[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 19 May 2011 18:32:33 -0300
From: Yuri Gonzaga <yuriggc@...il.com>
To: crypt-dev@...ts.openwall.com
Subject: Re: alternative approach
Hi!
I did the first attempt. I think I spent more time than I expected mainly
because the adaptation to Xilinx dev environment.
It is composed of a state machine to read the "n" input, calculate the
rounds and then output "l", "r" and "s" (byte-by-byte).
I put a directive `define to parameterize how much rounds are done per
single clock cycle.
The state machine has 8 states, but only 2 are related to rounds
calculation.
For NROUNDS = 2, synthesis results (Target Device: Virtex6
xc6vlx75t-3ff484):
Device Utilization Summary (estimated values)
[-] <?&ExpandedTable=DeviceUtilizationSummary(estimatedvalues)>
Logic Utilization
Used
Available
Utilization
Number of Slice Registers
35
93120
0%
Number of Slice LUTs
131
46560
0%
Number of fully used LUT-FF pairs
35
131
26%
Number of bonded IOBs
14
240
5%
Number of BUFG/BUFGCTRLs
2
32
6%
Maximum frequency: 261.356MHz
For NROUNDS = 4 (same target device):
Device Utilization Summary (estimated values)
[-] <?&ExpandedTable=DeviceUtilizationSummary(estimatedvalues)>
Logic Utilization
Used
Available
Utilization
Number of Slice Registers
35
93120
0%
Number of Slice LUTs
183
46560
0%
Number of fully used LUT-FF pairs
35
183
19%
Number of bonded IOBs
14
240
5%
Number of BUFG/BUFGCTRLs
2
32
6%
Maximum frequency: 147.432MHz
About the pipelining, how can we deal with the fact that there are
dependencies between r, l and s in calculations?
Will each stage have to store locally r, l and s?
The verilog code is attached, including simulation.
Regards,
Yuri Gonzaga
[ CONTENT OF TYPE text/html SKIPPED ]
[ CONTENT OF TYPE application/octet-stream SKIPPED ]
[ CONTENT OF TYPE application/octet-stream SKIPPED ]
Powered by blists - more mailing lists
Powered by Openwall GNU/*/Linux -
Powered by OpenVZ
