I wonder what kind of optimization gcc or llvm will do on that ? I'm not sure they will be able to infer that A(i)(k,l) is a good replacement for accessing the matrix: I mean, I'm not sure if they will find the right overloaded operator. In that case, it would cost a copy of a matrix, which can be pretty expensive.
Do someone know better on that topic ?
Cheers, David
On Sat, Aug 6, 2011 at 14:12, Ralf Philipp Denzer <Ralf.Denzer@web.de> wrote:
Hello Yajima,
just a remark: you can access the component (k,l) of the i-th matrix
A(i)
directly by, e.g.
A(i)(k,l) = 1.0;
So, it's not necessary to introduce 'local' matrices b and/or c. But
it's
not wrong to do so.
Bye
Ralf
On 08/04/2011 05:40 PM, Masanao Yajima wrote:
Hi Ralf
That is exactly what I was looking for.
Your code worked perfectly.
Thank you so much!
Cheers
Yajima
On Thu, Aug 4, 2011 at 12:33 AM, Ralf
Philipp Denzer <Ralf.Denzer@web.de>
wrote:
// set the correct size of matrix A(i)
A(i).resize(b.size1(), b.size2());
A(i) =b;
}
That should work ... (not tested)
You can use also:
std::vector<ublas::matrix<double> > A;
or
std::map<int, ublas::matrix<double> > A;
or other containers ...
Depends what you prefer ...
Bye
Ralf
On 08/04/2011 01:03 AM, Masanao Yajima wrote:
Hi List
Is there away to create a "set" of ublas matrix.
(by set I mean it in the most abstract way)
I would like to create bunch of matrix<double>
and store it in some form of a data structure
then retrieve them by an index and use it as a
matrix<double>.
Here is a vague pseudo code for what I would like
to do
something A;
for( int i=0;i<10;i++){
dmatrix b(10,10);
set values to b;
A(i) =b;
}
//then use it as
for( int i=0;i<10;i++){
dmatrix c = A(i);
do calculation with c
}
Or if you have better way to do something
similar, I'd appreciate your input.