tsfc: contract each block after unconcatenating

[pbrubeck]

Stack

Merge in this order, each PR retargets to main once the one above it lands:

#	PR	what
1	firedrakeproject/fiat#269	gem: sum factorise independent contractions
2	firedrakeproject/fiat#271	FInAT: select H(div)/H(curl) components with a Delta
3	firedrakeproject/fiat#268	FInAT: dual evaluate on each sub-element's own points
4	#5295	tsfc: dual evaluation against a Cofunction
5	#5289	p-multigrid: remove custom interpolation
6	#5297	tsfc: contract each block after unconcatenating

2 and 6 are a pair: neither is worth anything alone, though 2 is harmless on its
own — it changes how a component is selected, not what is selected.
5 and 6 are siblings on 4, in either order.

Only 4 carries a DROP BEFORE MERGE commit, pinning FIAT to
pbrubeck/fix/dual-enriched, which is now the top of the FIAT stack; 5 and 6
inherit it. Drop it once the FIAT side has landed.

compile_expression_dual_evaluation had a standing TODO:

# TODO: one should apply some GEM optimisations as in assembly,
# but we don't for now.

Apply one of them: contract each assignment once unconcatenate has split the
Concatenate nodes.

This matters for H(div) and H(curl). Those elements are a sum of blocks, each
mapping into one vector component, and with firedrakeproject/fiat#271 each
block selects its component with a Delta. Before unconcatenation the blocks are
branches of a Concatenate, so the Deltas never meet; afterwards each pair of
blocks is its own assignment, the Deltas multiply, and delta elimination cancels
every pair that maps to different components.

Paired with firedrakeproject/fiat#271. Neither PR does anything on its own —
without the Deltas there is nothing for the contraction to cancel, and without the
contraction the Deltas never meet.

Effect

Coarse to fine interpolation, as p-multigrid prolongation. Kernel stack
temporaries and wall clock:

	temporaries	time
NCF 1→4	832K → 68K	3.07×
NCE 1→4	307K → 110K	2.23×
NCE 1→3	131K → 61K	2.23×
NCF 1→3	223K → 38K	1.86×

Form assembly is unchanged; it does not go through this function.

Testing

Matrix-free mult/multTranspose against the assembled matrix, and an
interpolate/restrict round trip, for NCF/NCE on hexahedra and RTCF/RTCE on
quadrilaterals, spectral and fdm variants. Full FIAT suite on the paired branch.

🤖 Generated with Claude Code