Fixed point data types are less accurate (but fast in execution) alternative for floating point data types (single & double).
If you have a floating point model, then your calculation outputs in model will have real values with more accuracy. For example a signal PRESSURE can have value of 25.356568978.
If you convert same model to Fixed point, you need to scale each floating point signal to a fixed point type with required resolution.
Signals in your fixed point model will have accuracy based on scaling. For example PRESSURE can have value of 25.35, 25.40 etc (here resolution is 0.05)
If you test a floating point model and then convert it to Fixed point and test again, results will be different.
MIL test is used to check the correctness of logic/algorithm of your model.
SIL test used to check the correctness of generated code (there can be bug or issues with code generation process also)
So this is upto your project to decide which testings you need based on project complexity, model complexity, available tools etc.