Abstract: In order to verify the function of the floating-point unit in the microprocessor high-efficiently, numerous corner cases of floating-point arithmetic has been studied. Several arithmetics are introduced to solve the constriants of the intermediate results. Compared with the traditional function verification of floating-point unit, the floating-point number generator based on the constraint solved algorithms has widened the scope of the optional floating-point corner cases, which effectively improves the verification efficiency.Experimental results show that the UVM verification platform which integrated the generator can test one floating-point subunit efficiently, which achieves high coverage beyond 98%, within 12 hours of testing time.