New optimized string routines for Intel and alignment of stack.

[Florian Weimer]

On 06/11/2016 08:01 AM, Mike Frysinger wrote:
> On 07 Jun 2016 17:35, Joseph Myers wrote:
>> On Tue, 7 Jun 2016, Florian Weimer wrote:
>>> On 06/07/2016 07:56 AM, Carlos O'Donell wrote:
>>>> We have had several users that have built legacy applications
>>>> for 32-bit x86 with stack alignment that does not match the
>>>> ABI.
>>>
>>> Let's say the GNU project broke the i386 ABI, which is more accurate. The
>>> stack pointer alignment requirement is a recent change.
>>
>> I place the ABI change around 2006, when GCC stopped using
>> -mpreferred-stack-boundary=2 even with -Os, and so compatibility
>> requirements as applying to GLIBC_2.4 and older symbols.
>>
>> https://sourceware.org/ml/libc-alpha/2013-08/msg00372.html
>
> indeed, the ship has sailed.  i don't think trying to revert things after
> 10 years makes sense.
>
> https://groups.google.com/forum/#!msg/ia32-abi/T5s-UGmUO_E/0-9QgPuB8_MJ

Yeah, I didn't realize the only thing that might work is “preserve 
alignment, but don't rely on it”.  Performance-wise, this is the worst 
option.  But maybe we should do this in glibc.

Florian