This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH v2] aarch64: Hoist ZVA check out of the memset function
- From: Szabolcs Nagy <szabolcs dot nagy at arm dot com>
- To: siddhesh at sourceware dot org, libc-alpha at sourceware dot org
- Cc: nd at arm dot com, adhemerval dot zanella at linaro dot org
- Date: Thu, 05 Oct 2017 10:30:38 +0100
- Subject: Re: [PATCH v2] aarch64: Hoist ZVA check out of the memset function
- Authentication-results: sourceware.org; auth=none
- Authentication-results: spf=none (sender IP is ) smtp.mailfrom=Szabolcs dot Nagy at arm dot com;
- Nodisclaimer: True
- References: <1507031555-3046-1-git-send-email-siddhesh@sourceware.org> <59D507C1.4000101@arm.com> <e3c4dba2-9859-13f4-5c42-e0003f3c1107@sourceware.org>
- Spamdiagnosticmetadata: NSPM
- Spamdiagnosticoutput: 1:99
On 05/10/17 05:22, Siddhesh Poyarekar wrote:
> On Wednesday 04 October 2017 09:39 PM, Szabolcs Nagy wrote:
>>> +#if IS_IN (rtld)
>>> +# define MEMSET memset
>>> +# define INTERNAL_MEMSET
>>> +# define MEMSET_L(label) L(label)
>>> +# include <sysdeps/aarch64/memset.S>
>>> +#endif
>>
>> is this performance critical?
>> (i don't know if it's only used for early dynlinker startup code
>> or all libc internal memsets)
>>
>> if not then we don't need it to be different variant, it could be
>> just a copy of the _nozva/_defaultzva case
>
> This is used only in rtld, the libc internal one is __memset_generic.
> It is not really performance critical, but given that I'll be including
> one or the other anyway, I chose to include memset.S instead of
> memset_nozva.S. I can do the latter.
>
it is ok to use the generic one, it just seemed that there
is a generic and a default_zva variant which could be unified
(only one of them is necessary) and then that can be used
for rtld, generic and non-conventional zva size cases.