Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 11 Jul 2013 08:49:03 -0400
From: Rich Felker <dalias@...ifal.cx>
To: musl@...ts.openwall.com
Subject: Re: Thinking about release

On Thu, Jul 11, 2013 at 07:27:11AM +0200, Daniel Cegiełka wrote:
> 2013/7/11 Rich Felker <dalias@...ifal.cx>:
> > On Thu, Jul 11, 2013 at 10:44:16AM +1200, Andre Renaud wrote:
> 
> > Yes, I suspect so. One slightly crazy idea I had was to write the
> > function in C with just inline asm for the inner ldm/stm loop.
> 
> A bit of useful code (x86):
> 
> http://dpdk.org/browse/dpdk/tree/lib/librte_eal/common/include/rte_memcpy.h

On modern x86 (32-bit), this is slower than even the naive "rep movsb"
version. Some x86 chips have problems with rep movsb, so the version
in musl does a little bit more work (possibly more than it needs to)
to use "rep movsd".

On x86_64, there _may_ be faster approaches than the "rep movsq" we
have right now, but so far my impression is that they don't work on
baseline x86_64 (only later variants) and don't gain much.

Rich

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.