Re: Project progress

[c.]

On 6 Oct 2013, at 20:28, marco Vassallo <address@hidden> wrote:

> At the moment I'm doing something very close (I hope) to what you suggest, 
> just using nnz 
> to preallocate the vector. [1]

The "if" statement is not the only difference between the two implementations 
...

* I'd rather not use a ColumnVector to store and array of indices in [1],
  Array<octave_idx_type> would be better ...

* And you may use Array<double> for the matrix entries because you don't 
  need all the overhead from the ColumnVector class.

* For fast access to the Array entries you should get the pointer via the 
  fortran_vec () method.

* Using the SparseMatrix () constructor from Array<...> rather than 
  ColumnVector [3] you don't need to increment all indices as it uses 0-base 
indexing.

* I'm not quite sure statement [4] is correct (of course apart from the double 
semicolon ...)
  what is the idea behind the choice of setting the size to (nr * nz) / (i + 1)?

* You should remove all the verbose debug outputs like [5][6]...
  if you really need to write something to screan you should pass it through 
the pager.


c.


[1] 
http://sourceforge.net/p/octave/fem-fenics/ci/default/tree/src/assemble.cc#l108
[2] 
http://sourceforge.net/p/octave/fem-fenics/ci/default/tree/src/assemble.cc#l124
[3] 
http://sourceforge.net/p/octave/fem-fenics/ci/default/tree/src/assemble.cc#l136
[4] 
http://sourceforge.net/p/octave/fem-fenics/ci/default/tree/src/assemble.cc#l116
[5] 
http://sourceforge.net/p/octave/fem-fenics/ci/default/tree/src/assemble.cc#l192
[6] 
http://sourceforge.net/p/octave/fem-fenics/ci/default/tree/src/assemble.cc#l211