Re: complex float support and 64 bit hosts

[craig at jcb-sc dot com]

>We can perhaps use the same technique to solve our 64-bit mips problem with
>complex float.  The n32/n64 ABIs imply that complex float values are supposed
>to be returned in two FP registers.  There is still a problem for complex float
>parameters, as the ABI does not specify that they are passed in two FP
>registers.  Instead, it appears that they get passed in one general purpose
>register.  If we are willing to violate the ABI, we could pass them in two
>separate registers instead.

For now, it's probably best to assume that any ABI that doesn't clearly
and precisely mandate exactly what a complex type *is* (in terms of how
languages like C++ see them) and how it is to be passed, is broken wrt
to complex types.

In that case, GCC should at least get this stuff working by treating
a complex FOO type exactly as it does "struct { FOO r; FOO i; };".

Then, document any concerns about ABI ambiguity or compatibility in the
known-bugs list for the pertinent systems.

        tq vm, (burley)