From: "De Lara Guarch, Pablo" <pablo.de.lara.guarch@intel.com>
To: "Richardson, Bruce" <bruce.richardson@intel.com>
Cc: "dev@dpdk.org" <dev@dpdk.org>
Subject: Re: [dpdk-dev] [PATCH] hash: update jhash function with the latest available
Date: Fri, 17 Apr 2015 16:03:44 +0000 [thread overview]
Message-ID: <E115CCD9D858EF4F90C690B0DCB4D89727289E30@IRSMSX108.ger.corp.intel.com> (raw)
In-Reply-To: <20150416140115.GA10552@bricha3-MOBL3>
> -----Original Message-----
> From: Richardson, Bruce
> Sent: Thursday, April 16, 2015 3:01 PM
> To: De Lara Guarch, Pablo
> Cc: dev@dpdk.org
> Subject: Re: [dpdk-dev] [PATCH] hash: update jhash function with the latest
> available
>
> On Thu, Apr 16, 2015 at 02:26:59PM +0100, Pablo de Lara wrote:
> > Jenkins hash function was developed originally in 1996,
> > and was integrated in first versions of DPDK.
> > The function has been improved in 2006,
> > achieving up to 60% better performance, compared to the original one.
> >
> > Check out: http://burtleburtle.net/bob/c/lookup3.c
> >
> > This patch integrates that code in the rte_jhash library,
> > adding also a new function rte_jhash_word2,
> > that returns two different hash values, for a single key.
> >
>
> Should the addition of the new functionality not be a separate patch from
> the
> update to the existing code?
True, actually, I miss one extra function (2 in total).
> Also, do the new functions return the exact same values as the previous
> versions,
> just faster?
The new functions return different values from the previous version
AND faster (some cases, MUCH faster)
[...]
> > +#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
> > + case 12:
> > + c += k[2]; b += k[1]; a += k[0]; break;
> > + case 11:
> > + c += k[2]&0xffffff; b += k[1]; a += k[0]; break;
> > + case 10:
> > + c += k[2]&0xffff; b += k[1]; a += k[0]; break;
> > + case 9:
> > + c += k[2]&0xff; b += k[1]; a += k[0]; break;
> > + case 8:
> > + b += k[1]; a += k[0]; break;
> > + case 7:
> > + b += k[1]&0xffffff; a += k[0]; break;
> > + case 6:
> > + b += k[1]&0xffff; a += k[0]; break;
> > + case 5:
> > + b += k[1]&0xff; a += k[0]; break;
> > + case 4:
> > + a += k[0]; break;
> > + case 3:
> > + a += k[0]&0xffffff; break;
> > + case 2:
> > + a += k[0]&0xffff; break;
> > + case 1:
> > + a += k[0]&0xff; break;
> > +#else
> > + case 12:
> > + c += k[2]; b += k[1]; a += k[0]; break;
> > + case 11:
> > + c += k[2]&0xffffff00; b += k[1]; a += k[0]; break;
> > + case 10:
> > + c += k[2]&0xffff0000; b += k[1]; a += k[0]; break;
> > + case 9:
> > + c += k[2]&0xff000000; b += k[1]; a += k[0]; break;
> > + case 8:
> > + b += k[1]; a += k[0]; break;
> > + case 7:
> > + b += k[1]&0xffffff00; a += k[0]; break;
> > + case 6:
> > + b += k[1]&0xffff0000; a += k[0]; break;
> > + case 5:
> > + b += k[1]&0xff000000; a += k[0]; break;
> > + case 4:
> > + a += k[0]; break;
> > + case 3:
> > + a += k[0]&0xffffff00; break;
> > + case 2:
> > + a += k[0]&0xffff0000; break;
> > + case 1:
> > + a += k[0]&0xff000000; break;
> > +#endif
>
> Only the constants seem different in this block. Can we get rid of the
> #ifdefs using rte_XX_to_cpu() calls instead?
Will add that in next version.
>
> > + /* zero length strings require no mixing */
> > + case 0:
> > + return c;
> > + };
> > +#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
> > + } else if ((u.i & 0x1) == 0) {
> > + /* read 16-bit chunks */
> > + const uint16_t *k = (const uint16_t *)key;
> > + const uint8_t *k8;
> > +
> > + /* all but last block: aligned reads and different mixing */
> > + while (length > 12) {
> > + a += k[0] + (((uint32_t)k[1])<<16);
> > + b += k[2] + (((uint32_t)k[3])<<16);
> > + c += k[4] + (((uint32_t)k[5])<<16);
> > +
> > + __rte_jhash_mix(a, b, c);
> > +
> > + k += 6;
> > + length -= 12;
> > + }
> > +
> > + /* handle the last (probably partial) block */
> > + k8 = (const uint8_t *)k;
> > + switch (length) {
> > + case 12:
> > + c += k[4]+(((uint32_t)k[5])<<16);
> > + b += k[2]+(((uint32_t)k[3])<<16);
> > + a += k[0]+(((uint32_t)k[1])<<16);
> > + break;
> > + case 11:
> > + /* fall through */
> > + c += ((uint32_t)k8[10])<<16;
> > + case 10:
> > + c += k[4];
> > + b += k[2]+(((uint32_t)k[3])<<16);
> > + a += k[0]+(((uint32_t)k[1])<<16);
> > + break;
> > + case 9:
> > + /* fall through */
> > + c += k8[8];
> > + case 8:
> > + b += k[2]+(((uint32_t)k[3])<<16);
> > + a += k[0]+(((uint32_t)k[1])<<16);
> > + break;
> > + case 7:
> > + /* fall through */
> > + b += ((uint32_t)k8[6])<<16;
> > + case 6:
> > + b += k[2];
> > + a += k[0]+(((uint32_t)k[1])<<16);
> > + break;
> > + case 5:
> > + /* fall through */
> > + b += k8[4];
> > + case 4:
> > + a += k[0]+(((uint32_t)k[1])<<16);
> > + break;
> > + case 3:
> > + /* fall through */
> > + a += ((uint32_t)k8[2])<<16;
> > + case 2:
> > + a += k[0];
> > + break;
> > + case 1:
> > + a += k8[0];
> > + break;
> > + case 0:
> > + /* zero length requires no mixing */
> > + return c;
> > + }
> > +#endif
>
> No else block for this ifdef?
According to the code, for big endian, it only covers 4-byte alignment and rest of cases.
>
> > + } else {
> > + const uint8_t *k = (const uint8_t *)key;
> > +
> > + /* all but the last block: affect some 32 bits of (a, b, c) */
> > + while (length > 12) {
> > +#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
> > + a += k[0];
> > + a += ((uint32_t)k[1])<<8;
> > + a += ((uint32_t)k[2])<<16;
> > + a += ((uint32_t)k[3])<<24;
> > + b += k[4];
> > + b += ((uint32_t)k[5])<<8;
> > + b += ((uint32_t)k[6])<<16;
> > + b += ((uint32_t)k[7])<<24;
> > + c += k[8];
> > + c += ((uint32_t)k[9])<<8;
> > + c += ((uint32_t)k[10])<<16;
> > + c += ((uint32_t)k[11])<<24;
> > +#else
> > + a += ((uint32_t)k[0])<<24;
> > + a += ((uint32_t)k[1])<<16;
> > + a += ((uint32_t)k[2])<<8;
> > + a += ((uint32_t)k[3]);
> > + b += ((uint32_t)k[4])<<24;
> > + b += ((uint32_t)k[5])<<16;
> > + b += ((uint32_t)k[6])<<8;
> > + b += ((uint32_t)k[7]);
> > + c += ((uint32_t)k[8])<<32;
> > + c += ((uint32_t)k[9])<<16;
> > + c += ((uint32_t)k[10])<<8;
> > + c += ((uint32_t)k[11]);
> > +#endif
>
> Maybe find a better way to shorten/remove this #ifdef also. E.g. shorter
> ifdef defining macros for the different shift amounts, 0, 8, 16, 24.
Agree. Will change that in v2.
[...]
Thanks for the comments. I will send a v2 soon.
next prev parent reply other threads:[~2015-04-17 16:03 UTC|newest]
Thread overview: 62+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-04-16 13:26 Pablo de Lara
2015-04-16 14:01 ` Bruce Richardson
2015-04-17 16:03 ` De Lara Guarch, Pablo [this message]
2015-04-24 11:23 ` [dpdk-dev] [PATCH v2 0/6] update jhash function Pablo de Lara
2015-04-24 11:23 ` [dpdk-dev] [PATCH v2 1/6] test/hash: move hash function perf tests to separate file Pablo de Lara
2015-04-24 11:23 ` [dpdk-dev] [PATCH v2 2/6] test/hash: improve accuracy on cycle measurements Pablo de Lara
2015-04-24 11:23 ` [dpdk-dev] [PATCH v2 3/6] hash: update jhash function with the latest available Pablo de Lara
2015-04-24 11:23 ` [dpdk-dev] [PATCH v2 4/6] hash: add two new functions to jhash library Pablo de Lara
2015-04-24 11:23 ` [dpdk-dev] [PATCH v2 5/6] hash: remove duplicated code Pablo de Lara
2015-04-24 11:23 ` [dpdk-dev] [PATCH v2 6/6] hash: rename rte_jhash2 to rte_jhash_32b Pablo de Lara
2015-05-05 14:43 ` [dpdk-dev] [PATCH v3 0/6] update jhash function Pablo de Lara
2015-05-05 14:43 ` [dpdk-dev] [PATCH v3 1/6] test/hash: move hash function perf tests to separate file Pablo de Lara
2015-05-05 14:43 ` [dpdk-dev] [PATCH v3 2/6] test/hash: improve accuracy on cycle measurements Pablo de Lara
2015-05-05 14:43 ` [dpdk-dev] [PATCH v3 3/6] hash: update jhash function with the latest available Pablo de Lara
2015-05-06 0:35 ` Ananyev, Konstantin
2015-05-06 9:36 ` De Lara Guarch, Pablo
2015-05-06 16:11 ` Ananyev, Konstantin
2015-05-07 11:11 ` Ananyev, Konstantin
2015-05-05 14:43 ` [dpdk-dev] [PATCH v3 4/6] hash: add two new functions to jhash library Pablo de Lara
2015-05-05 14:43 ` [dpdk-dev] [PATCH v3 5/6] hash: remove duplicated code Pablo de Lara
2015-05-05 14:43 ` [dpdk-dev] [PATCH v3 6/6] hash: rename rte_jhash2 to rte_jhash_32b Pablo de Lara
2015-05-12 11:02 ` [dpdk-dev] [PATCH v4 0/6] update jhash function Pablo de Lara
2015-05-12 11:02 ` [dpdk-dev] [PATCH v4 1/6] test/hash: move hash function perf tests to separate file Pablo de Lara
2015-05-12 11:02 ` [dpdk-dev] [PATCH v4 2/6] test/hash: improve accuracy on cycle measurements Pablo de Lara
2015-05-12 11:02 ` [dpdk-dev] [PATCH v4 3/6] hash: update jhash function with the latest available Pablo de Lara
2015-05-12 11:02 ` [dpdk-dev] [PATCH v4 4/6] hash: add two new functions to jhash library Pablo de Lara
2015-05-12 11:02 ` [dpdk-dev] [PATCH v4 5/6] hash: remove duplicated code Pablo de Lara
2015-05-12 11:02 ` [dpdk-dev] [PATCH v4 6/6] hash: rename rte_jhash2 to rte_jhash_32b Pablo de Lara
2015-05-12 15:33 ` [dpdk-dev] [PATCH v4 0/6] update jhash function Neil Horman
2015-05-13 13:52 ` De Lara Guarch, Pablo
2015-05-13 14:20 ` Neil Horman
2015-05-18 16:14 ` Bruce Richardson
2015-05-22 10:16 ` [dpdk-dev] [PATCH v5 00/10] " Pablo de Lara
2015-05-22 10:16 ` [dpdk-dev] [PATCH v5 01/10] test/hash: move hash function perf tests to separate file Pablo de Lara
2015-05-22 10:16 ` [dpdk-dev] [PATCH v5 02/10] test/hash: improve accuracy on cycle measurements Pablo de Lara
2015-05-22 10:16 ` [dpdk-dev] [PATCH v5 03/10] test/hash: update key size range and initial values for testing Pablo de Lara
2015-05-22 10:16 ` [dpdk-dev] [PATCH v5 04/10] test/hash: change order of loops in hash function tests Pablo de Lara
2015-06-10 11:05 ` Bruce Richardson
2015-05-22 10:16 ` [dpdk-dev] [PATCH v5 05/10] test/hash: add new functional tests for hash functions Pablo de Lara
2015-05-22 10:16 ` [dpdk-dev] [PATCH v5 06/10] hash: update jhash function with the latest available Pablo de Lara
2015-06-10 11:07 ` Bruce Richardson
2015-05-22 10:16 ` [dpdk-dev] [PATCH v5 07/10] hash: add two new functions to jhash library Pablo de Lara
2015-05-22 10:16 ` [dpdk-dev] [PATCH v5 08/10] hash: remove duplicated code Pablo de Lara
2015-05-22 10:16 ` [dpdk-dev] [PATCH v5 09/10] hash: rename rte_jhash2 to rte_jhash_32b Pablo de Lara
2015-06-10 11:09 ` Bruce Richardson
2015-05-22 10:16 ` [dpdk-dev] [PATCH v5 10/10] test/hash: verify rte_jhash_1word/2words/3words Pablo de Lara
2015-06-10 15:25 ` [dpdk-dev] [PATCH v6 00/10] update jhash function Pablo de Lara
2015-06-10 15:25 ` [dpdk-dev] [PATCH v6 01/10] test/hash: move hash function perf tests to separate file Pablo de Lara
2015-06-10 15:25 ` [dpdk-dev] [PATCH v6 02/10] test/hash: improve accuracy on cycle measurements Pablo de Lara
2015-06-10 15:25 ` [dpdk-dev] [PATCH v6 03/10] test/hash: update key size range and initial values for testing Pablo de Lara
2015-06-10 15:25 ` [dpdk-dev] [PATCH v6 04/10] test/hash: change order of loops in hash function tests Pablo de Lara
2015-06-10 15:25 ` [dpdk-dev] [PATCH v6 05/10] test/hash: add new functional tests for hash functions Pablo de Lara
2015-06-10 15:25 ` [dpdk-dev] [PATCH v6 06/10] hash: update jhash function with the latest available Pablo de Lara
2015-06-10 15:25 ` [dpdk-dev] [PATCH v6 07/10] hash: add two new functions to jhash library Pablo de Lara
2015-06-10 15:25 ` [dpdk-dev] [PATCH v6 08/10] hash: remove duplicated code Pablo de Lara
2015-06-16 9:33 ` Thomas Monjalon
2015-06-16 10:31 ` De Lara Guarch, Pablo
2015-06-16 13:08 ` Thomas Monjalon
2015-06-10 15:25 ` [dpdk-dev] [PATCH v6 09/10] hash: rename rte_jhash2 to rte_jhash_32b Pablo de Lara
2015-06-10 15:25 ` [dpdk-dev] [PATCH v6 10/10] test/hash: verify rte_jhash_1word/2words/3words Pablo de Lara
2015-06-12 10:37 ` [dpdk-dev] [PATCH v6 00/10] update jhash function Bruce Richardson
2015-06-16 10:22 ` 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=E115CCD9D858EF4F90C690B0DCB4D89727289E30@IRSMSX108.ger.corp.intel.com \
--to=pablo.de.lara.guarch@intel.com \
--cc=bruce.richardson@intel.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).