First, this doesn't look to be parameterized.
Second, the comment is wrong unless there is a specific set of assumptions on the inputs.
Third, if such assumptions exist it is much better to write the code in a way the tools can use them.
for example, this is a 128 bit add (maybe, it looks like you are using 4 16 bit values which is 64b). It isn't clear that this never overflows/underflows into other 32b values.
I think you want to use for-generate with [32*myGenvar +: 32].