Register usage for the lpm_mult megafunction

Hi,

I was checking the resource utilization for the lpm_mult megafunction. I have used 16 bits inputs and a pipeline = 1.

Although knowing that is not the best-performance option, I choose an implementation on logic elements. The compilation report gives an utilization of 220 ALUTs and 86 dedicated logic registers for a Stratix III device.

However, for a pipeline equals to 1, I would expected a number of registers equals to 32, which is the number of bits for the result. Anyone knows why the compiler uses more number of registers and which is its purpose?

Thank you in advance.

P.D: for 32 bits inputs the number of registers grows to 400 !