From: Jianbo Liu <jianbo.liu@linaro.org>
To: "Sekhar, Ashwin" <Ashwin.Sekhar@cavium.com>
Cc: "tomasz.kantecki@intel.com" <tomasz.kantecki@intel.com>,
"Jacob, Jerin" <Jerin.JacobKollanukkaran@cavium.com>,
"dev@dpdk.org" <dev@dpdk.org>
Subject: Re: [dpdk-dev] [PATCH 5/5] examples/l3fwd: add neon support for l3fwd
Date: Wed, 3 May 2017 13:24:39 +0800 [thread overview]
Message-ID: <CAP4Qi3_YFeV33_D5zv0hDBAMtVW0kbzkY0qx18CTC_ygLcNWZw@mail.gmail.com> (raw)
In-Reply-To: <1493725624.3602.34.camel@caviumnetworks.com>
Hi Ashwin,
On 2 May 2017 at 19:47, Sekhar, Ashwin <Ashwin.Sekhar@cavium.com> wrote:
> Hi Jianbo,
>
> I tested your neon changes on thunderx. I am seeing a performance
> regression of ~10% for LPM case and ~20% for EM case with your changes.
> Did you see improvement on any arm64 platform with these changes. If
> yes, how much was the improvement?
Thanks for your reviewing and testing.
For some reason, I have not done much with the performance testing.
I'll send a new version later after tuning the performance.
Thanks!
Jianbo
>
> FYI, I had also tried vectorizing the l3fwd app with neon. Few of the
> optimizations that I can suggest that helped in my case.
>
> * Packet data prefetch is missing in the x86 sse version compared to
> the scalar version (l3fwd_lpm_send_packets vs
> l3fwd_lpm_no_opt_send_packets) . I couldn't understand why this was not
> done in x86. But adding the prefetch was improving performance for
> thunderx.
>
> * Offsets to some packet elements like eth_hdr, ip header, packet type
> etc. are recalculated in different functions. Calculating them once,
> caching them and passing them directly to different functions was
> improving performance.
>
> * There are 3 different loops in l3fwd_lpm_send_packets where we
> iterate over the packets. One each for processx4_step1 and
> processx4_step2 and one in send_packets_multi. Unifying these loops
> were also helping.
>
> Thanks and Regards
> Ashwin
>
next prev parent reply other threads:[~2017-05-03 5:24 UTC|newest]
Thread overview: 62+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-05-02 7:14 [dpdk-dev] [PATCH 1/5] examples/l3fwd: extract arch independent code from multi hash lookup Jianbo Liu
2017-05-02 7:14 ` [dpdk-dev] [PATCH 2/5] examples/l3fwd: rename l3fwd_em_sse.h to l3fwd_em_single.h Jianbo Liu
2017-05-02 9:40 ` Sekhar, Ashwin
2017-05-02 7:14 ` [dpdk-dev] [PATCH 3/5] examples/l3fwd: extract common code from multi packet send Jianbo Liu
2017-05-02 7:14 ` [dpdk-dev] [PATCH 4/5] examples/l3fwd: rearrange the code for lpm_l3fwd Jianbo Liu
2017-05-02 7:14 ` [dpdk-dev] [PATCH 5/5] examples/l3fwd: add neon support for l3fwd Jianbo Liu
2017-05-02 11:20 ` Sekhar, Ashwin
2017-05-02 11:47 ` Sekhar, Ashwin
2017-05-03 5:24 ` Jianbo Liu [this message]
2017-05-04 8:42 ` Jianbo Liu
2017-05-05 4:24 ` Sekhar, Ashwin
2017-05-05 5:43 ` Jianbo Liu
2017-05-09 8:10 ` Sekhar, Ashwin
2017-05-10 2:39 ` Jianbo Liu
2017-05-10 2:30 ` [dpdk-dev] [PATCH v2 0/7] accelerate examples/l3fwd with NEON on ARM64 platform Jianbo Liu
2017-05-10 2:30 ` [dpdk-dev] [PATCH v2 1/7] examples/l3fwd: extract arch independent code from multi hash lookup Jianbo Liu
2017-05-10 2:30 ` [dpdk-dev] [PATCH v2 2/7] examples/l3fwd: rename l3fwd_em_sse.h to l3fwd_em_sequential.h Jianbo Liu
2017-05-10 2:30 ` [dpdk-dev] [PATCH v2 3/7] examples/l3fwd: extract common code from multi packet send Jianbo Liu
2017-05-10 2:30 ` [dpdk-dev] [PATCH v2 4/7] examples/l3fwd: rearrange the code for lpm_l3fwd Jianbo Liu
2017-05-10 2:30 ` [dpdk-dev] [PATCH v2 5/7] examples/l3fwd: add neon support for l3fwd Jianbo Liu
2017-05-10 15:00 ` Sekhar, Ashwin
2017-05-11 3:16 ` Jianbo Liu
2017-05-11 4:14 ` Sekhar, Ashwin
2017-05-11 4:27 ` Sekhar, Ashwin
2017-05-11 6:11 ` Jianbo Liu
2017-05-10 2:30 ` [dpdk-dev] [PATCH v2 6/7] examples/l3fwd: add the times of hash multi-lookup for different Archs Jianbo Liu
2017-05-10 2:30 ` [dpdk-dev] [PATCH v2 7/7] examples/l3fwd: change the guard micro name for header file Jianbo Liu
2017-05-10 11:57 ` Sekhar, Ashwin
2017-05-11 9:25 ` [dpdk-dev] [PATCH v3 0/7] accelerate examples/l3fwd with NEON on ARM64 platform Jianbo Liu
2017-05-11 9:25 ` [dpdk-dev] [PATCH v3 1/7] examples/l3fwd: extract arch independent code from multi hash lookup Jianbo Liu
2017-05-11 9:25 ` [dpdk-dev] [PATCH v3 2/7] examples/l3fwd: rename l3fwd_em_sse.h to l3fwd_em_sequential.h Jianbo Liu
2017-05-11 9:25 ` [dpdk-dev] [PATCH v3 3/7] examples/l3fwd: extract common code from multi packet send Jianbo Liu
2017-05-11 9:25 ` [dpdk-dev] [PATCH v3 4/7] examples/l3fwd: rearrange the code for lpm_l3fwd Jianbo Liu
2017-05-11 9:25 ` [dpdk-dev] [PATCH v3 5/7] examples/l3fwd: add neon support for l3fwd Jianbo Liu
2017-05-11 9:49 ` Sekhar, Ashwin
2017-05-11 10:01 ` Jianbo Liu
2017-05-11 10:27 ` Sekhar, Ashwin
2017-05-12 2:40 ` Jianbo Liu
2017-05-11 9:25 ` [dpdk-dev] [PATCH v3 6/7] examples/l3fwd: add the times of hash multi-lookup for different Archs Jianbo Liu
2017-05-11 9:25 ` [dpdk-dev] [PATCH v3 7/7] examples/l3fwd: change the guard macro name for header file Jianbo Liu
2017-05-15 3:34 ` [dpdk-dev] [PATCH v4 0/8] accelerate examples/l3fwd with NEON on ARM64 platform Jianbo Liu
2017-05-15 3:34 ` [dpdk-dev] [PATCH v4 1/8] examples/l3fwd: extract arch independent code from multi hash lookup Jianbo Liu
2017-05-15 3:34 ` [dpdk-dev] [PATCH v4 2/8] examples/l3fwd: rename l3fwd_em_sse.h to l3fwd_em_sequential.h Jianbo Liu
2017-05-15 3:34 ` [dpdk-dev] [PATCH v4 3/8] examples/l3fwd: extract common code from multi packet send Jianbo Liu
2017-05-15 3:34 ` [dpdk-dev] [PATCH v4 4/8] examples/l3fwd: rearrange the code for lpm_l3fwd Jianbo Liu
2017-05-15 3:34 ` [dpdk-dev] [PATCH v4 5/8] arch/arm: add vcopyq_laneq_u32 for old version of gcc Jianbo Liu
2017-05-15 4:01 ` Jerin Jacob
2017-05-15 3:34 ` [dpdk-dev] [PATCH v4 6/8] examples/l3fwd: add neon support for l3fwd Jianbo Liu
2017-05-15 5:22 ` Sekhar, Ashwin
2017-05-15 3:34 ` [dpdk-dev] [PATCH v4 7/8] examples/l3fwd: add the times of hash multi-lookup for different Archs Jianbo Liu
2017-05-15 3:34 ` [dpdk-dev] [PATCH v4 8/8] examples/l3fwd: change the guard macro name for header file Jianbo Liu
2017-07-03 21:02 ` [dpdk-dev] [PATCH v4 0/8] accelerate examples/l3fwd with NEON on ARM64 platform Thomas Monjalon
2017-07-04 10:23 ` [dpdk-dev] [PATCH v5 " Jianbo Liu
2017-07-04 10:23 ` [dpdk-dev] [PATCH v5 1/8] examples/l3fwd: extract arch independent code from multi hash lookup Jianbo Liu
2017-07-04 10:23 ` [dpdk-dev] [PATCH v5 2/8] examples/l3fwd: rename l3fwd_em_sse.h to l3fwd_em_sequential.h Jianbo Liu
2017-07-04 10:24 ` [dpdk-dev] [PATCH v5 3/8] examples/l3fwd: extract common code from multi packet send Jianbo Liu
2017-07-04 10:24 ` [dpdk-dev] [PATCH v5 4/8] examples/l3fwd: rearrange the code for lpm_l3fwd Jianbo Liu
2017-07-04 10:24 ` [dpdk-dev] [PATCH v5 5/8] arch/arm: add vcopyq_laneq_u32 for old version of gcc Jianbo Liu
2017-07-04 10:24 ` [dpdk-dev] [PATCH v5 6/8] examples/l3fwd: add neon support for l3fwd Jianbo Liu
2017-07-04 10:24 ` [dpdk-dev] [PATCH v5 7/8] examples/l3fwd: add the times of hash multi-lookup for different Archs Jianbo Liu
2017-07-04 10:24 ` [dpdk-dev] [PATCH v5 8/8] examples/l3fwd: change the guard macro name for header file Jianbo Liu
2017-07-04 15:11 ` [dpdk-dev] [PATCH v5 0/8] accelerate examples/l3fwd with NEON on ARM64 platform Thomas Monjalon
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=CAP4Qi3_YFeV33_D5zv0hDBAMtVW0kbzkY0qx18CTC_ygLcNWZw@mail.gmail.com \
--to=jianbo.liu@linaro.org \
--cc=Ashwin.Sekhar@cavium.com \
--cc=Jerin.JacobKollanukkaran@cavium.com \
--cc=dev@dpdk.org \
--cc=tomasz.kantecki@intel.com \
/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 NNTP newsgroup(s).