Path: csiph.com!fu-berlin.de!uni-berlin.de!individual.net!not-for-mail
From: Bo Persson <bo@bo-persson.se>
Newsgroups: comp.lang.c++
Subject: Re: Examples of current platforms/architectures where sizeof(void*) > sizeof(int*) ?
Date: Wed, 1 Sep 2021 12:48:44 +0200
Lines: 34
Message-ID: <ip944dFfscpU1@mid.individual.net>
References: <sgglb9$c02$1@dont-email.me> <sgl066$j01$1@dont-email.me> <sgn0c7$1qv$2@gioia.aioe.org> <sgn2ub$kh6$2@dont-email.me> <sgn3r7$pch$1@dont-email.me> <ip8q2oFe188U1@mid.individual.net> <sgncmn$a6b$2@dont-email.me> <ip8sscFei0bU1@mid.individual.net> <sgnfof$v7j$1@dont-email.me> <ip8uptFesaeU2@mid.individual.net> <sgnk0m$s9o$2@dont-email.me>
Mime-Version: 1.0
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
X-Trace: individual.net pZ8pjn5I4jJHIyTWEF4A2wuehL/jkIxCs9MLKOA/q2vehFWv8K
Cancel-Lock: sha1:PDHtcEaOxGUxXSXFXhdEgvjHGwM=
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:78.0) Gecko/20100101 Thunderbird/78.13.0
In-Reply-To: <sgnk0m$s9o$2@dont-email.me>
Content-Language: sv
Xref: csiph.com comp.lang.c++:80964

On 2021-09-01 at 12:17, Bonita Montero wrote:
> Am 01.09.2021 um 11:17 schrieb Bo Persson:
>> On 2021-09-01 at 11:05, Bonita Montero wrote:
>>> Am 01.09.2021 um 10:44 schrieb Bo Persson:
>>>> On 2021-09-01 at 10:13, Bonita Montero wrote:
>>>>> Am 01.09.2021 um 09:57 schrieb Bo Persson:
>>>>>
>>>>>> Even on current byte addressed machines, malloc(1) gets us 8- or 
>>>>>> 16-bytes aligned allocations, partly to reduce memory fragmentation.
>>>>>
>>>>> No, just to limit the number of size-pools.
>>>>
>>>> And why do they have memory pools?  :-)
>>>
>>> To prevent that they would otherwise allocate larger blocks directly
>>> from the kernel, which is slower.
>>>
>>
>> So it's not that limiting allocations to just some sizes reduces 
>> memory fragmentation?
> 
> The discussion is about if there's fragmentation among different
> allocations, i.e. external fragmentation, with modern alllocations.
> Most allocations are small sized allocations, and with modern allo-
> cators theres no external fragmentation at all for this size class
> since the allocations are rounded up to 2^N (f.e. with mimalloc up
> to 8kB) and looked up in a pool which is also 2^N - the result is
> no fragementation among the allocations.
> 

Or, like someone said earlier:

Even on current byte addressed machines, malloc(1) gets us 8- or 
16-bytes aligned allocations, partly to reduce memory fragmentation.