Site Home   Archive Home   FAQ Home   How to search the Archive   How to Navigate the Archive   
Compare FPGA features and resources   

Threads starting:
1994JulAugSepOctNovDec1994
1995JanFebMarAprMayJunJulAugSepOctNovDec1995
1996JanFebMarAprMayJunJulAugSepOctNovDec1996
1997JanFebMarAprMayJunJulAugSepOctNovDec1997
1998JanFebMarAprMayJunJulAugSepOctNovDec1998
1999JanFebMarAprMayJunJulAugSepOctNovDec1999
2000JanFebMarAprMayJunJulAugSepOctNovDec2000
2001JanFebMarAprMayJunJulAugSepOctNovDec2001
2002JanFebMarAprMayJunJulAugSepOctNovDec2002
2003JanFebMarAprMayJunJulAugSepOctNovDec2003
2004JanFebMarAprMayJunJulAugSepOctNovDec2004
2005JanFebMarAprMayJunJulAugSepOctNovDec2005
2006JanFebMarAprMayJunJulAugSepOctNovDec2006
2007JanFebMarAprMayJunJulAugSepOctNovDec2007
2008JanFebMarAprMayJunJulAugSepOctNovDec2008
2009JanFebMarAprMayJunJulAugSepOctNovDec2009
2010JanFebMarAprMayJunJulAugSepOctNovDec2010
2011JanFebMarAprMayJunJulAugSepOctNovDec2011
2012JanFebMarAprMayJunJulAugSepOctNovDec2012
2013JanFebMarAprMayJunJulAugSepOctNovDec2013
2014JanFebMarAprMayJunJulAugSepOctNovDec2014
2015JanFebMarAprMayJunJulAugSepOctNovDec2015
2016JanFebMarAprMayJunJulAugSepOctNovDec2016
2017JanFebMarAprMayJunJulAugSepOctNovDec2017
2018JanFebMarAprMayJunJulAugSepOctNovDec2018
2019JanFebMarAprMayJunJulAugSepOctNovDec2019
2020JanFebMarAprMay2020

Authors:A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

Custom Search

Messages from 140300

Article: 140300
Subject: Re: FPGAs and Cryptography
From: goouse@twinmail.de
Date: Thu, 7 May 2009 22:57:55 -0700 (PDT)
Links: << >>  << T >>  << A >>
On 7 Mai, 22:13, "Xin Xiao" <s...@spam.com> wrote:
> Hi!
>
> I've heard that FPGAs are good for doing some cryptography... Do any of you
> have experience in cryptography projects and FPGAs?
>
> What tools are available? (better if they are free).
> I suppose I have to code in C/C++ and then synthesize that code to the
> hardware using a tool?
>
> Thanks,

Hi Xin,
yes, cryptographic hardware in FPGAs has the advantage of performance,
compared to any software solution.
There are papers that give some numbers like AES in a 100MHZ clocked
FPGA can acheive about 10 times more data througput than on a 3GHz
Pentium processor. (While consuming less energy!)


Almost any vendor has a free tool to begin with, but they are all
based on VHDL/Verilog.
Digital circuit design is different from programming algorithms. Be
aware of that.
Read some datasheets, to get the idea of FPGAs. Then proceed with
learning a Hardware Description Language.
Then you can try to express your algorithms in hardware.

Have a nice synthesis
  Eilert

Article: 140301
Subject: Help required on Ethernet with FPGA
From: "renupriya" <priya.renu@gmail.com>
Date: Fri, 08 May 2009 01:58:22 -0500
Links: << >>  << T >>  << A >>

Hi ..

 I'm working on a project which involves a design that has NiOS II
processor, DDRSDRAM, SSRAM, a custom Ethernet MAC and other components and
I've to run an application on it.. I use NiOS II embedded Evaluation
board(Cyclone III), quartus 8.1 web edition. 

Since I'm new to Altera, I would like to start with a simple design with
an ethernet MAC , then  send and recieve packets through it. I'm not strong
in Ethernet concepts too. I feel that the design with TSE MAC and socket
server example are too high at this level of me to understand.

 So can anyone send me a design  which has a MAC and also tell me how to
send and recieve packets over it? If possible, can you send me supporting
documents for mastering the concepts of ethernet. 

Thanks and Regards
Renu


Article: 140302
Subject: Re: OpenCores CAN/Ethernet cores
From: Haris <haris36@gmail.com>
Date: Fri, 8 May 2009 00:51:22 -0700 (PDT)
Links: << >>  << T >>  << A >>
On May 8, 12:53=A0am, Jukka Marin <jma...@pyy.embedtronics.fi> wrote:
> Dear All,
>
> I'm wondering whether to use 10/100 Ethernet and CAN controller chips
> (two of each) in a new design or just put cores from OpenCores inside
> an FPGA (which we will need in both cases). =A0I'm (still) new to FPGA's
> and have never used cores from OpenCores. =A0Has anybody used these cores=
?
> Are they reliable and ready for production use?
>
> The third alternative is to buy commercial IP blocks, but I'm afraid
> they might cost more than the old-fashioned way (the production volumes
> will be relatively small).
>
> I'd appreciate any facts and/or opinions. :)
>
> =A0 -jm

You can look at the ehternet cores from MANTARO NETWORKS (http://
www.mantaro.com/services/fpga.htm). They have very good costs for
there cores
as they have developed them from there partners in low costs

Article: 140303
Subject: Re: board with 2 gigabit ethernet connectors?
From: "Sebastien @ Sundance" <maury.sebastien@gmail.com>
Date: Fri, 8 May 2009 02:27:48 -0700 (PDT)
Links: << >>  << T >>  << A >>
On May 7, 12:25=A0pm, recoder <kurtulmeh...@gmail.com> wrote:
> Hi FPGA Gurus,
> =A0I need to get a Gigabit Ethernet message, change its destination
> IP,change some bits in the message(process) and send it out. I figure
> out that I need 2 GigE ports to do that. I don't want to pay for the
> MAC, so the board should be built with the Virtex 5 FPGAs or Altera
> alternatives.
> =A0Does a board like this exists?

Hello,
Yes, this dual GigE board exists at Sundance. Have a look at the
SMT945 from Sundance Multiprocessor Technology? It features a dual
Gigabit Ethernet ports (i.e. 2 GigE ports) and you can use it with the
SMT900 EVM that has a Virtex-5. You would then be able to implement
your logic in the Virtex-5 FPGA.
SMT945 Dual GigE mezzanine card: http://www.sundance.com/web/files/productp=
age.asp?STRFilter=3DSMT945
SMT900 Virtex-5 EVM: http://www.sundance.com/web/files/productpage.asp?STRF=
ilter=3DSMT900

Best Regards,
- Sebastien

Article: 140304
Subject: Re: Help required on Ethernet with FPGA
From: "Symon" <symon_brewer@hotmail.com>
Date: Fri, 8 May 2009 10:55:02 +0100
Links: << >>  << T >>  << A >>
http://www.fpga4fun.com/10BASE-T.html 



Article: 140305
Subject: Re: FPGAs and Cryptography
From: "Sebastien @ Sundance" <maury.sebastien@gmail.com>
Date: Fri, 8 May 2009 02:56:16 -0700 (PDT)
Links: << >>  << T >>  << A >>
On May 7, 9:13=A0pm, "Xin Xiao" <s...@spam.com> wrote:
> Hi!
>
> I've heard that FPGAs are good for doing some cryptography... Do any of y=
ou
> have experience in cryptography projects and FPGAs?
>
> What tools are available? (better if they are free).
> I suppose I have to code in C/C++ and then synthesize that code to the
> hardware using a tool?
>
> Thanks,

You have have a look at the Xilinx application note about encryption:
http://www.xilinx.com/support/documentation/white_papers/wp261.pdf

- Sebastien
www.sundance.com

Article: 140306
Subject: Re: Dual Port RAM Inference
From: Jonathan Bromley <jonathan.bromley@MYCOMPANY.com>
Date: Fri, 08 May 2009 13:42:38 +0100
Links: << >>  << T >>  << A >>
On Thu, 7 May 2009 20:52:47 -0700 (PDT), rickman wrote:

>On May 7, 9:06 pm, Peter Alfke wrote:
>> Here is a short Xilinx tutorial:

>I don't think anyone has a problem understanding how the parts work.
>The issue we are having is how to specify the hardware we want in the
>HDL.

Rick is right, but Peter's explanation is perhaps
timely.  It's comparatively easy to describe both
write-first and read-first behavior in HDL if there
is only one write port, but it's a lot more tricky
(especially in VHDL) if there are two write ports 
with independent clocks.  But the main problem is
as Peter pointed out: there are some corner cases
(for example, simultaneous write to the same location 
from both write ports) whose behavior is ill-defined.
Of course, sensible designers will ensure that
their memory controller never allows those ill-defined
cases to occur.  But you can't easily explain to
your synthesis tool that you have done so.  As a result,
the synth tool must take your HDL description at face
value, and do whatever it takes to ensure that the
finished hardware obeys that description *in all
cases* - hence the nonsense with logic (which 
wouldn't work right anyway) to avoid conflicting
writes to the same address, because your HDL says
that the behavior is well-defined but the memory
doesn't work quite like that.

As promised, I'll take this offline, do some work
on it over the weekend, and report back.  I haven't
yet seen a concise description of the right way to 
cope with this across a range of synth tools, so
it's something I'd like to do anyway.

If all else fails, I can grudgingly go back to
instantiation of wizard-generated components.
But I don't like doing that; it screws up my
design's portability, makes simulation tiresome
and slow, and makes the HDL code butt-ugly.
-- 
Jonathan Bromley, Consultant

DOULOS - Developing Design Know-how
VHDL * Verilog * SystemC * e * Perl * Tcl/Tk * Project Services

Doulos Ltd., 22 Market Place, Ringwood, BH24 1AW, UK
jonathan.bromley@MYCOMPANY.com
http://www.MYCOMPANY.com

The contents of this message may contain personal views which 
are not the views of Doulos Ltd., unless specifically stated.

Article: 140307
Subject: Quartus II negative bus dimensions in Schematic file
From: Chico <jcuello@gmail.com>
Date: Fri, 8 May 2009 05:57:52 -0700 (PDT)
Links: << >>  << T >>  << A >>
This question is Quartus II specific but I figured since this would
the proper place to find at least one person with this problem.

I want to use the fixed point package provided here:
http://www.accellera.org/apps/group_public/documents.php?wg_abbrev=vhdl

I can compile the library and use the files, but a problem arises when
I create a schematic file and try to synthesize. To use the fixed
point numbers I have to declare the bus width like so: var_name sfixed
(1 downto -30)

When I create a block out the file and try to synthesize it, I get an
error about illegal bus width "-30". Is there a way to get around
this? Or I cannot use negative numbers as a bus width?

Thanks

Article: 140308
Subject: Re: Quartus II negative bus dimensions in Schematic file
From: "KJ" <kkjennings@sbcglobal.net>
Date: Fri, 8 May 2009 09:04:22 -0400
Links: << >>  << T >>  << A >>

"Chico" <jcuello@gmail.com> wrote in message 
news:817c83f5-984d-43c7-9d00-9b5e79f2f581@o30g2000vbc.googlegroups.com...
> This question is Quartus II specific but I figured since this would
> the proper place to find at least one person with this problem.
>
> I want to use the fixed point package provided here:
> http://www.accellera.org/apps/group_public/documents.php?wg_abbrev=vhdl
>
> I can compile the library and use the files, but a problem arises when
> I create a schematic file and try to synthesize.

Don't create a 'schematic file' to synthesize.  Just use the VHDL packages 
directly.  I've used them with Quartus and not had any problems.

KJ 



Article: 140309
Subject: Re: Quartus II negative bus dimensions in Schematic file
From: Chico <jcuello@gmail.com>
Date: Fri, 8 May 2009 06:39:08 -0700 (PDT)
Links: << >>  << T >>  << A >>
On May 8, 9:04=A0am, "KJ" <kkjenni...@sbcglobal.net> wrote:
> "Chico" <jcue...@gmail.com> wrote in message
>
> news:817c83f5-984d-43c7-9d00-9b5e79f2f581@o30g2000vbc.googlegroups.com...
>
> > This question is Quartus II specific but I figured since this would
> > the proper place to find at least one person with this problem.
>
> > I want to use the fixed point package provided here:
> >http://www.accellera.org/apps/group_public/documents.php?wg_abbrev=3Dvhd=
l
>
> > I can compile the library and use the files, but a problem arises when
> > I create a schematic file and try to synthesize.
>
> Don't create a 'schematic file' to synthesize. =A0Just use the VHDL packa=
ges
> directly. =A0I've used them with Quartus and not had any problems.
>
> KJ

I was thinking of doing that, but I need to assign a PIN to my inputs/
outputs. Is there a way to do that without using a schematic file
through VHDL? I am using Quartus II 8.0.

Article: 140310
Subject: Re: FPGAs and Cryptography
From: Allan Herriman <allanherriman@hotmail.com>
Date: 08 May 2009 14:10:20 GMT
Links: << >>  << T >>  << A >>
On Thu, 07 May 2009 22:13:52 +0200, Xin Xiao wrote:

> Hi!
> 
> I've heard that FPGAs are good for doing some cryptography... Do any of
> you have experience in cryptography projects and FPGAs?

"Cryptography projects and FPGAs" describes aspects of my job.

FPGAs aren't necessarily the best approach though.  If you are using (or 
breaking!) standard algorithms, then there are off-the-shelf ASICs that 
compete favourably with FPGAs for development time, speed and power.  
E.g. google for Cavium Nitrox.

As an example of performance, some years ago I was able to generate 20Gb/
s of AES256 CTR mode keystream in a Xilinx Virtex2Pro part.  That design 
was I/O limited.  *Much* higher performance would be achievable in a more 
modern part or with a design that wasn't I/O limited.

(BTW, I've done more interesting things since that design, but I can't 
talk about them in a public forum.)

I imagine that past about 100Gb/s, the performance would be limited by 
thermal constraints rather than chip area or clock speed.  Newer FPGA 
families are better in this regard.

> What tools are available? (better if they are free). I suppose I have to
> code in C/C++ and then synthesize that code to the hardware using a
> tool?

I recommend HDLs (e.g. VHDL or Verilog) over languages that weren't 
designed for hardware description (e.g. C/C++).
The tools that come from the FPGA vendors are cheap or free and effective.

You can get open source cores (in VHDL or Verilog) for many of the 
commonly used cryptographic primitives (e.g. block ciphers like AES or 
DES).  I am not going to claim that the open source cores are efficient 
or achieve good performance, but they might be a good place to start.
I haven't found any good free cores for things like GCM or ECC or 
factoring.

Regards,
Allan

Article: 140311
Subject: Re: Can the complex DSP archetecture based on FPGA+DSP be replaced by
From: "Flemming@Sundance" <Flemming.C@sundance.com>
Date: Fri, 8 May 2009 07:24:32 -0700 (PDT)
Links: << >>  << T >>  << A >>
Hi,

The argument of getting a FPGA to replace a DSP is similar to the
=93Chicken & Egg=94 story. Not worth having!

I have on many occasion been involved with potential customers that
were =91sold=92 the idea that a FPGA could do everything and a DSP was not
needed. Similar, the DSP semiconductors suppliers, like TI, Analog
Device, etc. has been praising the level of integration that can be
found on =91Generic=92 DSP device and stated that a FPGA is surplus to
requirement.

My general feeling is that a DSP + FPGA offers everything. The DSP is
cheap, easy to program, runs cool and offers a lot of integration,
whereas the FPGAs are expensive, hard to program (compared to a
processor) =96 but offer much faster performance, support for any kind
of I/O interface, re-programmable, etc. =96 and if you really have a
high-quantity product, then you would surely want to prototype on a
DSP + FPGA, before moving to a custom ASIC device.

A number of suppliers can be found on the Web that offers development
boards with DSP + FPGAs and a growing number of software package can
be bought that makes it possible to use Model-based design for BOTH
the FPGA and the DSP.

 I suggest some R&D and watching YouTube clips.

Happy reading

Rgds

Flemming Christensen



Article: 140312
Subject: Question on using ODDR
From: "lioncat" <dominghang@hotmail.com>
Date: Fri, 08 May 2009 09:38:18 -0500
Links: << >>  << T >>  << A >>
Hello guys, I am working on a virtex 5 and trying to read out data from
internal distributed ROM at 250MHz. Then in order to increase data rate, I
choose use ODDR which is followed by OBUF. But in the MAP I have this kind
of errors at all the output of ODDR as following,

"ERROR:Pack:1569 - The dual data rate register
QPSK_wave/LOOP_DDR_A[6].ODDR_A
   failed to join an OLOGIC component as required."

I don't have any idea to solve this error. Is there someone can give me
some hint?

Another question, I have also tried to use OSERDES which is
parallel-to-serial converter before OBUF. Then the design pass the map but
in place and route it came out the warning as following,
"Route:436 - The router has detected an unroutable situation for one or
more connections. The router will finish the rest of the design and leave
them as unrouted, The cause of this behavior is either an issue with the
placement or unroutable placement constraints."
and occured in all the output pin of OSERDES. How to solve this? 

Is there any other method to increase data rate? When I reached to 500MHz,
the timing problem was always with me in post route simulation. 

Thank you for help!!



Article: 140313
Subject: Re: Dual Port RAM Inference
From: Mike Treseler <mtreseler@gmail.com>
Date: Fri, 08 May 2009 08:20:00 -0700
Links: << >>  << T >>  << A >>
Jonathan Bromley wrote:

> I haven't
> yet seen a concise description of the right way to 
> cope with this across a range of synth tools, so
> it's something I'd like to do anyway.

I long ago gave up trying, so I will be interested
to see your results.

> If all else fails, I can grudgingly go back to
> instantiation of wizard-generated components.
> But I don't like doing that; it screws up my
> design's portability, makes simulation tiresome
> and slow, and makes the HDL code butt-ugly.

I agree with the downside,
but I stick with Peter's "simple dual-port" block rams
which are portable across vendors and tools,
with the same template.

This is all I need for fifos.
I use multiple instances and my own synchronous
controllers to make more complicated memories.

  -- Mike Treseler

Article: 140314
Subject: Re: Question on using ODDR
From: "Symon" <symon_brewer@hotmail.com>
Date: Fri, 8 May 2009 16:25:13 +0100
Links: << >>  << T >>  << A >>
http://lmgtfy.com/?q=instantiate+oddr+virtex+5+site:xilinx.com

;-) 



Article: 140315
Subject: Re: Quartus II negative bus dimensions in Schematic file
From: "Wade Hassler" <wade_hassler@sbcglobal.net>
Date: Fri, 8 May 2009 08:31:13 -0700
Links: << >>  << T >>  << A >>

"Chico" <jcuello@gmail.com> wrote in message 
news:713706ae-f03d-439c-97c5-591e2a804fb1@z19g2000vbz.googlegroups.com...
On May 8, 9:04 am, "KJ" <kkjenni...@sbcglobal.net> wrote:
> "Chico" <jcue...@gmail.com> wrote in message
>
> news:817c83f5-984d-43c7-9d00-9b5e79f2f581@o30g2000vbc.googlegroups.com...
>
> > This question is Quartus II specific but I figured since this would
> > the proper place to find at least one person with this problem.
>
> > I want to use the fixed point package provided here:
> >http://www.accellera.org/apps/group_public/documents.php?wg_abbrev=vhdl
>
> > I can compile the library and use the files, but a problem arises when
> > I create a schematic file and try to synthesize.
>
> Don't create a 'schematic file' to synthesize. Just use the VHDL packages
> directly. I've used them with Quartus and not had any problems.
>
> KJ

I was thinking of doing that, but I need to assign a PIN to my inputs/
outputs. Is there a way to do that without using a schematic file
through VHDL? I am using Quartus II 8.0.


I don't know how VHDL would look, but the following works in Quartus
for Verilog code:

   input   SYNC_IN    /* synthesis altera_chip_pin_lc = "5"*/;
   output  SYNC_OUT   /* synthesis altera_chip_pin_lc = "39"*/;
   input    SW_2      /* synthesis altera_chip_pin_lc = "18"*/;
   input    SW_1      /* synthesis altera_chip_pin_lc = "21"*/;
   output   LED1      /* synthesis altera_chip_pin_lc = "37"*/;
   output   LED2      /* synthesis altera_chip_pin_lc = "22"*/;

Look in the Help file under Synthesis Attributes

Wade Hassler 



Article: 140316
Subject: Re: Question on using ODDR
From: "lioncat" <dominghang@hotmail.com>
Date: Fri, 08 May 2009 10:38:58 -0500
Links: << >>  << T >>  << A >>
>http://lmgtfy.com/?q=instantiate+oddr+virtex+5+site:xilinx.com
>
>;-) 
>
>
>
Thank you for reply. But I have already read the library and the template
of the ODDR. However my problem is in the map. The unit cannot be mapped. I
want to figure out how to solve this situation.

Article: 140317
Subject: Re: Quartus II negative bus dimensions in Schematic file
From: LittleAlex <alex.louie@email.com>
Date: Fri, 8 May 2009 08:42:33 -0700 (PDT)
Links: << >>  << T >>  << A >>
On May 8, 6:39 am, Chico <jcue...@gmail.com> wrote:
> On May 8, 9:04 am, "KJ" <kkjenni...@sbcglobal.net> wrote:
>
>
>
> > "Chico" <jcue...@gmail.com> wrote in message
>
> >news:817c83f5-984d-43c7-9d00-9b5e79f2f581@o30g2000vbc.googlegroups.com...
>
> > > This question is Quartus II specific but I figured since this would
> > > the proper place to find at least one person with this problem.
>
> > > I want to use the fixed point package provided here:
> > >http://www.accellera.org/apps/group_public/documents.php?wg_abbrev=vhdl
>
> > > I can compile the library and use the files, but a problem arises when
> > > I create a schematic file and try to synthesize.
>
> > Don't create a 'schematic file' to synthesize.  Just use the VHDL packages
> > directly.  I've used them with Quartus and not had any problems.
>
> > KJ
>
> I was thinking of doing that, but I need to assign a PIN to my inputs/
> outputs. Is there a way to do that without using a schematic file
> through VHDL? I am using Quartus II 8.0.

Yes.  This is well documented in the manual.  And the tutorial.  And
in several download-able examples.

AL

Article: 140318
Subject: Re: Can the complex DSP archetecture based on FPGA+DSP be replaced by ?FPGA
From: glen herrmannsfeldt <gah@ugcs.caltech.edu>
Date: Fri, 8 May 2009 15:57:02 +0000 (UTC)
Links: << >>  << T >>  << A >>
Flemming@Sundance <Flemming.C@sundance.com> wrote:
 
> The argument of getting a FPGA to replace a DSP is similar to the
> ?Chicken & Egg? story. Not worth having!
 
> I have on many occasion been involved with potential customers that
> were ?sold? the idea that a FPGA could do everything and a DSP was not
> needed. Similar, the DSP semiconductors suppliers, like TI, Analog
> Device, etc. has been praising the level of integration that can be
> found on ?Generic? DSP device and stated that a FPGA is surplus to
> requirement.

If the DSP can't do everything, and you need an FPGA, and the FPGA
with a soft processor can do what is needed, then the FPGA is likely
a good choice.  
 
> My general feeling is that a DSP + FPGA offers everything. The DSP is
> cheap, easy to program, runs cool and offers a lot of integration,

I won't disagree, though not so obvious.

> whereas the FPGAs are expensive, 

The prices are coming down pretty fast for the smaller ones.

> hard to program (compared to a processor) 

Well, you can design a simple (or not so simple) processor
into the FPGA and then program that.  There are a few specifically
designed for efficient FPGA implementation.

> but offer much faster performance, support for any kind
> of I/O interface, re-programmable, etc. ? and if you really have a
> high-quantity product, then you would surely want to prototype on a
> DSP + FPGA, before moving to a custom ASIC device.

FPGAs offer much higher performance if you can do many operations
in parallel.  That is often easier in an FPGA, but it is not
automatic.  The actual clock rate is likely slower in the FPGA,
but if you can do many more operations per clock cycle the
throughput can be very high.
 
> A number of suppliers can be found on the Web that offers development
> boards with DSP + FPGAs and a growing number of software package can
> be bought that makes it possible to use Model-based design for BOTH
> the FPGA and the DSP.

-- glen

Article: 140319
Subject: Re: FPGAs and Cryptography
From: glen herrmannsfeldt <gah@ugcs.caltech.edu>
Date: Fri, 8 May 2009 16:06:37 +0000 (UTC)
Links: << >>  << T >>  << A >>
Allan Herriman <allanherriman@hotmail.com> wrote:
> On Thu, 07 May 2009 22:13:52 +0200, Xin Xiao wrote:
 
>> I've heard that FPGAs are good for doing some cryptography... Do any of
>> you have experience in cryptography projects and FPGAs?
 
> "Cryptography projects and FPGAs" describes aspects of my job.
 
> FPGAs aren't necessarily the best approach though.  If you are using (or 
> breaking!) standard algorithms, then there are off-the-shelf ASICs that 
> compete favourably with FPGAs for development time, speed and power.  
> E.g. google for Cavium Nitrox.

For normal use, I might agree.  It is not so obvious that it is
the best way to build code breakers.  Also, it seems that encryption
is a small part of the Cavium Nitrox.  A dedicated ASIC (if one
exists) might be faster/cheaper.
 
> As an example of performance, some years ago I was able to generate 20Gb/
> s of AES256 CTR mode keystream in a Xilinx Virtex2Pro part.  That design 
> was I/O limited.  *Much* higher performance would be achievable in a more 
> modern part or with a design that wasn't I/O limited.

The other advantage to FPGA is that you can reprogram it when new
algorithms appear.  
 
> (BTW, I've done more interesting things since that design, but I can't 
> talk about them in a public forum.)
 
> I imagine that past about 100Gb/s, the performance would be limited by 
> thermal constraints rather than chip area or clock speed.  Newer FPGA 
> families are better in this regard.
> 
>> What tools are available? (better if they are free). I suppose I have to
>> code in C/C++ and then synthesize that code to the hardware using a
>> tool?
 
> I recommend HDLs (e.g. VHDL or Verilog) over languages that weren't 
> designed for hardware description (e.g. C/C++).
> The tools that come from the FPGA vendors are cheap or free and effective.

(snip)

The implementation is likely very different in serial C than in
an appropriately parallel HDL implementation.  (I believe systolic
arrays would work in this case.)

-- glen

Article: 140320
Subject: Re: Question on using ODDR
From: "Symon" <symon_brewer@hotmail.com>
Date: Fri, 8 May 2009 17:46:08 +0100
Links: << >>  << T >>  << A >>

"lioncat" <dominghang@hotmail.com> wrote in message 
news:gI2dnYJKlIWPz5nXnZ2dnUVZ_oGdnZ2d@giganews.com...

>>
> Thank you for reply. But I have already read the library and the template
> of the ODDR. However my problem is in the map. The unit cannot be mapped. 
> I
> want to figure out how to solve this situation.

So, even if you use the exact template they suggest, maybe with a really 
simple experimental design, the map doesn't work? Did you also try looking 
at XAPPs with DDR? What about the FDDRRSE part, did you try that?
HTH., Syms. 



Article: 140321
Subject: Re: Dual Port RAM Inference
From: peter@xilinx.com
Date: Fri, 8 May 2009 10:01:30 -0700 (PDT)
Links: << >>  << T >>  << A >>
On May 8, 8:20=A0am, Mike Treseler <mtrese...@gmail.com> wrote:
> Jonathan Bromley wrote:
> > I haven't
> > yet seen a concise description of the right way to
> > cope with this across a range of synth tools, so
> > it's something I'd like to do anyway.
>
> I long ago gave up trying, so I will be interested
> to see your results.
>
> > If all else fails, I can grudgingly go back to
> > instantiation of wizard-generated components.
> > But I don't like doing that; it screws up my
> > design's portability, makes simulation tiresome
> > and slow, and makes the HDL code butt-ugly.
>
> I agree with the downside,
> but I stick with Peter's "simple dual-port" block rams
> which are portable across vendors and tools,
> with the same template.
>
> This is all I need for fifos.
> I use multiple instances and my own synchronous
> controllers to make more complicated memories.
>
> =A0 -- Mike Treseler

This is not a Xilinx or Altera circuit design problem, nor is it a
VHDL problem.
It is a systems design issue.
My favorite solution is to clock the two ports on opposite clock
polarity edges.
"If you cannot solve it, avoid it."
Peter Alfke

Article: 140322
Subject: Re: FPGAs and Cryptography
From: Allan Herriman <allanherriman@hotmail.com>
Date: 08 May 2009 17:33:16 GMT
Links: << >>  << T >>  << A >>
On Fri, 08 May 2009 16:06:37 +0000, glen herrmannsfeldt wrote:
> 
> The other advantage to FPGA is that you can reprogram it when new
> algorithms appear.

Good point.  The field is quite dynamic as old algorithms are discovered 
to be broken and new algorithms are developed to replace them.  The 
timescale is such that there's a good chance that an algorithm might need 
to be replaced during the lifetime of a product.

Some examples:
MD5 is now known to be not collision resistant, and it probably shouldn't 
be used.  Sha-1 is broken in the cryptographic sense (but still has 
useful security).  IIRC it is no longer allowed for certain secrecy 
levels.
Sha-3, a new US hash standard is due in 2012, however (as with AES) we 
will know the details well in advance.
One of the DRNGs (the one developed by NSA that uses EC) in the NIST spec 
SP800-90 has a suspected backdoor that was discovered in 2007.

Regards,
Allan

Article: 140323
Subject: Re: Dual Port RAM Inference
From: Mike Treseler <mtreseler@gmail.com>
Date: Fri, 08 May 2009 11:07:58 -0700
Links: << >>  << T >>  << A >>
peter@xilinx.com wrote:

> This is not a Xilinx or Altera circuit design problem, nor is it a
> VHDL problem.
> It is a systems design issue.

Some existing FPGA structures like PLLs and dpram arbiters
can't be described by a synchronous process.
My synthesis code can either instance the netlist or use something else.

> My favorite solution is to clock the two ports on opposite clock
> polarity edges.

If I couldn't avoid two write ports, I would time slice one clock.

> "If you cannot solve it, avoid it."

True. But I can't avoid talking about it ;)

... and what is the time limit on 'cannot solve it'?


        -- Mike Treseler

Article: 140324
Subject: Re: Dual Port RAM Inference
From: rickman <gnuarm@gmail.com>
Date: Fri, 8 May 2009 11:54:21 -0700 (PDT)
Links: << >>  << T >>  << A >>
On May 8, 8:42 am, Jonathan Bromley <jonathan.brom...@MYCOMPANY.com>
wrote:
> On Thu, 7 May 2009 20:52:47 -0700 (PDT), rickman wrote:
> >On May 7, 9:06 pm, Peter Alfke wrote:
> >> Here is a short Xilinx tutorial:
> >I don't think anyone has a problem understanding how the parts work.
> >The issue we are having is how to specify the hardware we want in the
> >HDL.
>
> Rick is right, but Peter's explanation is perhaps
> timely.  It's comparatively easy to describe both
> write-first and read-first behavior in HDL if there
> is only one write port, but it's a lot more tricky
> (especially in VHDL) if there are two write ports
> with independent clocks.  But the main problem is
> as Peter pointed out: there are some corner cases
> (for example, simultaneous write to the same location
> from both write ports) whose behavior is ill-defined.
> Of course, sensible designers will ensure that
> their memory controller never allows those ill-defined
> cases to occur.  But you can't easily explain to
> your synthesis tool that you have done so.  As a result,
> the synth tool must take your HDL description at face
> value, and do whatever it takes to ensure that the
> finished hardware obeys that description *in all
> cases* - hence the nonsense with logic (which
> wouldn't work right anyway) to avoid conflicting
> writes to the same address, because your HDL says
> that the behavior is well-defined but the memory
> doesn't work quite like that.
>
> As promised, I'll take this offline, do some work
> on it over the weekend, and report back.  I haven't
> yet seen a concise description of the right way to
> cope with this across a range of synth tools, so
> it's something I'd like to do anyway.
>
> If all else fails, I can grudgingly go back to
> instantiation of wizard-generated components.
> But I don't like doing that; it screws up my
> design's portability, makes simulation tiresome
> and slow, and makes the HDL code butt-ugly.

I don't like the instantiation approach either.  But I don't pretend
to expect that inference is just a matter of finding the right way to
describe a dual port memory.  If the chip makers and the tool vendors
can't tell us how to do it, then I can only assume that is because it
doesn't work.  I think the way to solve this problem is not by trying
to be creative in writing code, rather a direct conversation with the
vendors is needed.

I tried looking in all sorts of Xilinx docs and the only one that has
example code for write first dual port block ram is the XST manual,
equivalent to the Verilog code shown earlier in the thread.  I tried
that and got the following error...

ERROR:Xst - You are apparently trying to describe a RAM with several
write ports for signal <Mram_RAM>. This RAM cannot be implemented
using distributed resources.
-->

So it clearly understands what I am trying to do, but still wants to
use distributed ram!

One thing I find odd is that they are using a variable for the RAM and
perform the write before the read.  But then they give two examples
for write first memory, one that has direct logic to mux the input
data and the memory data.  The other uses the sequence of the
assignments to infer the write first behavior.  Either way gives the
same error.

I guess I would have expected the vendors to know how to use their own
software.

Rick



Site Home   Archive Home   FAQ Home   How to search the Archive   How to Navigate the Archive   
Compare FPGA features and resources   

Threads starting:
1994JulAugSepOctNovDec1994
1995JanFebMarAprMayJunJulAugSepOctNovDec1995
1996JanFebMarAprMayJunJulAugSepOctNovDec1996
1997JanFebMarAprMayJunJulAugSepOctNovDec1997
1998JanFebMarAprMayJunJulAugSepOctNovDec1998
1999JanFebMarAprMayJunJulAugSepOctNovDec1999
2000JanFebMarAprMayJunJulAugSepOctNovDec2000
2001JanFebMarAprMayJunJulAugSepOctNovDec2001
2002JanFebMarAprMayJunJulAugSepOctNovDec2002
2003JanFebMarAprMayJunJulAugSepOctNovDec2003
2004JanFebMarAprMayJunJulAugSepOctNovDec2004
2005JanFebMarAprMayJunJulAugSepOctNovDec2005
2006JanFebMarAprMayJunJulAugSepOctNovDec2006
2007JanFebMarAprMayJunJulAugSepOctNovDec2007
2008JanFebMarAprMayJunJulAugSepOctNovDec2008
2009JanFebMarAprMayJunJulAugSepOctNovDec2009
2010JanFebMarAprMayJunJulAugSepOctNovDec2010
2011JanFebMarAprMayJunJulAugSepOctNovDec2011
2012JanFebMarAprMayJunJulAugSepOctNovDec2012
2013JanFebMarAprMayJunJulAugSepOctNovDec2013
2014JanFebMarAprMayJunJulAugSepOctNovDec2014
2015JanFebMarAprMayJunJulAugSepOctNovDec2015
2016JanFebMarAprMayJunJulAugSepOctNovDec2016
2017JanFebMarAprMayJunJulAugSepOctNovDec2017
2018JanFebMarAprMayJunJulAugSepOctNovDec2018
2019JanFebMarAprMayJunJulAugSepOctNovDec2019
2020JanFebMarAprMay2020

Authors:A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

Custom Search