Output of lpm_divide is x's when inputs are 0 in modelsim

Although zero divide by zero is mathematically NAN (not-a-number), a parallel divider should give a defined '0'/'1' bit vector.

I don't see a reason to expect an all '0' output.

The MegaFunction user manual specifies 'x' output for divide by zero. I think, this means it's up to the synthesis tool optimization to make individual bit either '0' or '1'. The simulator is apparently reflecting this ambiguity, you should run a gate level simulation to see the actual output. Forcing zero output for divide by zero requires additional logic anyway.