HDL code for the super-sample FIR Filters does optimize for powers of 2, zeros and symmetry. I've attached a simple model which has filters (each with 4 coefficients) processing 4 samples at a time to show the effect of these optimizations.
The _noOpt branch consumes 16 multipliers, the _pow2 branch consumes 8 multipliers (2 coefficients are powers of 2) as does the _zeros branch (with 2 zero coefficients) and the _symm branch consumes 8 multipliers with smaller adders (symmetry optimization).
You can run HDL code generation on the top-levl subsystem and a resource report is generated that will show these numbers for each subsystem below (which have the filters inside).
