[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20110613233332.GA24617@openwall.com>
Date: Tue, 14 Jun 2011 03:33:32 +0400
From: Solar Designer <solar@...nwall.com>
To: crypt-dev@...ts.openwall.com
Subject: Re: alternative approach
Yuri -
On Mon, Jun 13, 2011 at 07:55:27PM -0300, Yuri Gonzaga wrote:
> >
> > 2. All four LUT counts are a lot higher than those you had reported for
> > bflike. Are they for something else?
>
> Wasting some time to answer this question, I figured out that the first
> result (lower LUT count) was to the wrong version of verilog code.
> So, the correct result is that with higher LUT count.
This is confusing. Can you please non-ambiguously match code versions
to LUT counts, in one message you post in here? And explain why all
four LUT counts - for different kinds of pcadd() - are several times
higher than what you had reported for bflike before.
We can't work with unreliable, inconsistent, and self-contradictory
results like this.
> I ran again and the results are still the same. So, please, forget
> previous ones.
I have no problem forgetting the previous results, but I need new ones.
> Can you perhaps share the corresponding four pieces of code as well?
>
>
> `define NROUNDS_DIVIDED_BY_2 1
> `define PCADD(a, b, mask) (((a) ^ (b)) + (((a) & (b) & (mask)) << 1))
> //`define PCADD(a, b, mask) a^b^mask
> //`define PCADD(a, b, mask) a^b
> //`define PCADD(a, b, mask) a+b
OK. Perhaps it'd be a good idea not to omit the braces around (a), etc.
This shouldn't matter given your specific uses of PCADD(), but for
reliable testing we need to be making only the changes that we want to
see the effect of.
> (...)
>
> for(i = 0; i < `NROUNDS_DIVIDED_BY_2; i=i+1) begin
> r = r ^ `PCADD(s[l[3:0]],s[16+(l>>4)], 8'h55);
> l = l ^ `PCADD(s[r[3:0]],s[16+(r>>4)], 8'haa);
> end
OK. But I am still confused what entire Verilog source files your four
results applied to, and why they were several times higher than those
you had reported for bflike before.
Summary: please provide reliable results now.
If the LUT counts of 105 (Spartan-6) and 135 (Virtex-6) for bflike were
wrong, then please explain why.
Thanks,
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.
