In any case, during lower call,
the Args and Rety should be the same and not depend on any of this. For soft float, they are turning into Int64 in the case of double, for example. Those values usually just reflect the
actual function signature.
I have a workaround for this issue.
I will need to think about whether I need to do something to handle this errno case for mips16 fp. I will see how gcc handles this.