From: Ravi Kerur <rkerur@gmail.com>
To: dev@dpdk.org
Subject: [dpdk-dev] [PATCH v1 0/2] rte_memcmp functions
Date: Mon, 7 Mar 2016 14:59:43 -0800 [thread overview]
Message-ID: <1457391583-29604-1-git-send-email-rkerur@gmail.com> (raw)
This patch provides AVX/SSE based memcmp implementation on x86. For
other architectures supported by DPDK, rte_memcmp simply uses memcmp
function.
Following are preliminary performance numbers on Intel(R) Xeon(R) CPU E5-2620 v3 @ 2.40GHz
RTE>>memcmp_perf_autotest
*** RTE memcmp equal performance test results ***
*** Length (bytes), Ticks/Op. ***
*** 2, 4.8526 ***
*** 5, 5.4023 ***
*** 8, 4.5067 ***
*** 9, 5.4024 ***
*** 15, 7.2069 ***
*** 16, 4.5027 ***
*** 17, 4.5020 ***
*** 31, 4.5020 ***
*** 32, 4.5033 ***
*** 33, 5.1377 ***
*** 63, 6.9069 ***
*** 64, 6.9472 ***
*** 65, 9.6301 ***
*** 127, 13.5122 ***
*** 128, 10.8028 ***
*** 129, 11.7058 ***
*** 191, 14.4105 ***
*** 192, 14.4251 ***
*** 193, 16.2139 ***
*** 255, 18.0125 ***
*** 256, 17.1150 ***
*** 257, 18.9129 ***
*** 319, 20.7148 ***
*** 320, 20.7161 ***
*** 321, 22.5198 ***
*** 383, 24.3169 ***
*** 384, 22.5195 ***
*** 385, 24.3197 ***
*** 447, 26.1171 ***
*** 448, 26.1289 ***
*** 449, 27.9168 ***
*** 511, 29.7252 ***
*** 512, 29.7202 ***
*** 513, 27.9253 ***
*** 767, 38.7506 ***
*** 768, 36.9327 ***
*** 769, 38.7259 ***
*** 1023, 49.5368 ***
*** 1024, 49.5347 ***
*** 1025, 46.8414 ***
*** 1522, 68.4517 ***
*** 1536, 68.4522 ***
*** 1600, 67.5478 ***
*** 2048, 87.3674 ***
*** 2560, 106.2776 ***
*** 3072, 125.1937 ***
*** 3584, 144.1503 ***
*** 4096, 163.0243 ***
*** 4608, 181.9367 ***
*** 5632, 219.7613 ***
*** 6144, 238.6745 ***
*** 6656, 257.6009 ***
*** 7168, 276.5084 ***
*** 7680, 295.4162 ***
*** 8192, 314.3726 ***
*** 16834, 746.1065 ***
*** memcmp equal performance test results ***
*** Length (bytes), Ticks/Op. ***
*** 2, 9.0100 ***
*** 5, 8.1065 ***
*** 8, 9.1944 ***
*** 9, 9.0044 ***
*** 15, 9.0084 ***
*** 16, 10.0695 ***
*** 17, 9.0109 ***
*** 31, 9.9111 ***
*** 32, 9.9085 ***
*** 33, 9.9112 ***
*** 63, 12.6098 ***
*** 64, 12.6106 ***
*** 65, 12.6060 ***
*** 127, 19.8160 ***
*** 128, 19.8145 ***
*** 129, 20.7260 ***
*** 191, 26.1214 ***
*** 192, 26.1195 ***
*** 193, 26.1158 ***
*** 255, 30.6222 ***
*** 256, 30.6267 ***
*** 257, 31.5270 ***
*** 319, 36.0264 ***
*** 320, 36.0497 ***
*** 321, 36.9247 ***
*** 383, 40.5290 ***
*** 384, 40.5265 ***
*** 385, 41.4331 ***
*** 447, 45.9317 ***
*** 448, 45.9324 ***
*** 449, 45.9302 ***
*** 511, 50.4652 ***
*** 512, 50.4379 ***
*** 513, 51.3361 ***
*** 767, 67.5552 ***
*** 768, 67.5464 ***
*** 769, 67.5462 ***
*** 1023, 85.5579 ***
*** 1024, 85.5610 ***
*** 1025, 85.5582 ***
*** 1522, 120.6860 ***
*** 1536, 121.6064 ***
*** 1600, 126.1075 ***
*** 2048, 157.6208 ***
*** 2560, 208.8309 ***
*** 3072, 241.7587 ***
*** 3584, 276.1556 ***
*** 4096, 310.5865 ***
*** 4608, 343.8918 ***
*** 5632, 411.2264 ***
*** 6144, 445.3057 ***
*** 6656, 480.4620 ***
*** 7168, 512.5769 ***
*** 7680, 547.9394 ***
*** 8192, 582.7687 ***
*** 16834, 1456.4280 ***
*** RTE memcmp greater than performance test results ***
*** Length (bytes), Ticks/Op. ***
*** 1, 22.5862 ***
*** 8, 24.9140 ***
*** 15, 25.3942 ***
*** 16, 22.1721 ***
*** 32, 24.1650 ***
*** 64, 25.0849 ***
*** 128, 26.5515 ***
*** 256, 28.7055 ***
*** 512, 35.2811 ***
*** 1024, 44.4520 ***
*** 2048, 64.1331 ***
*** 4096, 103.9949 ***
*** 8192, 184.8077 ***
*** 16384, 345.6785 ***
*** memcmp greater than performance test results ***
*** Length (bytes), Ticks/Op. ***
*** 1, 22.6340 ***
*** 8, 25.5552 ***
*** 15, 25.4223 ***
*** 16, 25.1371 ***
*** 32, 26.7381 ***
*** 64, 27.4521 ***
*** 128, 29.7323 ***
*** 256, 35.8891 ***
*** 512, 46.0419 ***
*** 1024, 101.1564 ***
*** 2048, 159.8415 ***
*** 4096, 230.2136 ***
*** 8192, 366.2912 ***
*** 16384, 647.0217 ***
*** RTE memcmp less than performance test results ***
*** Length (bytes), Ticks/Op. ***
*** 1, 22.6627 ***
*** 8, 26.2665 ***
*** 15, 26.8192 ***
*** 16, 21.7960 ***
*** 32, 23.9878 ***
*** 64, 24.2074 ***
*** 128, 26.8111 ***
*** 256, 28.3444 ***
*** 512, 34.7882 ***
*** 1024, 44.4824 ***
*** 2048, 63.4154 ***
*** 4096, 101.4360 ***
*** 8192, 179.1029 ***
*** 16384, 333.9357 ***
*** memcmp less than performance test results ***
*** Length (bytes), Ticks/Op. ***
*** 1, 22.2894 ***
*** 8, 24.9805 ***
*** 15, 24.8632 ***
*** 16, 24.3448 ***
*** 32, 24.8554 ***
*** 64, 25.7541 ***
*** 128, 29.1831 ***
*** 256, 36.2345 ***
*** 512, 45.8233 ***
*** 1024, 103.4597 ***
*** 2048, 163.5588 ***
*** 4096, 232.7368 ***
*** 8192, 368.1143 ***
*** 16384, 649.0326 ***
Test OK
RTE>>quit
Ravi Kerur (2):
rte_memcmp functions using Intel AVX and SSE intrinsics
Test cases for rte_memcmp functions
app/test/Makefile | 31 +-
app/test/autotest_data.py | 19 +
app/test/test_memcmp.c | 250 +++++++
app/test/test_memcmp_perf.c | 396 +++++++++++
.../common/include/arch/arm/rte_memcmp.h | 60 ++
.../common/include/arch/ppc_64/rte_memcmp.h | 62 ++
.../common/include/arch/tile/rte_memcmp.h | 60 ++
.../common/include/arch/x86/rte_memcmp.h | 786 +++++++++++++++++++++
lib/librte_eal/common/include/generic/rte_memcmp.h | 175 +++++
9 files changed, 1838 insertions(+), 1 deletion(-)
create mode 100644 app/test/test_memcmp.c
create mode 100644 app/test/test_memcmp_perf.c
create mode 100644 lib/librte_eal/common/include/arch/arm/rte_memcmp.h
create mode 100644 lib/librte_eal/common/include/arch/ppc_64/rte_memcmp.h
create mode 100644 lib/librte_eal/common/include/arch/tile/rte_memcmp.h
create mode 100644 lib/librte_eal/common/include/arch/x86/rte_memcmp.h
create mode 100644 lib/librte_eal/common/include/generic/rte_memcmp.h
--
1.9.1
next reply other threads:[~2016-03-07 23:00 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-03-07 22:59 Ravi Kerur [this message]
2016-03-07 23:00 ` [dpdk-dev] [PATCH v1 1/2] rte_memcmp functions using Intel AVX and SSE intrinsics Ravi Kerur
2016-03-07 23:00 ` [dpdk-dev] [PATCH v1 2/2] Test cases for rte_memcmp functions Ravi Kerur
2016-05-26 9:05 ` Wang, Zhihong
2016-06-06 18:31 ` Ravi Kerur
2016-06-07 11:09 ` Wang, Zhihong
2017-01-02 20:41 ` Thomas Monjalon
2017-01-09 5:29 ` Wang, Zhihong
2017-01-09 11:08 ` Thomas Monjalon
2017-01-11 1:28 ` Wang, Zhihong
2016-05-25 8:56 ` [dpdk-dev] [PATCH v1 1/2] rte_memcmp functions using Intel AVX and SSE intrinsics Thomas Monjalon
2016-05-26 8:57 ` Wang, Zhihong
2018-12-20 23:30 ` Ferruh Yigit
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=1457391583-29604-1-git-send-email-rkerur@gmail.com \
--to=rkerur@gmail.com \
--cc=dev@dpdk.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 NNTP newsgroup(s).