Why there are high memory requirements for FunctionSpace with high order polynomials?

6 weeks ago by

I have a problem to create a FunctionSpace with high order polynomials (8) because it uses too much RAM (more than 8 GB).
The following example is calculated in FEniCS version 2017.2.0.

from dolfin import FunctionSpace, UnitCubeMesh
mesh = UnitCubeMesh(1,1,1)
V = FunctionSpace(mesh, 'CG', 8)
Community: FEniCS Project

1 Answer

6 weeks ago by
This is a bug/regression with switching to new UFLACS code generation between 2017.1 and 2017.2. If you run this


cat - > /tmp/test_uflacs_bug.ufl <<EOF
element = FiniteElement("P", tetrahedron, 8)

python3 -m ffc -v /tmp/test_uflacs_bug.ufl​

this produces in 120 seconds 127Mbytes header file which makes C++ compiler choke on it.

FFCx produces 5.6Mbytes source and I did not test what compiler says.

The problem reported: https://bitbucket.org/fenics-project/ffc/issues/176.
Please login to add an answer/comment or follow this question.

Similar posts:
Search »