Re: [PATCH] malloc: Revert fastbins to old-style atomics

[Carlos O'Donell <carlos at redhat dot com>]

On 1/16/19 10:33 AM, Florian Weimer wrote:
> * Carlos O'Donell:
> 
>> On 1/16/19 8:04 AM, Florian Weimer wrote:
>>> Commit 6923f6db1e688dedcf3a6556da76e0bf24a41872 ("malloc: Use current
>>> (C11-style) atomics for fastbin access") caused a substantial
>>> performance regression on POWER and Aarch64, and the old atomics,
>>> while hard to prove correct, seem to work in practice.
>>
>> Why is it slow?
>>
>> What do the benchmarks say?
>>
>> What does profiling show to be the slow instruction?
>>
>> How is the instruction sequence different from before that causes the
>> problem?
> 
> The other thread is here:
> 
>   <https://sourceware.org/ml/libc-alpha/2019-01/msg00379.html>

OK for master. Siddhesh needs to approve.

I think your change needs to go back in when 2.30 opens, it's such
a nice cleanup, and IBM and Arm need to do an RCA on their C11-style
atomics.

Reviewed-by: Carlos O'Donell <carlos@redhat.com>

-- 
Cheers,
Carlos.