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
2017JanFebMarApr2017

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 42025

Article: 42025
Subject: Re: Laying out the design
From: Ray Andraka <ray@andraka.com>
Date: Sat, 13 Apr 2002 03:45:20 GMT
Links: << >>  << T >>  << A >>
This is changing.  Altera has apparently recognized the value of allowing an
expert user to direct placement to improve performance.  This is especially
necessary in the 20K and new Stratix families because the local connections
between LABs make it possible to significantly alter the delay times.  In 10K, it
was not nearly as helpful since anything arithmetic had to go on the row routes
anyway.  Altera is listening very carefully to expert users, and I think it is
beginning to show in their offerings both in terms of the silicon and the
software.



Kevin Brace wrote:

>
> Altera attitude seems like users don't have to get too much in detail,
> and let the tool handle it, but shouldn't Altera listen to those power
> users?
>

--
--Ray Andraka, P.E.
President, the Andraka Consulting Group, Inc.
401/884-7930     Fax 401/884-7950
email ray@andraka.com
http://www.andraka.com

 "They that give up essential liberty to obtain a little
  temporary safety deserve neither liberty nor safety."
                                          -Benjamin Franklin, 1759



Article: 42026
Subject: Re: Marquis of Queensbury Rules
From: Ray Andraka <ray@andraka.com>
Date: Sat, 13 Apr 2002 03:48:27 GMT
Links: << >>  << T >>  << A >>
Competition is good.  It keeps the vendors honest, keeps the parts affordable
and gets us a better selection of parts to use and better features to play
with.  I'm all for it.

Peter Ormsby wrote:

> Both Altera and Xilinx have taken ideas from each other and used them to
> make their own products better.  This is good for all the engineers out
> there as they will have better devices today than they had five years ago -
> no matter which vendor's devices they decide to use.
>
> -Pete-

--
--Ray Andraka, P.E.
President, the Andraka Consulting Group, Inc.
401/884-7930     Fax 401/884-7950
email ray@andraka.com
http://www.andraka.com

 "They that give up essential liberty to obtain a little
  temporary safety deserve neither liberty nor safety."
                                          -Benjamin Franklin, 1759



Article: 42027
Subject: Re: PCI Bridge Question
From: Ray Andraka <ray@andraka.com>
Date: Sat, 13 Apr 2002 03:59:06 GMT
Links: << >>  << T >>  << A >>
You get what you pay for.  If you want a fully documented guaranteed bug
free core, I don't think a free core is likely to meet your needs.
  If you feel strongly about this, you could always pick up a few of the
cores, test them, post the testbenches and test results, and then document
them, I am quite sure no one would object.  Besides, I am sure it would give
you a wonderful learning experience.  One learns very quickly when debugging
another persons work, even if that other person is a very poor designer.
Please remember that no one got paid to do the thousands of hours of work
represented by those free cores.

Kevin Brace wrote:

>         No, I won't say that the ideal of the Opencores.org is bad, but
> the problem is, the people who work on projects seem to post their work
> there without adequately testing it.
> Perhaps for a small project (I will consider a PCI IP core a small
> project.) they might want to have a policy not to allow code to be
> available until all known bugs are fixed.
> Also, since the design will be open source, the authors will have to
> keep the design easy to understand if someone wants to modify or fix
> bugs themselves, but the authors of Opencores.org PCI IP core didn't
> seem to care about it.
>

--
--Ray Andraka, P.E.
President, the Andraka Consulting Group, Inc.
401/884-7930     Fax 401/884-7950
email ray@andraka.com
http://www.andraka.com

 "They that give up essential liberty to obtain a little
  temporary safety deserve neither liberty nor safety."
                                          -Benjamin Franklin, 1759



Article: 42028
Subject: Re: FPGA eval/dev boards with *serial* interface?
From: Ray Andraka <ray@andraka.com>
Date: Sat, 13 Apr 2002 04:03:34 GMT
Links: << >>  << T >>  << A >>
If you can find one, the old xchecker cable is serial.  I don't
think it is supported by the 4.x software though.  Probably the
best solution is to run out to your local computer shop and buy
one of those $15 parallel port boards and plug it into your
system.

Tom Loredo wrote:

> Hi-
>
> I'm experienced with 8-bit microcontrollers, but a complete
> newbie to FPGAs.  I'm considering them for an upcoming
> project, and a factor in my choice is availability of
> an affordable evaluation/development board with a *serial*
> interface.  So far I've been looking at parts from Xilinx
> and Atmel, but the eval boards I've come across all use
> the parallel port on the PC.  I only have a serial or USB
> port, so these are not suitable.  Any leads/suggesions
> are appreciated.
>
> Thanks!
> Tom Loredo

--
--Ray Andraka, P.E.
President, the Andraka Consulting Group, Inc.
401/884-7930     Fax 401/884-7950
email ray@andraka.com
http://www.andraka.com

 "They that give up essential liberty to obtain a little
  temporary safety deserve neither liberty nor safety."
                                          -Benjamin Franklin,
1759



Article: 42029
Subject: Re: Stupid .ngd file questions....
From: Philip Freidin <philip@fliptronics.com>
Date: Sat, 13 Apr 2002 06:28:00 GMT
Links: << >>  << T >>  << A >>
On Fri, 12 Apr 2002 22:27:15 +0000 (UTC), nweaver@CSUA.Berkeley.EDU (Nicholas
Weaver) wrote:
>OK, call me brain damaged, but searching the Xilinx web site isn't
>helping me yet.

OK, "you are brain damaged" .... Nope, I checked your other postings,
and you are not.

>I want to do some post placement, pre-routing manipulation of the
>design (well, actually, build TOOLS to do...).  Is there a way to
>extract/manipulate the .ngd file, including placement information?  Or
>is the format published somewhere?

So I did a litle test design, and I was able to change the location of a
CLB post placement, pre routing.

(all from a w2000 batch file)

First:

set des=test
set maptype=XC4013E-4-PQ208
ngdbuild -p %maptype% -u %des%.edn %des%.ngd > %des%.log
map -pr b -p %maptype% -o map.ncd -detail %des%.ngd %des%.pcf >> %des%.log
par -w -pl 5 -r -detail map.ncd %des%.ncd %des%.pcf >> %des%.log

The "-r" disables the routing phase, and you get an .NCD that is placed
but not routed. You can look at it in the FPGA editor.

Then:

xdl -ncd2xdl -nopips test.ncd

creates test.XDL

which I then edited, by changing the location of one of the CLBs.

Then:

xdl -xdl2ncd test.xdl test2.ncd

renamed test.pcf to test2.pcf

par -k -rl 2 test2.ncd test2pr.ncd

The "-k" is for re-entrant routing.


Then checke the resultant .NCD, to see if the moved CLB had survived
the surgery. It had.

I am looking forward to your placement repairer program that will
fixup the stupid data-path placement that the xilinx tools do.


All the best,

Philip Freidin


>ngd2{edif,vhdl,vlg} seems to just produce a simulation netlist,
>annotated with delays.  I want something I can go back & forth with.
>
>Thanks.

Philip Freidin
Fliptronics

Article: 42030
Subject: Slave serial loading of spartan II bitstream
From: George Hodges <nospam@here.please>
Date: Sat, 13 Apr 2002 07:40:08 +0100
Links: << >>  << T >>  << A >>
I've got my bitstream files and wish to load them in slave serial mode.

There are application notes but I think they might be a bit out of date.

Is this the right recipe ?

Take one .bit file
Drop the first 68 bytes so it starts with the 32 ones like the .mcs file
does.
Take each byte, and starting from the MSB clock it in through CCLK/DIN
being careful to observe timing constraints.

I find that the only way to make /INIT go low (DONE never goes high
unfortunately) is by clocking the stuff in for a second time then /INIT
drops after a couple hundred (repeatable number) bits. The clocks til
/INIT dropping is independant of how many zeros or ones I pad with after
the first bitstream. Almost as if the preamble is actually a postamble
and the entire stream needs turning back to front.

George

Article: 42031
Subject: webpack ISE
From: "crackeur" <crackeur@attbi.com>
Date: Sat, 13 Apr 2002 07:40:22 GMT
Links: << >>  << T >>  << A >>
This is what I want to do. I have a spec and I have some HDL written,
now I am going to try to target a spartII device, could some one please
point out a set of common tools needed? I am looking for HDL editor,
behavior and netlist level simulation tools, timing tools and ways to
download everything to my fpga, what tools in webpack ISE and other
components do I need to accomplish this goal?





Article: 42032
Subject: DLL property control in UCF
From: "Norman Yang" <norman@zh.t2-design.com>
Date: Sat, 13 Apr 2002 16:22:25 +0800
Links: << >>  << T >>  << A >>
Hello everyone:
I meet a problem in controlling VertexE DLL property in UCF file. I don't
know how to control the  Clock Divide Property of DLL in UCF. XILINX
datasheet mentioned about how to control it by symbol. And I also want to
know how to control it in HDL code and affect simulation. The default one is
always divide by 2. Thanks!

--
Best regards.
Norman
norman@zh.t2-design.com




Article: 42033
Subject: Re: bad experience with Xilinx ISE 4.1i and Xilinx hotline suppot
From: Eric Smith <eric-no-spam-for-me@brouhaha.com>
Date: 13 Apr 2002 01:47:18 -0700
Links: << >>  << T >>  << A >>
Theron Hicks <hicksthe@egr.msu.edu> writes:
>     Please enlighten me.  In 4.1 the simulation options are
> 1.    Simulate Behavioral VHDL Model
> 2.    Simulate Post-Translate VHDL Model
> 3.    Simulate Post-Map VHDL Model
> 4.    Simulate Post-Place & Route VHDL Model
> Can you clarify what each of these really accomplishes?
> 
> #1,  I assume is simulation of the code as written and assumes delta delays

I suppose it might also simulate any explicitly specified delays, though
that's obviously a big no-no for synthesis. 

> I am uncertain as to precisely what #2 and #3 do.

My guess is that at post-translate you have a pure RTL model, but that it
has not been partitioned to the specific primitives of the target FPGA,
since that's the function of mapping.

After mapping, the logic matches what's available in the FPGAs, but the
routing has not been done, so there's still not a complete delay model.

> #4 I assume is the actually delays including routing, etc.

Sounds right.

Perhaps someone at Xilinx can confirm?

Article: 42034
Subject: Re: Slave serial loading of spartan II bitstream
From: "Tim" <tim@rockylogic.com.nooospam.com>
Date: Sat, 13 Apr 2002 12:19:45 +0100
Links: << >>  << T >>  << A >>

George Hodges wrote

> I've got my bitstream files and wish to load them in slave serial mode.
>
> There are application notes but I think they might be a bit out of date.
>
> Is this the right recipe ?
>
> Take one .bit file
> Drop the first 68 bytes so it starts with the 32 ones like the .mcs file
> does.

Nope.  Check the FPGA FAQ site for the bit file format.  The header length
depends on the date, part type, and so on.

The rest of your recipe looks OK.




Article: 42035
Subject: Re: new to fpga's need insight
From: "Tim" <tim@rockylogic.com.nooospam.com>
Date: Sat, 13 Apr 2002 12:21:21 +0100
Links: << >>  << T >>  << A >>
Start at optimagic.com and fpga-faq.com

akhar wrote

> Hello, I have been looking around on the net for more information but it
> seems quite obscure (not totally but a bit) I am interested in learning more
> about FPGA's and how to program them. I have found www.fpgacpu.org to be
> quite interesting but no quite clear at least with what I know from the
> FPGA's. I wanted to know :
> - what is the native language to program an FPGA ( I saw that I can use C
> but I have not found the compiler , Can I use another language?(lisp,scheme,
> java,Smalltalk))
> - Are there limits to what I can program? (I would like to use them to
> program neural networks)
> - Is it possible to use OO paradigms?
> - what is the best recommended starter kit (altera's or xilinx's)
> - how do I know the number of gates I'll need for a project/application




Article: 42036
Subject: Re: new to fpga's need insight
From: Neil Franklin <neil@franklin.ch.remove>
Date: 13 Apr 2002 14:54:41 +0200
Links: << >>  << T >>  << A >>
"akhar" <akhar@videotron.ca> writes:

> - what is the native language to program an FPGA ( I saw that I can use C
> but I have not found the compiler , Can I use another language?(lisp,scheme,
> java,Smalltalk))

Well there exist no such thing as an "native" language. What there
exist are 4 entirely different approaches to programming and a few
languages or tools for each:

- high level code: Verilog, VHDL, AHDL, CUPL, Handel-C, ...
- high level schematic (drawing): whatever format the tool wants
- low level schematic: mainly FPGA Editor (only for Xilinx chips)
- low level code: JBits, cnets, PamDC, ...

Generally:

- Low level means that you select individual FPGA features, comparable
  with assembly language programming.
- High level means you write logic formulas and they are compiled,
  somewhat like high level programming.
- Code means an ASCII source files that are assembled/compiled.
- Schematic means some an drawing/CAD style program.


As for C: there exists high level (Handel-C) and low level (cnets and
PamDC) tools that use C as language. Handel-C tries to actually compile
C expressions to FPGAs. cnets and PamDC use C as "driver" language to
drive an library of "place this type of function here" calls.

As for other languages like Lisp, Scheme, etc: forget them. Even
Handel-C is at the limit of what is possible presently. And gets lots
of flak for not being up to it yet.


> - Are there limits to what I can program? (I would like to use them to
> program neural networks)

For one: Chip size. Remember the days when computers had 100k to 1M
of RAM, and no virtual memory? All them "out of memory" errors. FPGA
programming will remind you of that. Get a bigger chip...

More importantly though: FPGAs are not sequencial "do this, do that"
programming like processors. FPGAs are more "place this function here,
place that there" with data traveling from one unit to the next and
all units working all the time. So programming is splitting your
problem into units and placing them so that data travels fast.

So traditional languages are not what you want. You want an "layout
descriptor" language.


> - Is it possible to use OO paradigms?

No. Totally useless.

OO is all about managing sequential access to data. An FPGA is about
as OO as an layout of machines on an factory floor! Each Unit is one
instance of one class of processing (and not just a class for data
objects to be instantiated), and data is not in instanciated objects
that direct computation, but rather as stream of packets that travel
(and need to be directed) to the proper units.

So this is nearer to traditional "we know what data to expect"
programming.


> - what is the best recommended starter kit (altera's or xilinx's)

Roughly equal. That is why both firms are roughly same size. Altera
seems to win on complexer logic, Xilinx on faster Arithmetic. And whose
tools are better, that is an ongoing but low-intensity holy war :-).

For your neural networks (many small units with arithmetic and
connecting to near neighbors) I would recommend Xilinx.


> - how do I know the number of gates I'll need for a project/application

Forget gates. They are an near-useless marketing number.

Count Logic units (so called LUTs). Basically one LUT can compute any
1-bit function of 4 bits of input. And its associated FF can facultatively
store one bit of data (the output bit) before sending it further.

See the manufacturesrs data sheet for the range of LUT sizes they
make. Usually in the few-100 to few-10000 range.

To give an example: Xilinx XC2S200, for $50, is 4700 LUT/FF units.


--
Neil Franklin, neil@franklin.ch.remove http://neil.franklin.ch/
Hacker, Unix Guru, El Eng HTL/BSc, Sysadmin, Archer, Roleplayer
- Make your code truely free: put it into the public domain

Article: 42037
Subject: Re: synplify, quartus II 2.0
From: "Johnsonw10" <johnsonw10_NOSPAM@hotmail.com>
Date: Sat, 13 Apr 2002 13:06:19 GMT
Links: << >>  << T >>  << A >>

> BTW You have plenty of RAM I take it? You'll need it with both running.
>

Does anybody know why Quartus II 2.0 needs huge amount of memory? Just
opening up the GUI without any project opened takes 80M bytes of RAM on a
WinNT PC!!! Unbelievable!



Article: 42038
Subject: Re: new to fpga's need insight
From: Neil Franklin <neil@franklin.ch.remove>
Date: 13 Apr 2002 16:50:37 +0200
Links: << >>  << T >>  << A >>
"akhar" <akhar@videotron.ca> writes:

> I looked up Handel C and found  a lot of article talking about it but no
> official web site? is there one or is hcc the only compiler ?

Sorry, I never looked into them. Only know their name from what has been
discussed here on the group.

They are at the top of the "high level code" approach, I am "low level
code" style thinker and therefore programmer. I come from an El Eng
background , using 74(LS)xx(x) chips. While you seem to be coming from
an Comp Sci background used to HLLs.


> What sites do
> you recommend for learning how to program the FPGA's ?

Caveat: sites selected for my own use, so perhaps not good your style.

You could visit my computer and electronics links page:
http://neil.franklin.ch/Links/comp_electro.html

There scroll down to "FPGA CPU", as that was also for me the first
FPGA oriented site I found. From there on there are quite a few links
to (and into) various FPGA sites. The file is find-time ordered, so
don't imply anything else from ordering.

Most important (for me, low level code style) were the vendors data
sheets. All the important ones have direct links in each vendors
section, all fairly short after FPGA CPU.


--
Neil Franklin, neil@franklin.ch.remove http://neil.franklin.ch/
Hacker, Unix Guru, El Eng HTL/BSc, Sysadmin, Archer, Roleplayer
- Make your code truely free: put it into the public domain

Article: 42039
Subject: Re: new to fpga's need insight
From: "akhar" <akhar@videotron.ca>
Date: Sat, 13 Apr 2002 10:05:20 -0700
Links: << >>  << T >>  << A >>
Thanks for the pointers,
I looked up Handel C and found  a lot of article talking about it but no
official web site? is there one or is hcc the only compiler ? What sites do
you recommend for learning how to program the FPGA's ?

best regards

Stephane
"Neil Franklin" <neil@franklin.ch.remove> wrote in message
news:6ud6x3zrla.fsf@chonsp.franklin.ch...
> "akhar" <akhar@videotron.ca> writes:
>
> > - what is the native language to program an FPGA ( I saw that I can use
C
> > but I have not found the compiler , Can I use another
language?(lisp,scheme,
> > java,Smalltalk))
>
> Well there exist no such thing as an "native" language. What there
> exist are 4 entirely different approaches to programming and a few
> languages or tools for each:
>
> - high level code: Verilog, VHDL, AHDL, CUPL, Handel-C, ...
> - high level schematic (drawing): whatever format the tool wants
> - low level schematic: mainly FPGA Editor (only for Xilinx chips)
> - low level code: JBits, cnets, PamDC, ...
>
> Generally:
>
> - Low level means that you select individual FPGA features, comparable
>   with assembly language programming.
> - High level means you write logic formulas and they are compiled,
>   somewhat like high level programming.
> - Code means an ASCII source files that are assembled/compiled.
> - Schematic means some an drawing/CAD style program.
>
>
> As for C: there exists high level (Handel-C) and low level (cnets and
> PamDC) tools that use C as language. Handel-C tries to actually compile
> C expressions to FPGAs. cnets and PamDC use C as "driver" language to
> drive an library of "place this type of function here" calls.
>
> As for other languages like Lisp, Scheme, etc: forget them. Even
> Handel-C is at the limit of what is possible presently. And gets lots
> of flak for not being up to it yet.
>
>
> > - Are there limits to what I can program? (I would like to use them to
> > program neural networks)
>
> For one: Chip size. Remember the days when computers had 100k to 1M
> of RAM, and no virtual memory? All them "out of memory" errors. FPGA
> programming will remind you of that. Get a bigger chip...
>
> More importantly though: FPGAs are not sequencial "do this, do that"
> programming like processors. FPGAs are more "place this function here,
> place that there" with data traveling from one unit to the next and
> all units working all the time. So programming is splitting your
> problem into units and placing them so that data travels fast.
>
> So traditional languages are not what you want. You want an "layout
> descriptor" language.
>
>
> > - Is it possible to use OO paradigms?
>
> No. Totally useless.
>
> OO is all about managing sequential access to data. An FPGA is about
> as OO as an layout of machines on an factory floor! Each Unit is one
> instance of one class of processing (and not just a class for data
> objects to be instantiated), and data is not in instanciated objects
> that direct computation, but rather as stream of packets that travel
> (and need to be directed) to the proper units.
>
> So this is nearer to traditional "we know what data to expect"
> programming.
>
>
> > - what is the best recommended starter kit (altera's or xilinx's)
>
> Roughly equal. That is why both firms are roughly same size. Altera
> seems to win on complexer logic, Xilinx on faster Arithmetic. And whose
> tools are better, that is an ongoing but low-intensity holy war :-).
>
> For your neural networks (many small units with arithmetic and
> connecting to near neighbors) I would recommend Xilinx.
>
>
> > - how do I know the number of gates I'll need for a project/application
>
> Forget gates. They are an near-useless marketing number.
>
> Count Logic units (so called LUTs). Basically one LUT can compute any
> 1-bit function of 4 bits of input. And its associated FF can facultatively
> store one bit of data (the output bit) before sending it further.
>
> See the manufacturesrs data sheet for the range of LUT sizes they
> make. Usually in the few-100 to few-10000 range.
>
> To give an example: Xilinx XC2S200, for $50, is 4700 LUT/FF units.
>
>
> --
> Neil Franklin, neil@franklin.ch.remove http://neil.franklin.ch/
> Hacker, Unix Guru, El Eng HTL/BSc, Sysadmin, Archer, Roleplayer
> - Make your code truely free: put it into the public domain



Article: 42040
Subject: Re: new to fpga's need insight
From: Phil Hays <SpamPostmaster@attbi.com>
Date: Sat, 13 Apr 2002 17:47:16 GMT
Links: << >>  << T >>  << A >>
akhar wrote:
 
> Thanks for the pointers,
> I looked up Handel C and found a lot of article talking about it but no
> official web site?

http://www.celoxica.com/home.htm


> is there one or is hcc the only compiler ?

Synopsys and Forte have SystemC to Verilog or VHDL compilers.  SystemC
is a "standard" library of C++ classes.  Xilinx has a Java to Verilog
compiler that is "freeware" until this summer:

http://www.xilinx.com/ise/advanced/forge.htm

http://www.synopsys.com/cgi-bin/sld/ltl1.cgi#f2

http://www.forteds.com/products/cynthesizer.html


> What sites do
> you recommend for learning how to program the FPGA's ?

To do what?  If you want high speed digital signal processing in the
lowest cost possible FPGA, you may need different tools, attitudes and
skills than if you want to emulate a ASIC, or emulate obsolete hardware,
or communications equipment, or any of the various and sundry uses that
FPGAs are put to.  Some of the tools are useful for a narrow range of
types of designs.  For freeware tools, both Xilinx and Altera have
fairly nice packages for starting with VHDL or Verilog:

http://www.altera.com/products/software/sfw-quarwebmain.html

http://www.xilinx.com/webpack/index.html


> - Are there limits to what I can program? 

Of course.  Large FPGAs have ten's of thousands of slices and hundreds
of pins, and these limit the computation than can be done in a clock
cycle and the amount of data than be input or output in a clock cycle.  


>(I would like to use them to program neural networks)

Are you more interesting in learning about FPGAs, in learning about
neural networks or is there an application of neural networks you are
interested in?  Also, is this learning or for a real product?  If you
are mostly interested in an application of neural nets, you have a need
to speed it up relative to a software implementation, but don't need a
fully optimized design, you might want to look at one of the
HLL(HandelC, SystemC, Java).  If you want to learn lots about FPGAs, you
might want to start with a much lower level of abstraction (schematic or
VHDL physical netlist) and learn about primitives, placement and other
basics of FPGA design.


> - what is the best recommended starter kit (altera's or xilinx's)

I'd say Xilinx is somewhat ahead in general, this week.  Altera is
competitive, and has some advantages.


> - how do I know the number of gates I'll need for a project/application

Don't look at gate counts.  The devices don't have "gates".  The devices
have small 4 input "Look Up Tables" LUTS, carry chains, multipliers and
other special purpose logic, and larger memories, and some of these can
be used for other purposes.  The published gate counts make the
assumption that you can use some large fraction of all of these: and
real designs don't.  I do estimation by trying to identify the resource
that will be most used, and plan for the part that has enough of that
resource.  It's not easy.  Usually, however, the design will be limited
by internal memory or by LUTs.  In Xilinx speak, one slice = 1 LUT + 1
flipflop (single bit storage).


-- 
Phil Hays

Article: 42041
Subject: Re: prototyping an ASIC
From: Ken McElvain <ken@synplicity.com>
Date: Sat, 13 Apr 2002 17:50:46 GMT
Links: << >>  << T >>  << A >>

Take a look at

http://www.synplicity.com/literature/index.html#certify

You can now easily map a large ASIC design onto an off
the shelf board (fixed routing) with multiple FPGAs.  We
have a web page with links to suppliers (and would like more).

You may well need more capacity than you think.

1) The effective capacity of an FPGA depends on the type of
logic you are implementing.
2)  You will probably want to add a bunch of instrumentation
logic to your design and you need extra capacity for that.


Ken McElvain, CTO
Synplicity, Inc.


Gil Herbeck wrote:

> I need to prototype an ASIC design and am looking for
> advice on type of FPGA and on FPGA board as well.
> 
> The board needs to have an ARM9, external memory, an
> interface to a PC (serial is ok), the FPGA (or ASIC),
> and a header to plug in a daughter card with some pins
> routed to the FPGA.
> 
> The ASIC will have between 100K and 500K ASIC Logic
> Gates.  It will run at about 150 MHz.  It needs about
> 200 KB of internal RAM.  And it will have a lot of
> multipliers.  There will probably be some pipelining
> in the ASIC to meet speed - and probably deeper pipes
> in the FPGA.  We want to match the FPGA to the ASIC
> as closely as possible.
> 
> I think the key factors in FPGA selection are:
> - Capacity.  We want to fit in one FPGA.
> - Performance.  We want to run at speed.
> - "ASIC-like" synthesis library (see below)?
> - Availability of board described above.
> 
> "ASIC-like" synthesis library...  The datapath
> content on the ASIC may force us to use one of the
> datapath synthesis tools.  These tools don't support
> FPGA architectures directly.  I've heard that since
> the Actel architecture is "fine-grain" that it works
> best for these types of designs.
> 
> Any advice will be much appreciated.
> 
> Thanks,
> Gil
> 


Article: 42042
Subject: Re: Slave serial loading of spartan II bitstream
From: Philip Freidin <philip@fliptronics.com>
Date: Sat, 13 Apr 2002 18:29:00 GMT
Links: << >>  << T >>  << A >>
On Sat, 13 Apr 2002 07:40:08 +0100, George Hodges <nospam@here.please> wrote:
>I've got my bitstream files and wish to load them in slave serial mode.
>
>There are application notes but I think they might be a bit out of date.

App note 138 in particular should be read. It is up to date, as far as I
know.

     http://www.xilinx.com/xapp/xapp138.pdf

>Is this the right recipe ?
>
>Take one .bit file

Good start

>Drop the first 68 bytes so it starts with the 32 ones like the .mcs file
>does.

Various things affect the number of bytes to be skipped. See

   http://www.fpga-faq.com/FAQ_Pages/0026_Tell_me_about_bit_files.htm

for details.

>Take each byte, and starting from the MSB clock it in through CCLK/DIN
>being careful to observe timing constraints.

Right. MSB first.

>I find that the only way to make /INIT go low (DONE never goes high

This is NOT what you want! /INIT starts off low, indicating that the
FPGA is clearing all the config memory (called house-cleaning). It
then goes high for the duration of the configuration process. If it
goes low during configuration, this is an indication of an error.

>unfortunately) is by clocking the stuff in for a second time then /INIT
>drops after a couple hundred (repeatable number) bits.

Which is an indication of an error.

>The clocks till
>/INIT dropping is independant of how many zeros or ones I pad with after
>the first bitstream. Almost as if the preamble is actually a postamble
>and the entire stream needs turning back to front.

No. I can't tell what is going wrong with your config, but it isn't the
reversing the bitstream!  . Since DONE didn't go high, and /INIT didn't
go low, then I would suggest re-checking that you are shifting in the
correct bits. Another common problem is the quality of the CCLK that
you are supplying. People often don't give it enough attention because
it is not particularly fast. Even so, the signal integrity is very
important, and you should check that the rising and falling edges are
clean, fast, and monotonic. You need a fast scope (>300MHz) for this.

>George

Good luck,

Philip Freidin


Philip Freidin
Fliptronics

Article: 42043
Subject: Re: Attributes *and* generics!?
From: Mike Treseler <mike.treseler@flukenetworks.com>
Date: Sat, 13 Apr 2002 13:29:37 -0700
Links: << >>  << T >>  << A >>
Ray Andraka wrote:

> Even if you write your own you run into the attributes and generics
> issue. 
. . .

This presumes that making use of some 
device-specific nuance is required.
This do-it-yourself synthesis is certainly required
for many designs. For others, the lowest
common synthesizable subset for brand A and X
is good enough.


   -- Mike Treseler

Article: 42044
Subject: Re: webpack ISE
From: Kevin Brace <ihatespam99kevinbraceusenet@ihatespam99hotmail.com>
Date: Sat, 13 Apr 2002 15:40:31 -0500
Links: << >>  << T >>  << A >>
ISE WebPACK got all the tools you are asking for, so go ahead and
download the whole version from Xilinx. (About 160MB total download.
Takes about 14 hours if you are using a 56K modem to download it.)
Don't forget to download ModelSim XE-Starter at the same time.
Several Xilinx distributors like Insight Electronics or Avnet sell low
cost Spartan-II-based prototype boards, so you should be able to use
those to test your design.



Kevin Brace (In general, don't respond to me directly, and respond
within the newsgroup.)



crackeur wrote:
> 
> This is what I want to do. I have a spec and I have some HDL written,
> now I am going to try to target a spartII device, could some one please
> point out a set of common tools needed? I am looking for HDL editor,
> behavior and netlist level simulation tools, timing tools and ways to
> download everything to my fpga, what tools in webpack ISE and other
> components do I need to accomplish this goal?

Article: 42045
Subject: Re: PCI Bridge Question
From: Kevin Brace <ihatespam99kevinbraceusenet@ihatespam99hotmail.com>
Date: Sat, 13 Apr 2002 15:47:53 -0500
Links: << >>  << T >>  << A >>


Ray Andraka wrote:
> 
> You get what you pay for.  If you want a fully documented guaranteed bug
> free core, I don't think a free core is likely to meet your needs.
>   If you feel strongly about this, you could always pick up a few of the
> cores, test them, post the testbenches and test results, and then document
> them, I am quite sure no one would object.  Besides, I am sure it would give
> you a wonderful learning experience.  One learns very quickly when debugging
> another persons work, even if that other person is a very poor designer.
> Please remember that no one got paid to do the thousands of hours of work
> represented by those free cores.
> 


        Ray, is this reply intended for me or the original poster?
I agree that most of the time, free IP cores are pretty much "you get
what you pay for," but if Opencores.org is a little more strict about
the quality of the work getting uploaded, I think some of them will be
useful even in commercial designs.



Kevin Brace (In general, don't respond to me directly, and respond
within the newsgroup.)

Article: 42046
Subject: Re: Attributes *and* generics!?
From: Mike Treseler <mike.treseler@flukenetworks.com>
Date: Sat, 13 Apr 2002 13:48:48 -0700
Links: << >>  << T >>  << A >>
Martin Thompson wrote:

> Without wishing to ignite anything - the Altera LPMs use the generics
> for simulation, which Synplify passes through the EDIF netlist on the
> blackboxes and then the P&R tool deals with it from there.  Any chance
> of Xilinx imitating this?

The best thing about the Altera LPMs is that they are open source.
If you ignore the schematic string interface, you get a 
synthesizable template for useful hardware blocks that work
for both brands A and X through the EDIF netlist. 
Of course the coverage for brand A is better, since
they wrote it.

     -- Mike Treseler

Article: 42047
Subject: Re: new to fpga's need insight
From: Kevin Brace <ihatespam99kevinbraceusenet@ihatespam99hotmail.com>
Date: Sat, 13 Apr 2002 16:25:31 -0500
Links: << >>  << T >>  << A >>


akhar wrote:
> 
> Hello, I have been looking around on the net for more information but it
> seems quite obscure (not totally but a bit) I am interested in learning more
> about FPGA's and how to program them. I have found www.fpgacpu.org to be
> quite interesting but no quite clear at least with what I know from the
> FPGA's. I wanted to know :
> - what is the native language to program an FPGA ( I saw that I can use C
> but I have not found the compiler , Can I use another language?(lisp,scheme,
> java,Smalltalk))


        I personally won't like to use the phrase "programming an FPGA."
Although SRAM-based FPGAs have virtually unlimited programmability, I
will rather call it, "Developing a circuit for an FPGA."
        If you want to seriously learn how to design circuits in which
FPGAs can handle, learn Verilog or VHDL.
I personally will recommend learning Verilog, and play around with a
sample design that comes with free tools I will mention later.
        Besides http://www.fpgacpu.org, you can download free IP cores
from Opencores.org (http://www.opencores.org), but be aware that the
quality of the IP cores there isn't that great.



> - Are there limits to what I can program? (I would like to use them to
> program neural networks)


        Depends on the capacity of the FPGA.
For FPGAs that cost less than $30, the gate density of FPGAs are still
pretty small.
For only that much of money, all you will likely get will be roughly
about 50,000 ASIC (custom chip) gates.





> - Is it possible to use OO paradigms?


        Although there might be attempts to bring such concepts into the
FPGA world, I don't think it has worked too well, so you should forget
about it.



> - what is the best recommended starter kit (altera's or xilinx's)


       Well, I will become a partisan here.
If you a newbie of designing circuits for an FPGA, I recommend
downloading freely available design tools from various FPGA vendors.
However, the tools from Xilinx and Altera are the ones that are useful
in practice.
I will personally recommend using Xilinx's free tools because Altera
doesn't give you a free version of ModelSim (Yes, because it is free,
the version distributed by Xilinx is somewhat crippled, but it still
works at a reasonable speed as long as the design is not too big.), and
there aren't too many vendors selling low cost Altera FPGA-based
prototype boards.
From my own experience, Xilinx's free tools seem to run more stable,
faster, and requires less hardware resource than Altera's free tools.
For low cost Xilinx FPGA-based prototype boards, check out Insight
Electronics (http://www.insight-electronics.com) or Avnet
(http://www.avnet.com) website.
You can get one below $200.




> - how do I know the number of gates I'll need for a project/application
> 
> That's most of it I think
> 
> Best Regards
> 
> Stephane


        Design whatever you want to, and target the biggest chip that's
available.
Hopefully it will fit.



Kevin Brace (In general, don't respond to me directly, and respond
within the newsgroup.)

Article: 42048
Subject: Re: new to fpga's need insight
From: Peter Alfke <palfke@earthlink.net>
Date: Sat, 13 Apr 2002 22:34:40 GMT
Links: << >>  << T >>  << A >>


Phil Hays wrote:

> <snip, lots of good stuff>

> It's not easy.  Usually, however, the design will be limited
> by internal memory or by LUTs.  In Xilinx speak, one slice = 1 LUT + 1
> flipflop (single bit storage).

By definition, a slice is TWO LUTs and two flip-flops.
But let'snot argue about the arcane reasons for this definition...

Peter Alfke, Xilinx Applications



Article: 42049
Subject: virtex2 bufgce or not bufgce
From: Lasse Langwadt Christensen <langwadt@ieee.org>
Date: Sat, 13 Apr 2002 16:38:45 -0700
Links: << >>  << T >>  << A >>
Hi, 

I'm working on a design in a Virtex2 that uses both gated and 
ungated clk's i.e. a clk is distributed to modules in the 
design and some of them internally gate the clk to some of the logic. 

Can I just do the gating with a bufgce and run logic 
of both system clk and some of a gated version of that clk, 
or will there be so much skew between the two clks it's unsafe?

Or, will I have to change the code to run all clk's through 
a bufgce to have them aligned? 

thanks,

-Lasse
-- Lasse Langwadt Christensen, 
-- Aalborg, Denmark



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
2017JanFebMarApr2017

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