[m-rev.] for review: Fix abort in float_regs pass (bug #409).

[Julien Fischer]

On Tue, 6 Mar 2018, Peter Wang wrote:

> The float_regs pass did not consider this possibility: a term being
> deconstructed can have a ground inst with no higher-order inst
> information, but higher-order inst information is still available for
> the sub-unifications of the term's arguments. This is possible with the
> introduction of the "combined higher-order types and insts" feature.
> Losing the higher-order inst information led to an exception:
>
>    Unexpected: no higher order inst
>
> compiler/float_regs.m:
>    Handle the case above in unify_mode_set_rhs_final_inst.
>
>    Minor improvements in comments.
>
> tests/hard_coded/Mmakefile:
> tests/hard_coded/functor_ho_inst_float_reg.exp:
> tests/hard_coded/functor_ho_inst_float_reg.m:
>    Add test case.

That's fine.

Julien.