Re: [RFA] powerpc - extract a float return value

[Kevin Buettner <kevinb at redhat dot com>]

On Tue, 20 Apr 2004 20:03:48 +0200
Jerome Guitton <guitton@act-europe.fr> wrote:

> The way GDB extracts float return values in rs6000-tdep.c
> (rs6000_extract_return_value) seems dubious to me:
> 
> [...]
> memcpy (&dd, &regbuf[DEPRECATED_REGISTER_BYTE (FP0_REGNUM + 1)], 8);
> ff = (float) dd;
> memcpy (valbuf, &ff, sizeof (float));
> [...]
> 
> The cast will not work properly if the target and the host have not a similar
> float representation.

I agree.

> I propose to fix that by a call to convert_typed_floating. See patch in
> attachment.

This looks right to me.

The rs6000 return value code will need to be revamped to not refer to
DEPRECATED_REGISTER_BYTE.  And, in fact, it probably ought to be
restructured to use the mechanisms found in ppc-linux.c.  See
ppc_linux_return_value() and ppc_sysv_abi_return_value().  This
is not a barrier to getting your current patch in; I only mention
it in case you wish to tackle this bit of work too.

> I have not yet tested it against the testsuite, I will do that
> tomorrow. In the meantime, if you have comments I would be happy to
> address them!

Please let me know the results of your tests.  If they look good,
please commit your patch.

Thanks!

Kevin