Boost logo

Ublas :

Subject: Re: [ublas] bug on compressed_matrix ?
From: Hoang Giang Bui (hgbk2008_at_[hidden])
Date: 2013-01-16 13:44:08


On 01/16/13 19:30, Thomas Klimpel wrote:
> Hoang Giang Bui wrote:
>> Yes, the result is improved but it missed the last element. The matrix
>> has the last row is zero, therefore there should be another entry 141 in
>> the inde1_data vector. In fact, I expected P. index1_data().size() =187
>> and it was correct, just the last entry of P.index1_data() is wrong.
> Well, P.filled1()=186 and P. index1_data().size() is irrelevant, so your expectation was only partly correct. But I guess we both understand what actually happens. So, does the current treatment of empty trailing rows cause problems for you? Do you need to pass the data from matrix P to some external software that could handle empty rows? (Which software is this?) If it does cause problems, then maybe one could think about changing that behavior, but it's not really a bug.
>
> Regards,
> Thomas
> _______________________________________________
> ublas mailing list
> ublas_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/ublas
> Sent to: hgbk2008_at_[hidden]

Yes, I need the correct csr format as standard since we rely on ublas
for computing element stiffness matrix and global stiffness matrix stem
from structural problem. Furthermore, I need to take the transpose of
ublas matrix to feed in a subroutine to compute the restriction operator
in our multilevel solver. This subroutine uses the compressed data for
speed purpose. This behaviour causes segmentation false error in my code
since it read a negative number of entries in the last row.

BR
Giang Bui