On 11 April 2011 16:19, Richard Henderson <r...@twiddle.net> wrote:
> On 04/11/2011 08:09 AM, Peter Maydell wrote:
>> (4) I think you should be able to write a helper function for an
>> add as just
>> float32 HELPER(my_float_add)(float32 a, float32 b) {
>> return float32_add(a, b, status);
>> }
>
> While this is a laudable goal, this will fail for hosts that pass
> all structures by reference. This is true of, e.g. PPC32.
...but only if float32 is a struct, which is where we came in.
In the sane default configuration float32 is just a uint32_t
in disguise.
In other words, my point is that I'd prefer to give up[*] being
able to run with float32-is-a-struct rather than give up having
clean and straightforward helper functions.
[*] actually I suspect we've never actually had this capability
so we're not really giving anything up except philosophically...
-- PMM
