Hi,
I’m wondering if there’s anything similar to parameters from legacy dolfin inside dolfinx?parameters["form_compiler”] in dolfinx.
Best,
dokken
July 15, 2022, 8:43am
2
I’m DOLFINx you send in form_compiler_params as a dictionary to the functions that use it (I.e. assemble_*, expression, LinearProblem and NonlinearProblem.
For JIT parameters, see: JIT Parameters and visualization using Pandas — FEniCSx tutorial
For form compiler parameters, see for instance:https://docs.fenicsproject.org/dolfinx/v0.4.1/python/generated/dolfinx.jit.html?highlight=form_compiler_params#dolfinx.jit.ffcx_jit
dolfinx.fem.petsc.assemble_vector(b1, Lf)
end = time.time()
print("Time (C++, pass 1):", end - start)
b1.ghostUpdate(addv=PETSc.InsertMode.ADD, mode=PETSc.ScatterMode.REVERSE)
assert (b1 - b0.vector).norm() == pytest.approx(0.0)
# Assemble using generated tabulate_tensor kernel and Numba assembler
ffcxtype = "double _Complex" if np.issubdtype(PETSc.ScalarType, np.complexfloating) else "double"
b3 = Function(V)
ufcx_form, module, code = dolfinx.jit.ffcx_jit(
mesh.comm, L, form_compiler_params={"scalar_type": ffcxtype})
nptype = "complex128" if np.issubdtype(PETSc.ScalarType, np.complexfloating) else "float64"
# First 0 for "cell" integrals, second 0 for the first one, i.e. default domain
kernel = getattr(ufcx_form.integrals(0)[0], f"tabulate_tensor_{nptype}")
for i in range(2):
b = b3.x.array
b[:] = 0.0
start = time.time()
assemble_vector_ufc(b, kernel, (x_dofs, x), dofmap, num_owned_cells)
2 Likes
Perfect! Thank you so much!