Hello, It has been a while since I followed the mailing list with full attention. I did not find a good implementation of inplace_solve for triangular sparse...
Karl Meerbergen
Karl.Meerbergen@...
May 4, 2004 11:11 am
1504
Hello, I tried the product of a compressed_matrix with a dense vector and obtained quite different timings for row_major and column_major matrices: row_major...
Karl Meerbergen
Karl.Meerbergen@...
May 4, 2004 2:58 pm
1505
... No, it's wrong. There are 3 implementations of inplace_solve in triangular.hpp for each type of triangular matrix: - for a dense matrix: void inplace_solve...
Gunter Winkler
guwi17@...
May 4, 2004 4:18 pm
1506
... The factor seems very large, but it is true, that row_major * column is much faster than column_major * column. Did you use axpy_prod? Please look at the...
Gunter Winkler
guwi17@...
May 4, 2004 4:25 pm
1507
Hi Joerg, I cannot see this change in boost CVS yet. Have you checked it in? It's been a while since this was discussed here. Just a reminder that this...
... Thanks. The timings used prod. I now tried axpy_prod: this runs in 0.05 seconds for column_major. I have another question: which is the best way to define...
Karl Meerbergen
Karl.Meerbergen@...
May 5, 2004 8:08 am
1509
Hello, there have been some tries to update the documentation. I'd like to start another one. 1. I suggest adding a link to Effektive uBlas to the...
Gunter Winkler
guwi17@...
May 5, 2004 1:35 pm
1510
Hello Gunter, ... Good idea! I've added a extended version of you patch with the following additional section. <h2>Further information</h2> <h3>More...
Hello Karl, ... This is true. It is very hard for uBLAS to deal with symmetric sparse matrices efficiently. However if speed is not too important the...
Hi, I've been playing around a bit with gemm and so on, but bindings/blas3.hpp is full of comments // ! CAUTION this function assumes that all matrices...
Ian McCulloch
ianmcc@...
May 5, 2004 10:23 pm
1514
Hello, When I am using axpy_prod for sparse matrices, in one particular case v takes the value [4](-5.96311e-19,-2.14808e-18,3.36441e-18,-1.70254e-19) and cv...
Karl Meerbergen
Karl.Meerbergen@...
May 6, 2004 7:40 am
1515
... There is actually a sharper error bound: eps * norm_1( abs(v) + prod( abs(a), abs(w) ) where abs produces a vector_expression or matrix_expression Karl...
Karl Meerbergen
Karl.Meerbergen@...
May 6, 2004 7:46 am
1516
... This is a know problem. Please put this line before any ublas includes in your main program. #define BOOST_UBLAS_TYPE_CHECK_MIN real_type(1.0e-0) mfg ...
Gunter Winkler
guwi17@...
May 6, 2004 8:11 am
1517
... No. Original BLAS is written in Fortran, and native Fortran matrices are column major. But you can use ATLAS bindings -- ATLAS is BLAS rewritten in C and ...
... Ok, but you can still use the column major functions with row-major matrices, but you have to be careful about where the transposes appear. ie. if C=A*B...
Ian McCulloch
ianmcc@...
May 6, 2004 11:44 am
1519
Hello Gunter, Hello Karl, ... This is not a very nice problem. I think we should do something about it. The equals checks trigger in two circumstances I know...
... For my example, adding the #define removed the problem. In theory the upper bounds using the abs( ) function that I mentioned in another e-mail should...
Karl Meerbergen
Karl.Meerbergen@...
May 6, 2004 2:05 pm
1521
... I ran into the same problem while solving systems of linear equations iteratively: means compute r = A x - b -> update x -> recompute. When x is quite...
Gunter Winkler
guwi17@...
May 6, 2004 3:11 pm
1522
... I see. ... Sorry, I did not express myself well. I was describing the general case. Most of the equals check in uBLAS are position to pickup bugs in the...
Hello Gunter, Hello Karl, I have checked in a change which makes controlling the type check behaviour much easier. The BOOST_UBLAS_TYPE_CHECK macro is now...
Hello, I made some modifications to Mickaels proposition following the theory in Nick Higham's book. With these, the test does not fail, as it should not. Karl...
Karl Meerbergen
Karl.Meerbergen@...
May 7, 2004 2:43 pm
1525
Hello Karl, ... Do you have a reference for this. ... Looks like we are getting somewhere, thanks. ... I find the (1- size * epsilon) term rather curious. The...
... Hi Michael, The reference is Nicholas J. Higham. Accuracy and Stability of Numerical Algorithms. SIAM, Philadelphia. 1996, see www.siam.org This is an...
Karl Meerbergen
Karl.Meerbergen@...
May 10, 2004 7:00 am
1527
... For my example the new code works. I think the one-norm suggested by Karl is a good way to find all algorithmic problems of ublas. mfg Gunter...
Gunter Winkler
guwi17@...
May 10, 2004 9:18 am
1528
Hello, I have the following example that does not run as I would have expected: it runs forever. NDEBUG is defined. #include...
Karl Meerbergen
Karl.Meerbergen@...
May 10, 2004 1:59 pm
1529
... This program aborts in matrix_assign.hpp line 830 because the debug code tries to allocate a dense matrix of size (100000x100000) which is currently not ...
Gunter Winkler
guwi17@...
May 10, 2004 4:41 pm
1530
Hi, In the docs, matrix::clear is specified as "Clears the matrix.", and it isn't clear from that what exactly it does. The std:: container classes have a...
Ian McCulloch
ianmcc@...
May 11, 2004 11:50 am
1531
Hi, It appears that inner_prod(matrix, matrix) is not defined. Is there any reason why it should not? I was going to send in a sample definition, but then I...
Ian McCulloch
ianmcc@...
May 11, 2004 12:00 pm
1532
... Good point. In addition, what does it mean for sparse matrices? Are the arrays resized ? Karl...
Having problems with message search?
Fill out this form to ensure your group is one of the first to be migrated to the new message search system.
