Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Tue, 26 Jun 2012 17:09:55 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: build error for linux-x86-native (c3:fmt.c): undefined reference to `crypt'

On Tue, Jun 26, 2012 at 10:20:31AM +0200, magnum wrote:
> Actually we could just drop this make target. If you have to resort to a 
> 32-bit build, you don't have XOP or AVX anyway

Not necessarily.  Some people do run 32-bit operating systems (including
in VMs) on those recent machines.  Also, we don't even support 64-bit
builds for Windows yet (...but you're talking about a Linux only make
target here, so that's irrelevant).

> so linux-x86-sse2i is a better target.

Usually, yes.

I don't mind dropping linux-x86-native.  That may reduce confusion since
linux-x86-native is currently "marketed" as the best option, but it
usually is not.

> I was unsure about Atom, but googling "intel +atom +xop" 
> or "intel +atom +avx" both result in zero hits.

Yes, and besides XOP is an AMD-only thing.

BTW, both AVX and XOP are slow (like twice slower) with 32-bit builds on
Bulldozer, with SSE2 being faster in those builds.  There appears to be
a huge performance hit with VEX-encoded instructions in 32-bit mode on
Bulldozer.  I guess there was some kind of tradeoff that they resolved
in this way.

On Intel's CPUs, AVX is actually faster than SSE2 even in 32-bit builds.

Alexander

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.