[Bug target/77330] __float128 segfaults on 32-bit x86 due to 8-byte malloc alignment

["bernd.edlinger at hotmail dot de" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77330

--- Comment #11 from Bernd Edlinger <bernd.edlinger at hotmail dot de> ---
(In reply to joseph@codesourcery.com from comment #10)
> On Tue, 23 Aug 2016, bernd.edlinger at hotmail dot de wrote:
> 
> > Yes, but isn't that an ABI decision?
> 
> It's the sort of ABI decision that's fixed after it's appeared in a 
> release (whereas max_align_t is not the sort of thing that gets embedded 
> in external ABIs in a way that gets broken when increasing the alignment, 
> but something you use when writing your own allocator).  HJ's i386 ABI 
> document, version 1.1 is explicit about this ABI choice (and that for 
> _Decimal128, etc.), so presumably it's intended to be followed the same by 
> other implementations as well.

H.J's document did not mention any requirements on malloc.
It is only necessary if __float128 is declared a base type.
But __m512 is no base type, right?

Then MALLOC_ABI_ALIGNMENT should be changed to 128 at least for the i386.

For x864_64 this is already the case.

But note that it will be difficult to patch all glibc implementations
all over the world.  Maybe it should be a configure option then?