TSFC: support same-mesh ufl.Interpolate in Form kernels by pbrubeck · Pull Request #5258 · firedrakeproject/firedrake

pbrubeck · 2026-07-16T08:51:16Z

Summary

TSFC now compiles same-mesh ufl.Interpolate nodes in normal Form kernels, and this PR ports the feature from PyOP2 to pyop3.

Standalone Interpolate and Form(Interpolate) share one code generation path.
Same-mesh interpolation across a MixedFunctionSpace supports mat_type="aij" natively.
Interpolation into a VertexOnlyMesh uses the same TSFC code generation.
Cross-mesh interpolation stays on BaseFormAssembler.

SameMeshInterpolator no longer builds its own op3.loops. It sends the interpolation through the form assembler:

 SameMeshInterpolator._get_callable
-  _build_interpolation_callables      # hand-built op3.loop per sub-expression
+  get_form_assembler(Interpolate, access=...)
+    tsfc_interface.compile_form(..., access)
+      as_pyop3_local_kernel(ast, nargs, access)
+        op3.Function(..., requires_zeroed_output_arguments=True)

MIN and MAX are reduced into the output as MIN_WRITE and MAX_WRITE. TSFC kernels add their result to the output argument, so pyop3 starts these temporaries from zero (see #5543).

TwoFormAssembler._apply_bc now takes the field label of a BC from the BC's own space. Before, a rectangular form such as an interpolation from W = V * V into V failed.

Stacked on #5543, which adds MIN/MAX reductions to pyop3 kernel calls.
Depends on firedrakeproject/fiat#294.

Fixes #5468

Evidence

Before (merge of main without the fixes): 7 failures in tests/firedrake/regression/test_interpolate.py

test_subset_interpolation_reduction[min|max]     -> all entries stay at ±finfo.max
test_interpolate_periodic_coords_reduction[...]  -> [inf]
test_mixed_space_bcs                             -> MultipleUniqueEntriesException

After:

serial: interpolation, cross-mesh, vertex-only, MG grid transfer, pyop3 loop and TSFC tests  1006 passed
serial: BC, mixed, real-space, matrix and assembly regression tests                           740 passed
mpiexec -n 2  -m "parallel[2]" interpolation and vertex-only tests                            32 passed
mpiexec -n 3  -m "parallel[3]" interpolation and vertex-only tests                           271 passed

Merge Danger

Door: two-way

Blast Radius: interpolation

All same-mesh interpolation, and the Form kernels that contain it, now go through TSFC and the form assembler. op3.Function gets a new attribute, so kernel caches from before this change must be cleared with firedrake-clean.

The CI commits marked DROP BEFORE MERGE install the FIAT branch and must be dropped before merging.

Contribution assisted by Claude Code (Claude Opus 5.5).

🤖 Generated with Claude Code

pbrubeck · 2026-07-16T11:21:11Z

This PR should reimplement SameMeshInterpolator as a dispatcher to {Zero|One|Two}FormAssembler. We want to unify the TSFC codepaths for Form assembly and same-mesh interpolation.

Move get_runtime_tabulation_coordinates next to the other VertexOnlyMesh map helpers in interpolation.py, so that assemble.py does not add code inside _GlobalKernelBuilder or to imports that pyop3 rewrites. Keep unchanged the lines that pyop3 edits nearby: the firedrake.mesh import in interpolation.py, the cell-size assignment in set_cell_sizes, and the tsfc imports in mg/kernels.py. Move the new interpolation tests away from the test that pyop3 skips. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

The type pyop2.types.access.Access has no documentation that Sphinx can link to. Name the access descriptor in the parameter description, so that the docs need no extra nitpick exception. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Kernel calls with MIN_WRITE or MAX_WRITE access now pack a temporary that starts from the identity of the reduction, and unpack it with a "min" or "max" assignment. Assignments of type "min" and "max" now get their intents from assignment_type_as_intent, so they can be put in a loop. Fixes #5524 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Port the feature from PyOP2 to pyop3. The conflict resolutions come from pbrubeck/pyop3/form-interp-tsfc-port (927acc0), which ports the feature onto the pyop3 branch at eb98a70. The changes on main since eb98a70 are applied on top of that port. The result differs from the port only by those changes, with two exceptions: * firedrake/__init__.py exports both CommMismatchError from main and MismatchingFunctionSpaceError from the feature. * firedrake/interpolation.py does not take the change on main from float(f) to f.dat.global_data.item() in SameMeshInterpolator. The port assembles the interpolation through get_form_assembler, which returns zero-form results from ZeroFormAssembler. Therefore, the code that main changed no longer exists. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

TSFC kernels add their result to the output argument. A MIN_WRITE or MAX_WRITE temporary that starts from the identity of the reduction therefore stays at the identity. op3.Function now takes requires_zeroed_output_arguments, like the PyOP2 Kernel did. When it is set, these temporaries start from zero. as_pyop3_local_kernel sets it for TSFC kernels. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Stack the feature on #5543, which supports MIN and MAX reductions in pyop3. firedrake/interpolation.py keeps the feature's version. #5543 changes SameMeshInterpolator._get_tensor to accept MIN_RW and MAX_RW. The feature removes _get_tensor, and its Interpolator.__init__ already converts MIN_RW and MAX_RW to MIN_WRITE and MAX_WRITE. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

TwoFormAssembler._apply_bc required the row and column axes of the matrix to have the same field label for the index of the boundary condition. A rectangular form, such as an interpolation from a mixed space into a single space, does not satisfy this, and test_mixed_space_bcs failed. The label now comes from the field axis of the space of the boundary condition. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

# Conflicts: # firedrake/interpolation.py

Co-authored-by: Pablo Brubeck <brubeck@protonmail.com>