dolfin team mailing list archive

[Johan Jansson <jjan@xxxxxxxxxx>]

Garth N. Wells wrote:
> Ola Skavhaug wrote:
>   
> > I'm just barely able to compile dolfin on my computer
> > due to memory requirements during compilation of ProjectionLibrary. Almost 60%
> > of my memory was in use at peak. The consequence is that I cannot use scons
> > -j4 (or similar) to speed up compilation. 
> >
> > $ cat /proc/meminfo  | grep MemTotal
> > MemTotal:      2063108 kB
> >
> > Is this something we really want?
> >
> >     
>
>
> I also noticed a big slow down in compilation.
>
> I don't think we need ProjectionLibrary. A user can create the files 
> they need using FFC. We discussed some time ago also removing 
> ElementLibrary because user could create their own elements, but kept it 
> so that Functions could be read from file. This reasoning doesn't hold 
> for projections.
>
> Garth
>   
I think you're throwing out the baby with the bathwater here. The idea 
with having a projection library is to be able to have a C++ interface 
for projections. If you have to:

1. Write projection forms (which always look the same except the element).
2. Construct the form objects.
3. Solve the discrete system.

This is not a high-level projection interface.

The analogous Function read interface would be to:

1. Write the FFC description of the element.
2. Read the vector of degrees of freedom.
3. Construct the function object.

A much simpler fix is to remove some of the elements in the library. 
Currently the library (ElementLibrary and ProjectionLibrary have the 
same elements) consists of these elements:

"""
FiniteElement("Lagrange", "triangle", 1)
FiniteElement("Lagrange", "triangle", 2)
FiniteElement("Lagrange", "triangle", 3)
FiniteElement("Lagrange", "tetrahedron", 1)
FiniteElement("Lagrange", "tetrahedron", 2)
FiniteElement("Lagrange", "tetrahedron", 3)
FiniteElement("Discontinuous Lagrange", "triangle", 0)
FiniteElement("Discontinuous Lagrange", "triangle", 1)
FiniteElement("Discontinuous Lagrange", "triangle", 2)
FiniteElement("Discontinuous Lagrange", "tetrahedron", 0)
FiniteElement("Discontinuous Lagrange", "tetrahedron", 1)
FiniteElement("Discontinuous Lagrange", "tetrahedron", 2)
FiniteElement("Discontinuous Lagrange", "tetrahedron", 3)
VectorElement("Lagrange", "triangle", 1)
VectorElement("Lagrange", "triangle", 2)
VectorElement("Lagrange", "triangle", 3)
VectorElement("Lagrange", "tetrahedron", 1)
VectorElement("Lagrange", "tetrahedron", 2)
VectorElement("Lagrange", "tetrahedron", 3)
VectorElement("Discontinuous Lagrange", "triangle", 0)
VectorElement("Discontinuous Lagrange", "triangle", 1)
VectorElement("Discontinuous Lagrange", "triangle", 2)
VectorElement("Discontinuous Lagrange", "tetrahedron", 0)
VectorElement("Discontinuous Lagrange", "tetrahedron", 1)
VectorElement("Discontinuous Lagrange", "tetrahedron", 2)
VectorElement("Discontinuous Lagrange", "tetrahedron", 3)
FiniteElement("Brezzi-Douglas-Marini", "triangle", 1)
"""

I propose the reduced list:


"""
FiniteElement("Lagrange", "triangle", 1)
FiniteElement("Lagrange", "triangle", 2)
FiniteElement("Lagrange", "tetrahedron", 1)
FiniteElement("Lagrange", "tetrahedron", 2)
FiniteElement("Discontinuous Lagrange", "triangle", 0)
FiniteElement("Discontinuous Lagrange", "triangle", 1)
FiniteElement("Discontinuous Lagrange", "triangle", 2)
FiniteElement("Discontinuous Lagrange", "tetrahedron", 0)
FiniteElement("Discontinuous Lagrange", "tetrahedron", 1)
FiniteElement("Discontinuous Lagrange", "tetrahedron", 2)
VectorElement("Lagrange", "triangle", 1)
VectorElement("Lagrange", "triangle", 2)
VectorElement("Lagrange", "tetrahedron", 1)
VectorElement("Lagrange", "tetrahedron", 2)
VectorElement("Discontinuous Lagrange", "triangle", 0)
VectorElement("Discontinuous Lagrange", "triangle", 1)
VectorElement("Discontinuous Lagrange", "triangle", 2)
VectorElement("Discontinuous Lagrange", "tetrahedron", 0)
VectorElement("Discontinuous Lagrange", "tetrahedron", 1)
VectorElement("Discontinuous Lagrange", "tetrahedron", 2)
FiniteElement("Brezzi-Douglas-Marini", "triangle", 1)
"""

This cuts compilation and memory size by about half.

Current list:

user    2m43.682s

Reduced list:

user    1m11.344s

What do you think?

 Johan