Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix for Cofunction self-assignment via interpolation #3939

Open
wants to merge 5 commits into
base: master
Choose a base branch
from
Open

Fix for Cofunction self-assignment via interpolation #3939

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
b8c5252
Fix for Cofunction self-assignment via interpolation
jrmaddison Dec 18, 2024
98f8c1e
Update firedrake/interpolation.py
jrmaddison Dec 18, 2024
63b74ca
Add a more complicated test for dual interpolation where the tensor i…
jrmaddison Dec 19, 2024
a2b6002
== -> np.allclose
jrmaddison Dec 19, 2024
2dc087a
Add an explanatory comment
jrmaddison Dec 19, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 6 additions & 1 deletion firedrake/interpolation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -859,7 +859,12 @@ def _interpolate(self, *function, output=None, transpose=False, **kwargs):
V = self.V
result = output or firedrake.Function(V)
with function.dat.vec_ro as x, result.dat.vec_wo as out:
mul(x, out)
if x.id != out.id:
jrmaddison marked this conversation as resolved.
Show resolved Hide resolved
mul(x, out)
else:
out_ = out.duplicate()
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't a no-op be the right thing?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think so, e.g.

u = fd.Function(space, name="u")
v = fd.Function(space, name="v")
w = fd.Cofunction(space.dual(), name="w")
...
fd.assemble(fd.action(
    fd.adjoint(fd.derivative(interpolate(v * u, space), u)),
    w), tensor=w)

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see, it only is a no-op in the MFE from the issue: assemble(L(w), tensor=w) with L being the identity.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe it'd be good to add a more complicated test then

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done. I think there are also be issues with complex here (transpose should be Hermitian for the adjoint actions?).

mul(x, out_)
out_.copy(result=out)
return result

else:
Expand Down
7 changes: 7 additions & 0 deletions tests/firedrake/regression/test_interp_dual.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,13 @@ def f1(mesh, V1):
return Function(V1).interpolate(expr)


def test_interp_self(V1):
a = assemble(TestFunction(V1) * dx)
b = assemble(TestFunction(V1) * dx)
a.interpolate(a)
assert (a.dat.data_ro == b.dat.data_ro).all()
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not sure if this should be a np.allclose test

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think you're right as written. No flops should be performed so that should be an equality.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think it's actually no flops, it's interpolation which happens to be the identify except for roundoff.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Changed to np.allclose, I'm pretty sure now that's correct



def test_assemble_interp_operator(V2, f1):
# Check type
If1 = Interpolate(f1, V2)
Expand Down
Loading