inhibit inlining.

local bases.  This is necessary since the PDELab version needs to do this too,
or be less general.

our context.

quadrature-point loops (!).

highest-numbered coordinate axis.  This should keep similar-indexed entities
together.

runs for different number of threads.

the vector.  Output the norms in high precision.

unlocked concurrent writes.

anything, i.e. there _are_ races.  For benchmarking the impact of locking.