The two options produce the same results. The only difference being how the results are being computed internally. Some users prefer to have high customization on their generated code when using Simulink Coder or Embedded Coder; this option provides the ability to do so.
There are situations that can arise when using fixed-point data types where you can obtain different results, but if you are obtaining different or unexpected results then you would probably want to reconsider the fraction and word length of your fixed-point data types.
If you are receiving the same results and are either not performing any code generation on your model then I would ignore this option and just go along with whichever you feel most comfortable with