The context is the patch on bug 1637328: there must be code in the compiler for handling the case when an InterpEntry accepts or returns a v128, because we generate an InterpEntry for exported functions with that signature, yet when those entries are called we will throw before we get to the code that handles the v128 value. If it's easy to do so and does not compromise other functionality (eg debugging) it would be better to generate no entry stub, or a simple entry stub that just crashes.

We should also consider whether there are other cases like this.

This is more complicated because of the debug code in the shell, which has a backdoor for v128. So we could avoid generating stubs only in the browser, or we could make stubs for the v128 case lazy / lazier. Probably not urgent either way.