>  size 15,826,565
>>>  sha1sum 057dbaa30d0ef75e84db30aeda623a2561b0d547
>>>
>>>  Note that this build does not include pre-build octave-forge packages.
>>>
>>>  To install octave-forge packages
>>>
>>>   From octave prompt, execute
>>>  cd <install dir>/src
>>>  build_packages
>>>
>>>  you can install octave-forge packages but I have not tested yet.
>>>
>>>  At this moment, this is temporal and not described in my web site.
>>>  http://www.tatsuromatsuoka.com/octave/Eng/Win/
>>
>>
>> Thanks a lot Tatsuro.
>>
>> I will forward your email to my colleague who needed this.
>>
>> @++
>> Julien
>>
>
>
> @Julien
>
>

Correction
> Excuse me I have forgotten to clean up the directory so that some files and
> directory for 4.2.0 remains.
> Perhaps this will have side effect other than sizes


Perhaps this will *not* have side effect other than sizes


> But I am building again after clean up.
>
> Tatsuro


Tatsuro

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

tmacchant
----- Original Message -----

> From: Tatsuro MATSUOKA 
> To: tmacchant Julien Bect "octave-maintainers
> Cc:
> Date: 2017/3/16, Thu 16:11
> Subject: Re: Octave on Windows with large arrays ?
>
>
>
>
>
> ----- Original Message -----
>>  From: Tatsuro MATSUOKA
>>  To: Julien Bect  "octave-maintainers
>>  Cc:
>>  Date: 2017/3/16, Thu 15:37
>>  Subject: Re: Octave on Windows with large arrays ?
>>
>>  ----- Original Message -----
>>>  From: Julien Bect 
>>>  To: octave-maintainers
>>>  Date: 2017/3/16, Thu 14:51
>>>  Subject: Re: Octave on Windows with large arrays ?
>>>
>>>  Le 16/03/2017 à 03:53, Tatsuro MATSUOKA a écrit :
>>>>   @Julien
>>>>
>>>>   I have built octave-4.2.1 for windows with 64 bits indexing.
>>>>
>>>>   http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1-LA-0.7z
>>>>   size  256,085,305 byte
>>>>
>>>>   sha1sum 62e50e31a0b3c2fe9e570958accd7be91e9f9529
>>>>
>>>>   Source
>>>>   http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1.tar.lz
>>>>   size 15,826,565
>>>>   sha1sum 057dbaa30d0ef75e84db30aeda623a2561b0d547
>>>>
>>>>   Note that this build does not include pre-build octave-forge
> packages.
>>>>
>>>>   To install octave-forge packages
>>>>
>>>>    From octave prompt, execute
>>>>   cd <install dir>/src
>>>>   build_packages
>>>>
>>>>   you can install octave-forge packages but I have not tested yet.
>>>>
>>>>   At this moment, this is temporal and not described in my web site.
>>>>   http://www.tatsuromatsuoka.com/octave/Eng/Win/
>>>
>>>
>>>  Thanks a lot Tatsuro.
>>>
>>>  I will forward your email to my colleague who needed this.
>>>
>>>  @++
>>>  Julien
>>>
>>
>>
>>  @Julien
>>
>>
>
> Correction
>>  Excuse me I have forgotten to clean up the directory so that some files and
>
>>  directory for 4.2.0 remains.
>>  Perhaps this will have side effect other than sizes
>
>
> Perhaps this will *not* have side effect other than sizes
>
>
>>  But I am building again after clean up.
>>
>>  Tatsuro
>
>
> Tatsuro


I manually remove 4.2.0 related files and directories from archived directory and re-compress it on windows.

New binary (....LA-0.7z => .....LA-1.7z)
http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1-LA-1.7z
size  209,082,028 byte
sha1sum 5ec10ea6c1fe4023e785d5eaecfa6b289684bddb


Source
http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1.tar.lz
size 15,826,565
sha1sum 057dbaa30d0ef75e84db30aeda623a2561b0d547



I have executed __run_test_suite__ but hangs.

  libinterp\corefcn\lsode.cc-tst .............................. LSODE--  AT T (=R1) AND STEP SIZE H (=R2), THE
       CORRECTOR CONVERGENCE FAILED REPEATEDLY
       OR WITH ABS(H) = HMIN
      In above,  R1 =  0.0000000000000D+00   R2 =  0.3802108489572D-09
 LSODE--  AT T (=R1) AND STEP SIZE H (=R2), THE
       CORRECTOR CONVERGENCE FAILED REPEATEDLY
       OR WITH ABS(H) = HMIN
      In above,  R1 =  0.0000000000000D+00   R2 =  0.6424715644617D-09
 LSODE--  AT T (=R1) AND STEP SIZE H (=R2), THE
       CORRECTOR CONVERGENCE FAILED REPEATEDLY
       OR WITH ABS(H) = HMIN
      In above,  R1 =  0.0000000000000D+00   R2 =  0.4165001171641D-09
 PASS      2/5
                                                                  FAIL    3
  libinterp\corefcn\lu.cc-tst .................................panic: Segmentation fault -- stopping myself...


Mmmmmmmm.

Anyway I will test again against binary I am now re-building after clean up.

Tatsuro

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

Julien Bect-2
Le 16/03/2017 à 09:11, Tatsuro MATSUOKA a écrit :
I have executed __run_test_suite__ but hangs.

  libinterp\corefcn\lsode.cc-tst .............................. LSODE--  AT T (=R1) AND STEP SIZE H (=R2), THE
       CORRECTOR CONVERGENCE FAILED REPEATEDLY
       OR WITH ABS(H) = HMIN
      In above,  R1 =  0.0000000000000D+00   R2 =  0.3802108489572D-09
 LSODE--  AT T (=R1) AND STEP SIZE H (=R2), THE
       CORRECTOR CONVERGENCE FAILED REPEATEDLY
       OR WITH ABS(H) = HMIN
      In above,  R1 =  0.0000000000000D+00   R2 =  0.6424715644617D-09
 LSODE--  AT T (=R1) AND STEP SIZE H (=R2), THE
       CORRECTOR CONVERGENCE FAILED REPEATEDLY
       OR WITH ABS(H) = HMIN
      In above,  R1 =  0.0000000000000D+00   R2 =  0.4165001171641D-09
 PASS      2/5
                                                                  FAIL    3
  libinterp\corefcn\lu.cc-tst .................................panic: Segmentation fault -- stopping myself...


Mmmmmmmm.

Anyway I will test again against binary I am now re-building after clean up.


Hi Tatsuro,

I haven't tried myself, but my colleague confirmed the crash with lsode.

There were also some failed tests with det and eig (see below).  Have you already seen these ?

@++
Julien


***** assert (det ([1, 2; 3, 4]), -2, 10*eps)
!!!!! test failed
ASSERT errors for:  assert (det ([1, 2; 3, 4]),-2,10 * eps)
 
  Location  |  Observed  |  Expected  |  Reason
     ()           2            -2        Abs err 4 exceeds tol 2.2204e-015

***** assert (det (single ([1, 2; 3, 4])), single (-2), 10*eps ("single"))
!!!!! test failed
ASSERT errors for:  assert (det (single ([1, 2; 3, 4])),single (-2),10 * eps ("single"))
 
  Location  |  Observed  |  Expected  |  Reason
     ()           2            -2        Abs err 4 exceeds tol 1.1921e-006

***** test
 A = diag([10^-16, 10^-15]);
 chol_qz_accuracy (A, A, true, false);
!!!!! test failed
ASSERT errors for:  assert (isequal (A * V2, A * V2 * D2),is_chol_accurate)
 
  Location  |  Observed  |  Expected  |  Reason
     ()           1            0         Abs err 1 exceeds tol 0


Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

tmacchant
----- Original Message -----
>From: Julien Bect 
>To: [hidden email]
>Date: 2017/3/17, Fri 06:06
>Subject: Re: Octave on Windows with large arrays ?
>
>
>Le 16/03/2017 à 09:11, Tatsuro MATSUOKA a écrit :
>
>I have executed __run_test_suite__ but hangs.   libinterp\corefcn\lsode.cc-tst .............................. LSODE--  AT T (=R1) AND STEP SIZE H (=R2), THE
       CORRECTOR CONVERGENCE FAILED REPEATEDLY
       OR WITH ABS(H) = HMIN
      In above,  R1 =  0.0000000000000D+00   R2 =  0.3802108489572D-09
 LSODE--  AT T (=R1) AND STEP SIZE H (=R2), THE
       CORRECTOR CONVERGENCE FAILED REPEATEDLY
       OR WITH ABS(H) = HMIN
      In above,  R1 =  0.0000000000000D+00   R2 =  0.6424715644617D-09
 LSODE--  AT T (=R1) AND STEP SIZE H (=R2), THE
       CORRECTOR CONVERGENCE FAILED REPEATEDLY
       OR WITH ABS(H) = HMIN
      In above,  R1 =  0.0000000000000D+00   R2 =  0.4165001171641D-09
 PASS      2/5
                                                                  FAIL    3
  libinterp\corefcn\lu.cc-tst .................................panic: Segmentation fault -- stopping myself... Mmmmmmmm. Anyway I will test again against binary I am now re-building after clean up.
>
>Hi Tatsuro,
>
>I haven't tried myself, but my colleague confirmed the crash with
      lsode.
>
>There were also some failed tests with det and eig (see below). 
      Have you already seen these ?

>
>@++
>Julien
>
>
>________________________________
>
>***** assert (det ([1, 2; 3, 4]), -2, 10*eps)
>!!!!! test failed
>ASSERT errors for:  assert (det ([1, 2; 3, 4]),-2,10 * eps)

>  Location  |  Observed  |  Expected  |  Reason
>     ()           2            -2        Abs err 4 exceeds tol 2.2204e-015
>
>***** assert (det (single ([1, 2; 3, 4])), single (-2), 10*eps ("single"))
>!!!!! test failed
>ASSERT errors for:  assert (det (single ([1, 2; 3, 4])),single (-2),10 * eps ("single"))

>  Location  |  Observed  |  Expected  |  Reason
>     ()           2            -2        Abs err 4 exceeds tol 1.1921e-006
>
>***** test
> A = diag([10^-16, 10^-15]);
> chol_qz_accuracy (A, A, true, false);
>!!!!! test failed
>ASSERT errors for:  assert (isequal (A * V2, A * V2 * D2),is_chol_accurate)

>  Location  |  Observed  |  Expected  |  Reason
>     ()           1            0         Abs err 1 exceeds tol 0
>
>


I have built again octave-4.2.1 with LargeArray after clean up but results are the same.

>I haven't tried myself, but my colleague confirmed the crash with
      lsode.


crash happens on lu but not onlsode

  libinterp\corefcn\lu.cc-tst .................................panic: Segmentation fault -- stopping myself...


on lsode error happens.

Hang on lu is easy to reproduce.

>> A=rand(5);
>> [L,U,P]=lu(A);

panic: Segmentation fault -- stopping myself...
attempting to save variables to 'octave-workspace'...
save to 'octave-workspace' complete

>There were also some failed tests with det and eig (see below). 
      Have you already seen these ?


I have also seen this.

On octave-4.0.3 for windows with Large Arrays, I have not seen such errors on __run_test_suite__.

Unfortunately octave-4.2.x for windows with Large Arrays has many fault, it cannot be usable.

octave-4.0.3 for windows with Large Arrays can be downloaded from 
http://www.tatsuromatsuoka.com/octave/Eng/Win/


Tatsuro

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

tmacchant
----- Original Message -----

> From: Tatsuro MATSUOKA >
> To: Julien Bect ; "octave-maintainers
> Cc:
> Date: 2017/3/17, Fri 08:35
> Subject: Re: Octave on Windows with large arrays ?
>
> ----- Original Message -----
>> From: Julien Bect 
>> To: [hidden email]
>> Date: 2017/3/17, Fri 06:06
>> Subject: Re: Octave on Windows with large arrays ?
>>
>>
>> Le 16/03/2017 à 09:11, Tatsuro MATSUOKA a écrit :
>>
>> I have executed __run_test_suite__ but hangs.  
> libinterp\corefcn\lsode.cc-tst .............................. LSODE--
>  AT T (=R1) AND STEP SIZE H (=R2), THE
>        CORRECTOR CONVERGENCE FAILED REPEATEDLY
>        OR WITH ABS(H) = HMIN
>       In above,  R1 =  0.0000000000000D+00   R2 =  0.3802108489572D-09
>  LSODE--  AT T (=R1) AND STEP SIZE H (=R2), THE
>        CORRECTOR CONVERGENCE FAILED REPEATEDLY
>        OR WITH ABS(H) = HMIN
>       In above,  R1 =  0.0000000000000D+00   R2 =  0.6424715644617D-09
>  LSODE--  AT T (=R1) AND STEP SIZE H (=R2), THE
>        CORRECTOR CONVERGENCE FAILED REPEATEDLY
>        OR WITH ABS(H) = HMIN
>       In above,  R1 =  0.0000000000000D+00   R2 =  0.4165001171641D-09
>  PASS      2/5
>                                                                   FAIL    3
>   libinterp\corefcn\lu.cc-tst .................................panic:
> Segmentation fault -- stopping myself... Mmmmmmmm. Anyway I will test again
> against binary I am now re-building after clean up.
>>
>> Hi Tatsuro,
>>
>> I haven't tried myself, but my colleague confirmed the crash with
>       lsode.
>>
>> There were also some failed tests with det and eig (see below). 
>       Have you already seen these ?
>>
>> @++
>> Julien
>>
>>
>> ________________________________
>>
>> ***** assert (det ([1, 2; 3, 4]), -2, 10*eps)
>> !!!!! test failed
>> ASSERT errors for:  assert (det ([1, 2; 3, 4]),-2,10 * eps)
>>  
>>   Location  |  Observed  |  Expected  |  Reason
>>      ()           2            -2        Abs err 4 exceeds tol 2.2204e-015
>>
>> ***** assert (det (single ([1, 2; 3, 4])), single (-2), 10*eps
> ("single"))
>> !!!!! test failed
>> ASSERT errors for:  assert (det (single ([1, 2; 3, 4])),single (-2),10 * eps
> ("single"))
>>  
>>   Location  |  Observed  |  Expected  |  Reason
>>      ()           2            -2        Abs err 4 exceeds tol 1.1921e-006
>>
>> ***** test
>>  A = diag([10^-16, 10^-15]);
>>  chol_qz_accuracy (A, A, true, false);
>> !!!!! test failed
>> ASSERT errors for:  assert (isequal (A * V2, A * V2 * D2),is_chol_accurate)
>>  
>>   Location  |  Observed  |  Expected  |  Reason
>>      ()           1            0         Abs err 1 exceeds tol 0
>>
>>
>
>
> I have built again octave-4.2.1 with LargeArray after clean up but results are
> the same.
>
>> I haven't tried myself, but my colleague confirmed the crash with
>       lsode.
>
>
> crash happens on lu but not onlsode
>
>   libinterp\corefcn\lu.cc-tst .................................panic:
> Segmentation fault -- stopping myself...
>
>
> on lsode error happens.
>
> Hang on lu is easy to reproduce.
>
>>>  A=rand(5);
>>>  [L,U,P]=lu(A);
>
> panic: Segmentation fault -- stopping myself...
> attempting to save variables to 'octave-workspace'...
> save to 'octave-workspace' complete
>
>> There were also some failed tests with det and eig (see below). 
>       Have you already seen these ?
>
>
> I have also seen this.
>
> On octave-4.0.3 for windows with Large Arrays, I have not seen such errors on
> __run_test_suite__.
>
> Unfortunately octave-4.2.x for windows with Large Arrays has many fault, it
> cannot be usable.
>
> octave-4.0.3 for windows with Large Arrays can be downloaded from 
> http://www.tatsuromatsuoka.com/octave/Eng/Win/
>
 

Perhaps my build failed without --enable-fortran-int64.
(That option was not required for mxe-build before I think )

I am trying rebuild 4.2.1 adding this option.

Tatsuro 

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

John W. Eaton
Administrator
On 03/17/2017 03:13 AM, Tatsuro MATSUOKA wrote:

> Perhaps my build failed without --enable-fortran-int64.
> (That option was not required for mxe-build before I think )
>
> I am trying rebuild 4.2.1 adding this option.

That is not an option for the 4.2.1 configure script.  With 4.2.1,
--enable-64 should cause Octave to use 64-bit integers for array
dimensions and indexing, and to expect the Fortran libraries like BLAS
and LAPACK to also be compiled to use 64-bit integers for indexing.

Was -fdefault-integer-8 added to FFLAGS by the configure script?  That
should also happen automatically.

To know what happened, I'd have to see the output displayed when running
the configure script and maybe the config.log file.

jwe



Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

tmacchant
In reply to this post by Julien Bect-2
--- jwe

> On 03/17/2017 03:13 AM, Tatsuro MATSUOKA wrote:
>
> > Perhaps my build failed without --enrable-fortran-int64.
> > (That option was not required for mxe-build before I think )
> >
> > I am trying rebuild 4.2.1 adding this option.
>
> That is not an option for the 4.2.1 configure script.  With 4.2.1, --enable-64 should cause Octave to use 64-bit integers for array dimensions and indexing, and to expect the Fortran libraries like BLAS and LAPACK to also be compiled to use 64-bit integers for indexing.
>
> Was -fdefault-integer-8 added to FFLAGS by the configure script?  That should also happen automatically.
>
> To know what happened, I'd have to see the output displayed when running the configure script and maybe the config.log file.
>
> jwe
>
>
Excuse me for my inadequate writing.

As written in the thead, --enrable-fortran-int64 is configure option of mxe- octave.

http://octave.1599824.n4.nabble.com/mxe-octave-enable-fortran-int64-is-required-for-64-bits-index-build-td4682442.html

Tatsuro

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

John W. Eaton
Administrator
On 03/17/2017 01:04 PM, [hidden email] wrote:

> As written in the thead, --enrable-fortran-int64 is configure option of mxe- octave.

Sorry, I missed that part.

Yes, because now it is possible to build the development version of
Octave with support for 64-bit indexing and still use 32-bit indexing in
the BLAS and other libraries, we need to control the type of index used
for those libraries separately from whether Octave uses 64-bit indexing.
  Sorry for the confusion, most likely due to not having a separate
mxe-octave branch for building stable Octave.  OTOH, I'm not sure we
really want to add that complexity.

At first I was surprised that Octave was successfully configured with
--enable-64 when the libraries are using 32-bit indexing, but then I
remembered that this is a cross build, so we can't run the program that
is necessary to detect that problem.  So we have to trust that the
options have been selected correctly.  Sorry.

jwe


Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

tmacchant
----- Original Message -----

> From: John W. Eaton 
> To: "tmacchant; Julien Bect ; "octave-maintainers
> Cc:
> Date: 2017/3/18, Sat 04:21
> Subject: Re: Octave on Windows with large arrays ?
>
> On 03/17/2017 01:04 PM, [hidden email] wrote:
>
>>  As written in the thead, --enrable-fortran-int64 is configure option of
> mxe- octave.
>
> Sorry, I missed that part.
>
> Yes, because now it is possible to build the development version of Octave with
> support for 64-bit indexing and still use 32-bit indexing in the BLAS and other
> libraries, we need to control the type of index used for those libraries
> separately from whether Octave uses 64-bit indexing.  Sorry for the confusion,
> most likely due to not having a separate mxe-octave branch for building stable
> Octave.  OTOH, I'm not sure we really want to add that complexity.
>
> At first I was surprised that Octave was successfully configured with
> --enable-64 when the libraries are using 32-bit indexing, but then I remembered
> that this is a cross build, so we can't run the program that is necessary to
> detect that problem.  So we have to trust that the options have been selected
> correctly.  Sorry.
>
> jwe



Sorry for late reply.

I understand the situation.
I rebuild octave-4.2.1 for windows with large arrays with --enable-fortran-int64 for mxe-octave.

This time __run_test_suite__ works with out hang. (3 FAIL exist, tolerance and round off? issue?)

Thanks for detailed and comprehensive explanation.

Tatsuro

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

tmacchant
In reply to this post by tmacchant


----- Original Message -----
> From: Tatsuro MATSUOKA 
> To: tmacchantJulien Bect ; "octave-maintainers
> Cc:
> Date: 2017/3/17, Fri 16:13
> Subject: Re: Octave on Windows with large arrays ?
>


[snip]

> Perhaps my build failed without --enable-fortran-int64.
> (That option was not required for mxe-build before I think )
>
> I am trying rebuild 4.2.1 adding this option.
>
> Tatsuro 


@Julien 

I have re-built octave-4.2.1 for windows with 64 bits indexing adding --enable-fortran-int64.
This time __run_test_suite__ runs to the end.
Test results are included in the directory "run_test" in the package.

Binary archive
http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1-LA-2.7z
size  19,4466,920 byte 
sha1sum 461e6695231e7ab094e13430a8fdd51c58a17663


Source  
http://www.tatsuromatsuoka.com/octave/Eng/Win/octave-4.2.1.tar.lz
size 15,826,565 
sha1sum 057dbaa30d0ef75e84db30aeda623a2561b0d547 

Notes 

1. This build does not include pre-build octave-forge packages. 

To install octave-forge packages 

From octave prompt, execute 
cd <install dir>/src 
build_packages 

I have not tested octave-forge.

2. Default blas is reference blas.


To use openblas,
In bin directory, use explorer 
rename libblas.dll to other name.
copy and paste libopenblas.dll by other name in bin folder
rename copied libopenblas.dll to libblas.dll

Tatsuro


Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

Julien Bect-2
In reply to this post by tmacchant
Le 21/03/2017 à 02:14, Tatsuro MATSUOKA a écrit :

> ----- Original Message -----
>
>> From: John W. Eaton
>> To: "tmacchant; Julien Bect ; "octave-maintainers
>> Cc:
>> Date: 2017/3/18, Sat 04:21
>> Subject: Re: Octave on Windows with large arrays ?
>>
>> On 03/17/2017 01:04 PM, [hidden email] wrote:
>>
>>>   As written in the thead, --enrable-fortran-int64 is configure option of
>> mxe- octave.
>>
>> Sorry, I missed that part.
>>
>> Yes, because now it is possible to build the development version of Octave with
>> support for 64-bit indexing and still use 32-bit indexing in the BLAS and other
>> libraries, we need to control the type of index used for those libraries
>> separately from whether Octave uses 64-bit indexing.  Sorry for the confusion,
>> most likely due to not having a separate mxe-octave branch for building stable
>> Octave.  OTOH, I'm not sure we really want to add that complexity.
>>
>> At first I was surprised that Octave was successfully configured with
>> --enable-64 when the libraries are using 32-bit indexing, but then I remembered
>> that this is a cross build, so we can't run the program that is necessary to
>> detect that problem.  So we have to trust that the options have been selected
>> correctly.  Sorry.
>>
>> jwe
> Sorry for late reply.
>
> I understand the situation.
> I rebuild octave-4.2.1 for windows with large arrays with --enable-fortran-int64 for mxe-octave.
>
> This time __run_test_suite__ works with out hang. (3 FAIL exist, tolerance and round off? issue?)

Hi Tatsuro,

Thank you for your work.  Did you upload this new build on your web site ?

@++
Julien


Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Octave on Windows with large arrays ?

Julien Bect-2
Le 21/03/2017 à 07:13, Julien Bect a écrit :

I understand the situation.
I rebuild octave-4.2.1 for windows with large arrays with --enable-fortran-int64 for mxe-octave.

This time __run_test_suite__ works with out hang. (3 FAIL exist, tolerance and round off? issue?)

Hi Tatsuro,

Thank you for your work.  Did you upload this new build on your web site ?


Just saw the other mail.  Sorry for the noise.  @++ Julien.

Loading...