[PATCH] [gdb/testsuite] Fix gdb.arch/i386-avx.exp with clang

[Tom de Vries]

On 11/5/21 2:52 PM, Andrew Burgess wrote:
> * Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> [2021-11-05 14:35:20 +0100]:
> 
>> On 11/5/21 2:20 PM, Pedro Alves wrote:
>>> On 2021-11-05 13:15, Tom de Vries wrote:
>>>> On 11/5/21 1:55 PM, Pedro Alves wrote:
>>>>> On 2021-11-05 12:23, Tom de Vries via Gdb-patches wrote:
>>>>>
>>>>>>> No, but in gdb/testsuite/lib/attribute.h we do setup a compatibility
>>>>>>> macro for 'noclone', so there's definitely precedent for using
>>>>>>> attributes that might not be supported everywhere.
>>>>>>>
>>>>>>
>>>>>> Right, I'm aware of this, but that's a typical case where we have no
>>>>>> portable alternative.
>>>>>
>>>>> We actually do -- _Alignas is standard C11.  This fixes the test as well:
>>>>>
>>>>>   _Alignas(32) v8sf_t data[] =
>>>>>
>>>>
>>>> I was referring to the noclone, but ok, I was not aware of the _Alignas,
>>>> good to know, thanks.
>>>>
>>>> Anyway, in the latest version this is not relevant anymore, since the
>>>> precise alignment implementation has an extra benefit, as explained in
>>>> the post.
>>>>
>>>
>>> OOC, is that benefit important here?
>>>
>>
>> So, this is the post I mentioned (
>> https://sourceware.org/pipermail/gdb-patches/2021-November/183183.html ).
>>
>> Well, the benefit is that it prevents accidental overalignment, which is
>> the reason that this problem escaped detection and/or fixing for so long.
>>
>> Without that, I could do a thinko and specify too small an alignment and
>> have the test passing accidentally, only to fail in a different setup.
> 
> I'm still not convinced.  The test doesn't appear to be about the
> alignment, 

Correct.

> but about accessing the feature specific registers, so I
> don't see how making a mistake with the alignment would be different
> to any other bug - eventually it gets spotted and fixed.

Right, and my perception is that using precise alignment gives me
confidence that I fixed it properly.

Thanks,
- Tom