This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH, AArch64] Add optimized strchrnul
- From: Richard Earnshaw <rearnsha at arm dot com>
- To: Andrew Pinski <pinskia at gmail dot com>, Kyle McMartin <kmcmarti at redhat dot com>
- Cc: Marcus Shawcroft <marcus dot shawcroft at gmail dot com>, "libc-alpha at sourceware dot org" <libc-alpha at sourceware dot org>
- Date: Wed, 10 Dec 2014 09:38:20 +0000
- Subject: Re: [PATCH, AArch64] Add optimized strchrnul
- Authentication-results: sourceware.org; auth=none
- References: <539AD11E dot 50507 at arm dot com> <CAFqB+PyiWb1Bj+UJa2HRvyCMhn-H57r3Wn05mHpnOn1n_qhzrQ at mail dot gmail dot com> <20141203174546 dot GG4170 at redacted dot bos dot redhat dot com> <CA+=Sn1kkD88LSLyzxVyAdc5zkmsrHJP5at1fvsnLfA0G7fg+=w at mail dot gmail dot com>
On 09/12/14 22:38, Andrew Pinski wrote:
> On Wed, Dec 3, 2014 at 9:45 AM, Kyle McMartin <kmcmarti@redhat.com> wrote:
>> On Wed, Nov 05, 2014 at 02:02:22PM +0000, Marcus Shawcroft wrote:
>>> On 13 June 2014 11:23, Richard Earnshaw <rearnsha@arm.com> wrote:
>>>> Here is an optimized implementation of __strchrnul. The simplification
>>>> that we don't have to track precisely why the loop terminates (match or
>>>> end-of-string) means we have to do less work in both setup and the core
>>>> inner loop. That means this should never be slower than strchr.
>>>>
>>>> As with strchr, the use of LD1 means we do not need different versions
>>>> for big-/little-endian.
>>>>
>>>> <date> Richard Earnshaw <rearnsha@arm.com>
>>>>
>>>> * sysdeps/aarch64/strchrnul.S: New file.
>>>
>>> Committed. + NEWS entry. /Marcus
>>
>> Hrm, I've bisected this commit as the cause of localedef segfaulting as
>> part of install-locales... it ends up crashing weirdly inside of
>> vsprintf's internals. Have either of you seen anything similar?
>
> I just ran into this with an ILP32 compiled glibc (with thunderX
> tuning and some extra patches) where printf was crashing. The problem
> is v15 is being used but it is a callee saved register. Here is the
> documentation from the ABI:
> Registers v8-v15 must be preserved by a callee across subroutine
> calls; the remaining registers (v0-v7, v16-v31)
> do not need to be preserved (or should be preserved by the caller).
>
> Thanks,
> Andrew Pinski
>
>>
>> regards, Kyle
>
Well spotted! I'll push a fix round the various implementations of this
ASAP.
R.