Hello, the attached test program does not compile with current ublas cvs (using gcc 3.3.1, see compile.log). It does compile with current boost cvs. It does...
Hello Georg, Thanks for the report I can reproduce it on multiple compilers. I don't have a fix yet but I know when things wrong. The bug was introduced with...
When a coordinate matrix is resized, filled_ is set to zero. Although the data are copied in the arrays index1_data_, index2_data_ and value_data_, they are no...
Karl Meerbergen
Karl.Meerbergen@...
Jun 4, 2004 9:07 am
1576
... This looks like a bug. Resizing sparse matrices is a difficult problem. Thus the strategie was to clear the matrix when it is resized. For coordinate ...
Gunter Winkler
guwi17@...
Jun 4, 2004 10:10 am
1577
... In the case where the matrix is cleared anyway (e.g. compressed_matrix) I suppose it does not make sense to have the preserve argument. At least it is very...
Karl Meerbergen
Karl.Meerbergen@...
Jun 4, 2004 1:38 pm
1578
... Looks like multiple problems! The documentation for resize states that 'The content of the XX_matrix is preserved', in the case of all three gappy types....
Gunter, Karl, ... Looking at coordinate_matrix does not seem to correctly implement a preserving resizing in all cases. The existing preserve will only work if...
Can anyone recommend a fast way to do the following computation? I have three matrices, D, D^(-1/2), and W. D and D^(-1/2) are diagonal, and W is sparse and...
Steve, I'm not quite sure what you mean by "fast" in "there isn't fast support for sparse-sparse matrix multiplication in uBLAS". GluCat ( http://glucat.sf.net...
Steve, ... Is Paul's expression is the one we really want to compute? Am I missing something, but if W is symmetric and so is d then we can rearrange to = I -...
Michael, Reply below. Best regards ... This would always work if diagonal matrices always commute with symmetric matrices. Unfortunately this is not the case....
Paul, ... Oh blast! Serves me right for doing a back of the envelope calculation. I used the standard symmetric matrix rule that A' * B' = (B * A)' I missed...
(Sorry my reply to this thread took so long) ... And I should have answered already a week ago: In case of 'gemm' it's easy to play with the transpose. With...
... I don't think anybody will disagree. But I'm afraid you will not hear an OK from Joerg I have not seen Joerg for a long time on the ml. I think Michael...
Hi Toon, ... I bypassed the mailing list and talked with Joerg by phone a couple of weeks ago. He still has lost of ideas regarding the future of uBLAS. And...
... No, it should work for non-square matrices (modulo bugs ;). If C is row-major, then you can rewrite the product: C^T = (A * B)^T = B^T * A^T and now the...
Ian McCulloch
ianmcc@...
Jun 10, 2004 1:38 pm
1589
Hi, Sorry, I have A correction: ... [...] ... Sorry, that doesn't make sense: I misread 'square' for 'rectangular' there. What dies it mean to have a...
Ian McCulloch
ianmcc@...
Jun 11, 2004 2:10 pm
1590
... I never mentioned 'non-rectangular' IIRC? ... Of course, I overlooked that option. ... Indeed....
To clarify, the example I posted earlier compiles but the type 'result_type' is not as expected. It is of type boost::mpl::void_ and thus declaring a variable...
1.) That means, changing in traits.hpp to #ifndef BOOST_NO_TEMPLATE_PARTIAL_SPECIALIZATION template<class T1, class T2> struct promote_traits { typedef...
Hi, ... What about char and unsigned char ? Here is a small problem: the limit for the typelists in mpl is 10 as default value (do not know where to change ...
... Yes but it's kind of natural to use short's while char's are less natural. Another option would be not to use MPL but implement the promote_traits as...
Currently UBLAS_TYPE_CHECK will automatically be defined when compiling in debug mode. However when UBLAS_TYPE_CHECK is defined, ublas is not portable to...
Hello, ... OK, I just read a little bit in the standard and I think the solution adding the "short" and "unsigned short" is nearly correct. "4.5 Integral...
Hi Toon, ... Looks good. It is probably worth mentioning that the current uBLAS CVS (not Boost CVS) already has such a mechanism. Since your aim was to get...
Hi Patrick, Hi Toon, I just read through the whole of the thread "Compilation problem with VC 7.1]" with interest. ... Agreed. The existing mechanism using an...
Hi all, I have recently been experimenting with getting uBLAS to play nicely with some legacy code that uses simple C-arrays to represent vectors and matrices,...
