Re: Options of fixing biggest alignment in PR target/38736

["Richard Guenther" <richard dot guenther at gmail dot com>]

On Fri, Jan 9, 2009 at 4:59 PM, Ian Lance Taylor <iant@google.com> wrote:
> "Richard Guenther" <richard.guenther@gmail.com> writes:
>
>> On Fri, Jan 9, 2009 at 4:31 PM, Ian Lance Taylor <iant@google.com> wrote:
>>> "Richard Guenther" <richard.guenther@gmail.com> writes:
>>>
>>>>> There are many ways to align data without exposing it in the
>>>>> ABI--e.g., the alignment of a global array is not part of the ABI, in
>>>>> that nothing breaks if the alignment is increased.  Also, there are
>>>>> many programs which simply don't care about an external ABI.
>>>>
>>>> I'd say if programs want to use vectorization they should use
>>>> alignof (vector_type) instead of a magic attribute((aligned(max))).
>>>
>>> It's not that programs explicitly want to use vectorization.  It's
>>> that auto-vectorization happens on ordinary scalar types, and programs
>>> want to permit that to happen without worrying about the details.
>>
>> Sure.  The vectorizer increases alignment of variables for this reason.
>> But as people cannot adjust their memory allocator behavior they
>> cannot use attribute((aligned(max))) for that either.  Or do you suggest
>> they should stick that on every decl?  (which is what the autovectorizer
>> does)
>
> People can and do adjust their memory allocator.  Doing that in
> portable code requires something like attribute((aligned)).

Well.  As HJ says, an attribute((aligned)) is not the best way to
communicate this.  You would need sth like

 char a[] __attribute__((aligned));

and then query alignof (a) for your alignment bit-twiddling inside
your allocator.  Certainly possible, but also "interesting" use
of attribute((aligned)) ;)

Richard.