[Ltrace-devel] Current status of git tree on ppc64le

[Petr Machata]

"Thierry Fauck at linux.vnet.ibm.com" <thierry at linux.vnet.ibm.com> writes:

>  * With GGC 4.9.2 , we get a message:
>
>         /tmp/lt-V2vnjpabM5.c:73:10: note: the ABI of passing
>    homogeneous float aggregates will change in a future GCC release

Per this commit:

	https://gcc.gnu.org/ml/gcc-patches/2014-07/msg01173.html

... it seems like the new convention puts arguments to FPR first, GPR
next, and memory last, while the old (buggy) ABI only put to FPR what it
could and to memory the rest.

But allocate_hfa in sysdeps/linux-gnu/ppc/fetch.c handles this case,
there's an explicit code to use part of r10 for the first argument that
doesn't fit to an FPR.

How comes we are not seeing failures then?  Clearly we have functions in
the test suite that will be impacted by the change.

>  * So I am proposing the following patch to get rid of it:

The patch didn't apply using "git am", so I commited it under my name
and credited you.  It's not pushed yet, I got distracted by some bugs in
Dwarf backend that that I noticed when re-testing.

Thanks,
Petr