Some microprocessors have multiply-add fused floating point units with a reduced multiply array, requiring two passes through the array for operations involving double precision multiplies. The paper ...