redistribution of vectors

[Stephan Kramer]

Barry Smith wrote:
>
> On Feb 10, 2009, at 5:01 PM, Stephan Kramer wrote:
>
>>>
>
>>>>
>>>>
>>>
>>> You can certainly use a VecScatter to redistribute a vector. You
>>> already have the
>>> indices from MatGetSubMatrix() so it should be easy.
>>>
>>> However, you can also use VecLoadIntoVector() and provide an initial
>>> partitioning.
>>>
>>>  Matt
>>>
>> Ah brilliant, I must have overlooked these options (there are so many 
>> in PETSc!). In that case is there also something like 
>> VecLoadIntoVector for MatLoad, i.e. where I specify the distribution 
>> beforehand, otherwise I'll just go for the MatGetSubMatrix() + 
>> VecScatter().
>
>    We don't currently have a MatLoadIntoVector(). The current plan is 
> to actually remove VecLoadIntoVector() and have both VecLoad() and 
> MatLoad() be essentially "load into" operations where one is free to 
> define the layout of the vector and matrix before or not thus getting 
> the full spectrum of possibilities in one clean interface. Sadly we 
> haven't done this reorganization due to having many other things to do.
>>
>>
>> Just a small related question, for MatGetSubMatrix where I ask for 
>> the rows owned by each process but all columns, should I really 
>> assemble an index set ranging over all global indices? This seems a 
>> bit wasteful and non-scalable. Or should I put in the effort and, 
>> using information I have over my halo regions, work out which columns 
>> might have nonzeros for these rows. Or am I missing something and is 
>> there an easier short-cut?
>
>    It is true that it is not totally scalable but then dumping some 
> huge honkying sparse matrix to a file and reading it in is not 
> scalable either. We've never had a problem with this yet for 
> reasonable matrix sizes. Rather than fixing this MatGetSubMatrix() , 
> in your case, having a better MatLoad() merged with 
> MatLoadIntoMatrix() would be the thing I recommend optimizing. But 
> frankly until you get up to 1 billion unknowns what is there now is 
> probably good enough.
>
>    Barry
>
Good point. No, wasn't really planning on using this for anything that 
huge. Thanks a lot for your answers

Cheers
Stephan
>>
>>
>> Thanks a lot
>> Stephan
>>
>>>
>>>> Cheers
>>>> Stephan
>>>>
>>
>
>