public inbox for netdev@vger.kernel.org 
 help / color / mirror / Atom feed
From: Ingo Molnar <mingo@kernel•org>
To: Pavel Machek <pavel@ucw•cz>
Cc: Thomas Gleixner <tglx@linutronix•de>,
	David Laight <David.Laight@ACULAB•COM>,
	'Rahul Lakkireddy' <rahul.lakkireddy@chelsio•com>,
	"x86@kernel•org" <x86@kernel•org>,
	"linux-kernel@vger•kernel.org" <linux-kernel@vger•kernel.org>,
	"netdev@vger•kernel.org" <netdev@vger•kernel.org>,
	"mingo@redhat•com" <mingo@redhat•com>,
	"hpa@zytor•com" <hpa@zytor•com>,
	"davem@davemloft•net" <davem@davemloft•net>,
	"akpm@linux-foundation•org" <akpm@linux-foundation•org>,
	"torvalds@linux-foundation•org" <torvalds@linux-foundation•org>,
	"ganeshgr@chelsio•com" <ganeshgr@chelsio•com>,
	"nirranjan@chelsio•com" <nirranjan@chelsio•com>,
	"indranil@chelsio•com" <indranil@chelsio•com>,
	Andy Lutomirski <luto@kernel•org>,
	Peter Zijlstra <a.p.zijlstra@chello•nl>,
	Fenghua Yu <fenghua.yu@intel•com
Subject: Re: [RFC PATCH 0/3] kernel: add support for 256-bit IO access
Date: Tue, 3 Apr 2018 12:36:55 +0200	[thread overview]
Message-ID: <20180403103655.oa235p3h65twf4ct@gmail.com> (raw)
In-Reply-To: <20180403084932.GA3926@amd>


* Pavel Machek <pavel@ucw•cz> wrote:

> > > > Yeah, so generic memcpy() replacement is only feasible I think if the most 
> > > > optimistic implementation is actually correct:
> > > > 
> > > >  - if no preempt disable()/enable() is required
> > > > 
> > > >  - if direct access to the AVX[2] registers does not disturb legacy FPU state in 
> > > >    any fashion
> > > > 
> > > >  - if direct access to the AVX[2] registers cannot raise weird exceptions or have
> > > >    weird behavior if the FPU control word is modified to non-standard values by 
> > > >    untrusted user-space
> > > > 
> > > > If we have to touch the FPU tag or control words then it's probably only good for 
> > > > a specialized API.
> > > 
> > > I did not mean to have a general memcpy replacement. Rather something like
> > > magic_memcpy() which falls back to memcpy when AVX is not usable or the
> > > length does not justify the AVX stuff at all.
> > 
> > OK, fair enough.
> > 
> > Note that a generic version might still be worth trying out, if and only if it's 
> > safe to access those vector registers directly: modern x86 CPUs will do their 
> > non-constant memcpy()s via the common memcpy_erms() function - which could in 
> > theory be an easy common point to be (cpufeatures-) patched to an AVX2 variant, if 
> > size (and alignment, perhaps) is a multiple of 32 bytes or so.
> 
> How is AVX2 supposed to help the memcpy speed?
> 
> If the copy is small, constant overhead will dominate, and I don't
> think AVX2 is going to be win there.

There are several advantages:

1)

"REP; MOVS" (also called ERMS) has a significant constant "setup cost".

In the scheme I suggested (and if it's possible) then single-register AVX2 access 
on the other hand has a setup cost on the "few cycles" order of magnitude.

2)

AVX2 have various non-temporary load and store behavioral variants - while "REP; 
MOVS" doesn't (or rather, any such caching optimizations, to the extent they 
exist, are hidden in the microcode).

> If the copy is big, well, the copy loop will likely run out of L1 and maybe even 
> out of L2, and at that point speed of the loop does not matter because memory is 
> slow...?

In many cases "memory" will be something very fast, such as another level of 
cache. Also, on NUMA "memory" can also be something locally wired to the CPU - 
again accessible at ridiculous bandwidths.

Nevertheless ERMS is probably wins for the regular bulk memcpy by a few percentage 
points, so I don't think AVX2 is a win in the generic large-memcpy case, as long 
as continued caching of both the loads and the stores is beneficial.

Thanks,

	Ingo

  reply	other threads:[~2018-04-03 10:36 UTC|newest]

Thread overview: 47+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-03-19 14:20 [RFC PATCH 0/3] kernel: add support for 256-bit IO access Rahul Lakkireddy
2018-03-19 14:20 ` [RFC PATCH 1/3] include/linux: add 256-bit IO accessors Rahul Lakkireddy
2018-03-19 14:20 ` [RFC PATCH 2/3] x86/io: implement 256-bit IO read and write Rahul Lakkireddy
2018-03-19 14:43   ` Thomas Gleixner
2018-03-20 13:32     ` Rahul Lakkireddy
2018-03-20 13:44       ` Andy Shevchenko
2018-03-21 12:27         ` Rahul Lakkireddy
2018-03-20 14:40       ` David Laight
2018-03-21 12:28         ` Rahul Lakkireddy
2018-03-20 14:42       ` Alexander Duyck
2018-03-21 12:28         ` Rahul Lakkireddy
2018-03-22  1:26         ` Linus Torvalds
2018-03-22 10:48           ` David Laight
2018-03-22 17:16             ` Linus Torvalds
2018-03-19 14:20 ` [RFC PATCH 3/3] cxgb4: read on-chip memory 256-bits at a time Rahul Lakkireddy
2018-03-19 14:53 ` [RFC PATCH 0/3] kernel: add support for 256-bit IO access David Laight
2018-03-19 15:05   ` Thomas Gleixner
2018-03-19 15:19     ` David Laight
2018-03-19 15:37       ` Thomas Gleixner
2018-03-19 15:53         ` David Laight
2018-03-19 16:29           ` Linus Torvalds
2018-03-20  8:26         ` Ingo Molnar
2018-03-20  8:38           ` Thomas Gleixner
2018-03-20  9:08             ` Ingo Molnar
2018-03-20  9:41               ` Thomas Gleixner
2018-03-20  9:59                 ` David Laight
2018-03-20 10:54                 ` Ingo Molnar
2018-03-20 13:30                   ` David Laight
2018-04-03  8:49                   ` Pavel Machek
2018-04-03 10:36                     ` Ingo Molnar [this message]
2018-03-20 14:57           ` Andy Lutomirski
2018-03-20 15:10             ` David Laight
2018-03-21  0:39               ` Andy Lutomirski
2018-03-20 18:01           ` Linus Torvalds
2018-03-21  6:32             ` Ingo Molnar
2018-03-21 15:45               ` Andy Lutomirski
2018-03-22  9:36                 ` Ingo Molnar
2018-03-21  7:46             ` Ingo Molnar
2018-03-21 18:15               ` Linus Torvalds
2018-03-22  9:33                 ` Ingo Molnar
2018-03-22 17:40                   ` Alexei Starovoitov
2018-03-22 17:44                     ` Andy Lutomirski
2018-03-22 10:35                 ` David Laight
2018-03-22 12:48                   ` David Laight
2018-03-22 17:07                     ` Linus Torvalds
2018-03-19 15:27 ` Christoph Hellwig
2018-03-20 13:45   ` Rahul Lakkireddy

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=20180403103655.oa235p3h65twf4ct@gmail.com \
    --to=mingo@kernel$(echo .)org \
    --cc=David.Laight@ACULAB$(echo .)COM \
    --cc=a.p.zijlstra@chello$(echo .)nl \
    --cc=akpm@linux-foundation$(echo .)org \
    --cc=davem@davemloft$(echo .)net \
    --cc=fenghua.yu@intel$(echo .)com \
    --cc=ganeshgr@chelsio$(echo .)com \
    --cc=hpa@zytor$(echo .)com \
    --cc=indranil@chelsio$(echo .)com \
    --cc=linux-kernel@vger$(echo .)kernel.org \
    --cc=luto@kernel$(echo .)org \
    --cc=mingo@redhat$(echo .)com \
    --cc=netdev@vger$(echo .)kernel.org \
    --cc=nirranjan@chelsio$(echo .)com \
    --cc=pavel@ucw$(echo .)cz \
    --cc=rahul.lakkireddy@chelsio$(echo .)com \
    --cc=tglx@linutronix$(echo .)de \
    --cc=torvalds@linux-foundation$(echo .)org \
    --cc=x86@kernel$(echo .)org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html
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