[deal.II] Trouble getting deal.II working on bgq

[Jie Cheng]

Happy new year everyone

I have been trying to install deal.II on my university's blue gene Q system 
recently. The IBM XL compiler wouldn't work because it doesn't support 
C++11 features. An old thread in the mailing list also suggests that it is 
an dead end... I also tried GCC compilers, unfortunately we only have GCC 
4.7.2 available and dealii requires 4.8 or higher... I do need the latest 
version of dealii so using an old version is not acceptable. 

Is there a simple way to get deal.II working with GCC 4.7.2? If not, how to 
install a newer GCC on a cross-compiling system like bgq?

Thank you
Jie

-- 
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see 
https://groups.google.com/d/forum/dealii?hl=en
--- 
You received this message because you are subscribed to the Google Groups 
"deal.II User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to dealii+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Re: [deal.II] New deal.II 8.5.1 is 20% slower than deal.II 8.0.0

[luca.heltai]

This depends on the finite element. For FEQ, yes, for other finite elements, 
definitely not (RaviarThomas, Nedelec, etc.).

Also, the code was wrong with the numbering, of course. On the left you should 
put the global dof number associated to i, not i. 

Even if the local matrix is identical, storing it with the correct numbering 
into a big sparse matrix is only marginally expensive, while saving a lot in 
extracting local dofs.

One vmult, followed by memory contiguous access at every cell, is much cheaper 
than searching in the global vector for local dofs, then performing one local 
multiplication (maybe with one identical Vandermonde matrix).

L. 

> On 31 Dec 2017, at 6:09, Praveen C  wrote:
> 
> 
> 
>> On 30-Dec-2017, at 11:40 PM, luca.heltai  wrote:
>> 
>> I’m thinking of a matrix of size (n_quadrature_points x n_active_cells) x 
>> n_dofs, and then  you slice the results cellwise instead of repeatedly 
>> calling get_function_values.
>> 
>> once:
>> 
>> M[q+active_cell_index*n_dofs_per_cell, i] = fe_values.shape_value(i,q);
> 
> This looks like a Vandermonde matrix which would be identical on every cell. 
> In that case, it is not necessary to store it for every cell.
> 
> Thanks
> praveen
> 
> -- 
> The deal.II project is located at http://www.dealii.org/
> For mailing list/forum options, see 
> https://groups.google.com/d/forum/dealii?hl=en
> --- 
> You received this message because you are subscribed to the Google Groups 
> "deal.II User Group" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email to dealii+unsubscr...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

-- 
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see 
https://groups.google.com/d/forum/dealii?hl=en
--- 
You received this message because you are subscribed to the Google Groups 
"deal.II User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to dealii+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.