[Trilinos-Users] [EXTERNAL] Tpetra::MatrixMarket: Write and Read MultiVector

[Andrey Prokopenko]

Hi Sander,

Mark Hoemmen pushed the fix (I checked it on your example and it seems 
to work, test64 and test64_2 are the same now).
It should appear in the public repo some time tomorrow.
If you don't have access to it, I've attached the patch.

-Andrey
On 02/09/2015 07:21 AM, Sander Schaffner wrote:
> Hello again,
>
> the new version helpt in the sense that it reads now the file. But 
> still it behaves like described before:
>
> [1 2 3 ... 63] -> [0 1 64 65 2 3 ...]
> [64 65 66 ... 127] -> [32 33 96 97 34 35 ...]
>
> I wrote a small test program: First I create a 64x2 MV on 32 cores and 
> write into the first MV  [1 2 3 ... 63] and into the second [64 65 66 
> ... 127]. Then I write out the map and the MV. After that I read both 
> files to create a new MV to finally write it out again. Since I 
> changed nothing inbetween and only read and write a valid file i 
> expect two identical files. But this is not the case.
>
> Can someone tell me if I'm doing something wrong in my Code, or is it 
> maybe a bug?
>
> Thanks
>
> Sander
>
> Output first File (test_64_2):
>
> %%MatrixMarket matrix array real general
> 64 2
> 0
> 1
> 64
> 65
> 2
> 3
> 66
> 67
> 4
> 5
> 68
> 69
> 6
> 7
> ...
>
> Second (test_64):
>
> %%MatrixMarket matrix array real general
> 64 2
> 0
> 1
> 32
> 33
> 64
> 65
> 96
> 97
> 2
> 3
> 34
> 35
> 66
> 67
>
> Am 06.02.2015 um 19:04 schrieb Andrey Prokopenko:
>> Hi Sander,
>>
>> I believe that was fixed (see 
>> https://software.sandia.gov/bugzilla/show_bug.cgi?id=6264).
>> Which version of Trilinos do you use?
>>
>> -Andrey
>> On 02/06/2015 02:25 AM, Sander Schaffner wrote:
>>> Hi
>>>
>>> I'm trying to write a MultiVector with 
>>> Tpetra::MatrixMarket::Writer.writeDenseFile. The goal is to read the 
>>> same MultiVector afterwards with 
>>> Tpetra::MatrixMarket::Reader.readDenseFile. But there are some 
>>> reasons why this doesn't work:
>>>
>>> If i generate a MultiVector with two Vectors (dimension 128) where 
>>> the first holds the values 0 and the second 1 I get this output on 
>>> 32 cores:
>>>
>>> %%MatrixMarket matrix array real general
>>> 128 2
>>> 0
>>> 0
>>> 0
>>> 0
>>>
>>> 1
>>> 1
>>> 1
>>> 1
>>>
>>> 0
>>> 0
>>> 0
>>> 0
>>>
>>> 1
>>> 1
>>> 1
>>> 1
>>>
>>> and so on. So I get first the local part of processor 0 for both 
>>> Vectors and so on.
>>>
>>> The reader does not accept this file. It complains about the spaces. 
>>> If I manually get ridd of them it reads the file. But then it 
>>> handles the first half of the values as Vector 1 and the second part 
>>> as Vector 2. So in the End we have the following:
>>>
>>> [0 0 0 0 0 ... 0] -> [0 0 0 0 1 1 1 1 ...]
>>> [1 1 1 1 ... 1] -> [0 0 0 0 1 1 1 1 ...]
>>>
>>> Is there a way of directly read in a MatrixMarket file created by 
>>> the Writer? What other options are there to write/read MultiVectors?
>>>
>>> Any help is appreciated.
>>>
>>> Sander
>>> _______________________________________________
>>> Trilinos-Users mailing list
>>> Trilinos-Users at software.sandia.gov
>>> https://software.sandia.gov/mailman/listinfo/trilinos-users
>>
>
>
>
> Am 06.02.2015 um 19:04 schrieb Andrey Prokopenko:
>> Hi Sander,
>>
>> I believe that was fixed (see 
>> https://software.sandia.gov/bugzilla/show_bug.cgi?id=6264).
>> Which version of Trilinos do you use?
>>
>> -Andrey
>> On 02/06/2015 02:25 AM, Sander Schaffner wrote:
>>> Hi
>>>
>>> I'm trying to write a MultiVector with 
>>> Tpetra::MatrixMarket::Writer.writeDenseFile. The goal is to read the 
>>> same MultiVector afterwards with 
>>> Tpetra::MatrixMarket::Reader.readDenseFile. But there are some 
>>> reasons why this doesn't work:
>>>
>>> If i generate a MultiVector with two Vectors (dimension 128) where 
>>> the first holds the values 0 and the second 1 I get this output on 
>>> 32 cores:
>>>
>>> %%MatrixMarket matrix array real general
>>> 128 2
>>> 0
>>> 0
>>> 0
>>> 0
>>>
>>> 1
>>> 1
>>> 1
>>> 1
>>>
>>> 0
>>> 0
>>> 0
>>> 0
>>>
>>> 1
>>> 1
>>> 1
>>> 1
>>>
>>> and so on. So I get first the local part of processor 0 for both 
>>> Vectors and so on.
>>>
>>> The reader does not accept this file. It complains about the spaces. 
>>> If I manually get ridd of them it reads the file. But then it 
>>> handles the first half of the values as Vector 1 and the second part 
>>> as Vector 2. So in the End we have the following:
>>>
>>> [0 0 0 0 0 ... 0] -> [0 0 0 0 1 1 1 1 ...]
>>> [1 1 1 1 ... 1] -> [0 0 0 0 1 1 1 1 ...]
>>>
>>> Is there a way of directly read in a MatrixMarket file created by 
>>> the Writer? What other options are there to write/read MultiVectors?
>>>
>>> Any help is appreciated.
>>>
>>> Sander
>>> _______________________________________________
>>> Trilinos-Users mailing list
>>> Trilinos-Users at software.sandia.gov
>>> https://software.sandia.gov/mailman/listinfo/trilinos-users
>>
>

-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-TpetraCore-Possible-temporary-fix-for-Bug-6288.patch
Type: text/x-patch
Size: 14031 bytes
Desc: not available
URL: <http://software.sandia.gov/pipermail/trilinos-users/attachments/20150211/bdd51b19/attachment.bin>