Optimized hash routines needed!

Corni · **Joined:** 15 Nov 2008, 20:45 **Posts:** 34

jci wrote:

neinbrucke wrote:

lol, you don't get valid hashes with md4 at all? or just not while using sse2 ?

The "single" is doing valid hashes, just not the SSE "quad" version.
I am sure it's some idiot simple thing, i'm just having trouble finding it :cry:

I needed >3 hours to find out that a shift is not a rotate while I implemented it, and that you've implement the roll at your own (use the rightshift-function which shifts in zeros, not the sign bit!)

jci · **Joined:** 05 Nov 2007, 01:55 **Posts:** 103

Corni wrote:

jci wrote:

neinbrucke wrote:

lol, you don't get valid hashes with md4 at all? or just not while using sse2 ?

The "single" is doing valid hashes, just not the SSE "quad" version.
I am sure it's some idiot simple thing, i'm just having trouble finding it :cry:

I needed >3 hours to find out that a shift is not a rotate while I implemented it, and that you've implement the roll at your own (use the rightshift-function which shifts in zeros, not the sign bit!)

Yeah, SSE doesn't have packed rotates until AVX i think :evil:

here's my unfolded ROL: (uses a tmp var i have laying around)

Code: Select all

// ((X << s) | (X >> (32-s)))
#define ROL(X, s, tmp) \
   tmp = X;  X = psrld(X, 32 - s); tmp = pslld(tmp, s); X = por(X, tmp);

jci · **Joined:** 05 Nov 2007, 01:55 **Posts:** 103

Just thought i would share my own versions of MD4/MD5, see the bottom of this thread for links.

AFAIK, i don't see any (general purpose, x86, non-SSE) implementation being significantly faster than those for any given length

dzhugashvili · **Joined:** 20 Jul 2009, 02:02 **Posts:** 1

jci, fast_md5, fast_md4, and NEW_MD4 are not SSE2-capable, is it? I keep seeing your benchmarks for quad and octo SSE2-accelerated MD5 reference, but where can I get the source code for it?

GenTiradentes · **Joined:** 19 May 2010, 02:22 **Posts:** 1

I'm new to this forum, but I've been working on a cracker of my own for some time now.

I have three versions of the MD4/NTLM algorithm written, one which is plain C, one which is SSE2 accelerated, and one which is MMX accelerated. (MMX acceleration is functional, but immature, and not well-optimized yet.) The first two are rather fast, and will be further improved in the future. My code is GPL v3 licensed. All that I ask is that if you use my code, you tell me what you're doing with it, (I'm curious) and if the code is mostly my own, that you credit me somewhere. (Obviously, if you decide to use any of my code, your usage must comply with the GPL v3 license.)

My code can be found at http://lavernasbrute.googlecode.com

I like to keep the trunk relatively stable, so it's probably not quite as fast as the latest version. I have an unstable branch which has the latest revisions and performance improvements.

The actual hashing code can be found in (BRANCH_NAME)/source/hashing

Cheers.

Optimized hash routines needed!

Who is online