From: Cupertino Miranda <cupertino.miranda@oracle.com>
To: Szabolcs Nagy <Szabolcs.Nagy@arm.com>
Cc: Wilco Dijkstra <Wilco.Dijkstra@arm.com>, libc-alpha@sourceware.org
Subject: Re: [PATCH] AArch64: Optimize strlen
Date: Thu, 02 Mar 2023 11:49:48 +0000 [thread overview]
Message-ID: <87lekf9xur.fsf@oracle.com> (raw)
In-Reply-To: <Y8FN4C/7ttByIbe+@arm.com>
Hi Szabolcs,
I am attempting to reproduce the presented performance improvements on a
Ampere Altra processor.
Can you please detail some more what was your setup and how did you
measured it.
BTW, I am not looking to discredit the work, but rather to be able
to replicate the results on our end to evaluate backporting your patches.
Best regards,
Cupertino
Szabolcs Nagy via Libc-alpha writes:
> The 01/12/2023 15:53, Wilco Dijkstra wrote:
>> Optimize strlen by unrolling the main loop. Large strings are 64% faster on
>> modern CPUs. Passes regress.
>
> please commit it, thanks.
>
> Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
>
>
>>
>> ---
>>
>> diff --git a/sysdeps/aarch64/strlen.S b/sysdeps/aarch64/strlen.S
>> index b3c92d9dc9b3c52e29e05ebbb89b929f177dc2cf..133ef933425fa260e61642a7840d73391168507d 100644
>> --- a/sysdeps/aarch64/strlen.S
>> +++ b/sysdeps/aarch64/strlen.S
>> @@ -43,12 +43,9 @@
>> #define dend d2
>>
>> /* Core algorithm:
>> -
>> - For each 16-byte chunk we calculate a 64-bit nibble mask value with four bits
>> - per byte. We take 4 bits of every comparison byte with shift right and narrow
>> - by 4 instruction. Since the bits in the nibble mask reflect the order in
>> - which things occur in the original string, counting trailing zeros identifies
>> - exactly which byte matched. */
>> + Process the string in 16-byte aligned chunks. Compute a 64-bit mask with
>> + four bits per byte using the shrn instruction. A count trailing zeros then
>> + identifies the first zero byte. */
>>
>> ENTRY (STRLEN)
>> PTR_ARG (0)
>> @@ -68,18 +65,25 @@ ENTRY (STRLEN)
>>
>> .p2align 5
>> L(loop):
>> - ldr data, [src, 16]!
>> + ldr data, [src, 16]
>> + cmeq vhas_nul.16b, vdata.16b, 0
>> + umaxp vend.16b, vhas_nul.16b, vhas_nul.16b
>> + fmov synd, dend
>> + cbnz synd, L(loop_end)
>> + ldr data, [src, 32]!
>> cmeq vhas_nul.16b, vdata.16b, 0
>> umaxp vend.16b, vhas_nul.16b, vhas_nul.16b
>> fmov synd, dend
>> cbz synd, L(loop)
>> -
>> + sub src, src, 16
>> +L(loop_end):
>> shrn vend.8b, vhas_nul.8h, 4 /* 128->64 */
>> sub result, src, srcin
>> fmov synd, dend
>> #ifndef __AARCH64EB__
>> rbit synd, synd
>> #endif
>> + add result, result, 16
>> clz tmp, synd
>> add result, result, tmp, lsr 2
>> ret
>>
next prev parent reply other threads:[~2023-03-02 11:49 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-01-12 15:53 Wilco Dijkstra
2023-01-13 12:26 ` Szabolcs Nagy
2023-03-02 11:49 ` Cupertino Miranda [this message]
2023-03-02 13:07 ` Wilco Dijkstra
2023-03-02 14:29 ` Cupertino Miranda
-- strict thread matches above, loose matches on Subject: below --
2015-04-15 12:34 [PATCH][AArch64] " Wilco Dijkstra
2015-02-27 15:06 Wilco Dijkstra
2015-01-16 17:39 Wilco Dijkstra
2015-07-06 14:42 ` Marcus Shawcroft
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=87lekf9xur.fsf@oracle.com \
--to=cupertino.miranda@oracle.com \
--cc=Szabolcs.Nagy@arm.com \
--cc=Wilco.Dijkstra@arm.com \
--cc=libc-alpha@sourceware.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).