Site Home Archive Home FAQ Home How to search the Archive How to Navigate the Archive
Compare FPGA features and resources
Threads starting:
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
Sylvain Giroudon <giroudon@NOSPAM-montrouge.tt.slb.com> writes: >I red http://www.starbridgesystem.com/Pages/about.html recently. >Is this thesis in the same domain/goal as what the guys >from StarBridge are doing? The aims may be the same but I'd imagine Star Bridge's machine performs rather poorly for the majority of computing tasks. FPGA-based machines tend to perform extremely well for a particular domain of tasks - in some cases orders of magnitude faster than general-purpose processors - but perform very, very poorly for the majority of computing tasks. For instance running a compiler on an FPGA-only system could realistically be expected to be orders of magnitude _slower_ than a current-day conventional processor. My aim was to see if the domain over which reconfigurable logic was applicable could be expanded by adapting the architecture to the task at hand. FPGAs were after all designed as a rapid-development platform for static hardware designs and not as computing elements at all. This is why most current attempts at FPGA-based computing machines team FPGA logic arrays with conventional processors. The conventional processors take up the slack where the FPGAs aren't suitable for the job. My thesis describes a type of logic array which is more suitable for general computing tasks than FPGAs are. Whether it's good enough to challenge conventional processors on their own ground is for you to decide. Keep in mind that funds were not available to create a hardware implementation of the device I describe so it's a purely theoretical system. Simulations indicated that the idea definitely had potential - but a great deal more work is required before a full implementation could be realised. Thanks for your interest, Zik SaleebaArticle: 15326
Hi: I have a concern about the synthesis of a state machine. I am implementing a 16-bit mealy state machine using the "One Hot" and "fastest & smallest" options for compliling in FPGA Express. It turns out that one of the outputs of the state machine is functionally set when one of two (logical OR) of these states is set. However, the EDIF netlist out of the tool gives 6 levels of logic to implement this 2-input OR gate. In fact, it is a function of nearly all (14 to be precise) state signals, as opposed to the 2 I would expect. Even more curious is the fact that I see some of these mutually exclusive One-Hot state signals entering LUTs with AND-gate functions. But the output of the AND gate can never be set if the inputs are truly mutually exclusive. Maybe the compiler is providing a "safe" representation, allowing for recovery from illegal combinations of states, but I specifically opted for "fastest & smallest" instead. Any ideas? Thanks, Mike -----------== Posted via Deja News, The Discussion Network ==---------- http://www.dejanews.com/ Search, Read, Discuss, or Start Your OwnArticle: 15327
Are you coding in Verilog or VHDL? If you are coding in Verilog and using a case statement for your 1 hot combinational logic you need to add the comment //synopsys full_case parallel_case on the same line as the case statement. This will reduce the logic considerably. There should be something analagous for VHDL. If you are coding your 1 hot with if statements (if (state[15] ...) you will have problems with Synopsys deciding whether to encode an output function as the OR of 14 1's or 2 0's (assuming 16 states). That problem seems to be avoided if you use the case construct. You can only avoid this with coding style. I have some sample state machines (in verilog) that I can attach that will show all of this and an additional synopsys 1-hot synthesis bug. Turns out that synopsys can decide to replace your "last" 1 hot state with an "all off" state. It won't tell you, but you will be short 1 ff. It then uses a pile of LUT's to detect that state, thereby increasing design size and slowing your design. Xilinx is looking at it at this time. The correct workaround seems to be "use synplicity". bruce michaellewis@my-dejanews.com wrote in message <7cs1ap$gbj$1@nnrp1.dejanews.com>... > > >Hi: > >I have a concern about the synthesis of a state machine. I >am implementing a 16-bit mealy state machine using the >"One Hot" and "fastest & smallest" options for compliling >in FPGA Express. It turns out that one of the outputs of >the state machine is functionally set when one of two >(logical OR) of these states is set. However, the EDIF >netlist out of the tool gives 6 levels of logic to implement >this 2-input OR gate. In fact, it is a function of nearly all >(14 to be precise) state signals, as opposed to the 2 I would >expect. > >Even more curious is the fact that I see some of these >mutually exclusive One-Hot state signals entering LUTs >with AND-gate functions. But the output of the AND >gate can never be set if the inputs are truly mutually >exclusive. Maybe the compiler is providing a "safe" >representation, allowing for recovery from illegal >combinations of states, but I specifically opted for >"fastest & smallest" instead. > >Any ideas? > >Thanks, >Mike > >-----------== Posted via Deja News, The Discussion Network ==---------- >http://www.dejanews.com/ Search, Read, Discuss, or Start Your OwnArticle: 15328
Go back to the older version tools you used before (just a guess, hehe) Assuming you actually removed all the reset statements so there is no reset net, I can't understand why it would change. I think the placement tools use random number seeds based on your xnf file. Perhaps you could try different seeds until you got a placement that works better. bruce Zhen Luo wrote in message <36F09361.1005121B@ee.princeton.edu>... >I have a design that used a global reset signal on XC4000e. Later on I >removed this signal since I decided to utilize the initial "0" state of >all FFs. However, I could never synthesize my design at the clock rate >it used to be. Anyone has the same exprience here? How could I get >around this problem? > >-- Zhen > >Article: 15329
In article <36f15dc2@news.nwlink.com>, Alex Rast <arast@inficom.com> wrote: >Many apologies if the terminology I used offends. I didn't mean it as a >precise, rigidly-defined term with an accepted, standard meaning but rather as >simply a word to convey a concept. How might you phrase the same question >incorporating more acceptable terminology? > I didn't mean to come across as harsh as I did in the last post. The typical question is "How can I make the CLB do f(a,b,c,d)?" Where a,b,c,d are inputs to the CLB. I am still not sure what function is served by trying to generate a fixed CLB set of outputs that appear to be independent of inputs. Perhaps you could enlighten us with a reason behind this need as an alternative approach may be better suited to your needs. >Yes, I actually noticed that I posed the question based on an earlier series >of notes. I noted the same condition you outlined. However, I do still find >one output combination that *does* seem to be impossible, namely YB=1, Y=1, >XB=0. Does this look correct? > You're correct it is functionally impossible to have the above set of outputs in Virtex if you are not allowed to use the BY input as you previously indicated. >I'd gotten so >wound up in trying to think of creative ways of generating certain output >combinations that I knew my head wasn't completely clear. Still curious about this need for specific outputs without mentioning how it relates to inputs. Using a CLB as pwr/gnd source would be huge waste and you don't need more than one pwr and gnd source in a design. Ed McGettiganArticle: 15330
Anybody have any experience with the Xilinx PCI Core and the Orca 3TP12? Willing to share a comparison of the two? -- Tim DavisArticle: 15331
Well, I was using FOUNDATION V1.3, I guess it is old enough :). I removed all the reset net, I couldn't understand why it would change, either. I just got FOUNDATION V1.5, I will try on that to see it works better. Zhen Bruce Nepple wrote: > Go back to the older version tools you used before (just a guess, hehe) > > Assuming you actually removed all the reset statements so there is no reset > net, I can't understand why it would change. > > I think the placement tools use random number seeds based on your xnf file. > Perhaps you could try different seeds until you got a placement that works > better. > > bruce > > Zhen Luo wrote in message <36F09361.1005121B@ee.princeton.edu>... > >I have a design that used a global reset signal on XC4000e. Later on I > >removed this signal since I decided to utilize the initial "0" state of > >all FFs. However, I could never synthesize my design at the clock rate > >it used to be. Anyone has the same exprience here? How could I get > >around this problem? > > > >-- Zhen > > > >Article: 15332
In article <7cs49q$25n$1@feynman.xsj.xilinx.com>, mcgett@feynman.xsj.xilinx.com (Ed McGettigan) wrote: >In article <36f15dc2@news.nwlink.com>, Alex Rast <arast@inficom.com> wrote: .. >The typical question is "How can I make the CLB do f(a,b,c,d)?" >Where a,b,c,d are inputs to the CLB. I am still not sure what >function is served by trying to generate a fixed CLB set of outputs >that appear to be independent of inputs... No, the outputs are not necessarily independent of inputs. But they *are* independent of each other. Thus it would be nice to be able to generate XB=f(a,b,c,d), Y=g(a,b,c,d) YB=h(a,b,c,d) h<>q(f,g), g<>r(f,h), f<>s(g,h). Note that a,b,c,and d need not be the same inputs for each output, and some of them could be always null. >>... I do still find >>one output combination that *does* seem to be impossible, namely YB=1, Y=1, >>XB=0. Does this look correct? >You're correct it is functionally impossible to have the above set >of outputs in Virtex if you are not allowed to use the BY input as >you previously indicated. Which appears, at least on surface, to be simply a pass-through input, i.e. YB=BY, not YB=F(BY). So it barely counts. I'm a little surprised, however, at the broken symmetry between XB and YB - it's possible to have an XB output that is independent of X and Y, but apparently not a YB output. Alex Rast arast@inficom.comArticle: 15333
This is a multi-part message in MIME format. --------------060ADD051CB09F9BA6374B23 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Hi all, I am working on an 8 bit processor and I get an synthesis error that I can't figure out. I have tried to order the code in many different ways, but I always get an error relating to the 'event on one of the signals (inc_pc). Here is the error message I get: -- error line 75: This use of clock edge specification not supported This code simulates fine using Mentor Graphics. Thanks for your time. Jamie Morken --------------060ADD051CB09F9BA6374B23 Content-Type: text/plain; charset=us-ascii; name="jampc.vhd" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="jampc.vhd" -- Program Counter Module -- -- Jamie Morken, Cayce Marston -- Created: March 11, 1999 -- -- REVISION HISTORY -- -- DATE BY WHOM DESCRIPTION -- March 15/99 CM,JM Modified logic -- March 16/99 CM,JM Added load_pc input signal for branch/return -- PC now loadable -- pc.vhd LIBRARY ieee; -- Include the ieee library USE ieee.std_logic_1164.ALL; -- Logical naming of the 1164 part USE work.cpu_pkg.ALL; PACKAGE pc_pkg IS -- Package declaration for the PC COMPONENT pc -- Include component memory PORT ( value_in : IN std_logic_vector(bus_width DOWNTO 0); reset_in : IN std_logic; inc_pc : IN std_logic; read_write : IN std_logic; oe : IN std_logic; b_bus_out : OUT std_logic_vector(bus_width DOWNTO 0) ); END COMPONENT; END pc_pkg; LIBRARY ieee; USE ieee.std_logic_1164.ALL; USE ieee.std_logic_signed.ALL; --USE ieee.std_logic_arith.ALL; USE work.cpu_pkg.ALL; ENTITY pc IS PORT ( value_in : IN std_logic_vector(bus_width DOWNTO 0); reset_in : IN std_logic; inc_pc : IN std_logic; read_write : IN std_logic; oe : IN std_logic; b_bus_out : OUT std_logic_vector(bus_width DOWNTO 0) ); END; ARCHITECTURE behave OF pc IS SIGNAL pc_reg: std_logic_vector(bus_width DOWNTO 0); BEGIN PROCESS (inc_pc, reset_in, read_write, oe, value_in) BEGIN IF (reset_in = '0') THEN -- reset pc_reg pc_reg <= (OTHERS =>'0'); ELSE IF (inc_pc'event AND inc_pc = '0') THEN -- this is the line --with the error. pc_reg <= pc_reg + 1; ELSE IF read_write = '1' THEN -- Read from PC, output to b_bus IF oe = '0' THEN b_bus_out <= pc_reg; -- Output PC to b_bus END IF; END IF; IF read_write = '0' THEN -- Write to PC pc_reg <= value_in; -- Read input regardless of oe END IF; IF oe = '1' THEN b_bus_out <= (OTHERS =>'Z'); -- disable output END IF; END IF; END IF; END PROCESS; END behave; -- error line 75: This use of clock edge specification not supported --------------060ADD051CB09F9BA6374B23--Article: 15334
Hi, I think the problem is that you use an ELSE after the IF signal'EVENT .... this kind of edge specification is not synthetisable !! You can increment (and reset) your pc_reg in an other process kerim el imem Jamie Morken a écrit dans le message <36F21330.216AB342@uvic.ca>... >Hi all, > >I am working on an 8 bit processor and I get an synthesis error that I >can't figure out. I have tried to order the code in many different >ways, but I always get an error relating to the 'event on one of the >signals (inc_pc). Here is the error message I get: >-- error line 75: This use of clock edge specification not supported >This code simulates fine using Mentor Graphics. >Thanks for your time. >Jamie Morken >Article: 15335
Jamie Morken wrote: < snip > > I get an synthesis error that I can't figure out. I have tried to > order the code in many different ways, but I always get an error > relating to the 'event on one of the signals (inc_pc). Here is the > error message I get: -- error line 75: This use of clock edge > specification not supported Your code looks a bit strange to me... I can't see the point of using an ELSE statement after "IF clk = '1' AND clk'EVENT THEN" I also think that the b_bus_out assignments should be done out of this process : b_bus_out doesn't change on reset and is independant of your clock signal. I also have a question : what is the value of b_bus_out when read_write = '0' ? See my version of your architecture below ------------------------------------------------------------------------ ARCHITECTURE behave OF pc IS SIGNAL pc_reg: std_logic_vector(bus_width DOWNTO 0); BEGIN b_bus_out <= (OTHERS =>'Z') WHEN oe = '1' -- disable output ELSE pc_reg WHEN read_write = '1' ELSE ?? ; -- any default value there ? PROCESS (inc_pc, reset_in, read_write) BEGIN IF (reset_in = '0') THEN -- reset pc_reg pc_reg <= (OTHERS =>'0'); ELSIF read_write = '0' THEN -- Asynchronous write to PC pc_reg <= value_in; -- Read input regardless of oe ELSIF (inc_pc'event AND inc_pc = '0') THEN pc_reg <= pc_reg + 1; END IF; END PROCESS; END behave; Nicolas MATRINGE DotCom S.A. Conception electronique 16 rue du Moulin des Bruyeres Tel 00 33 1 46 67 51 11 92400 COURBEVOIE Fax 00 33 1 46 67 51 01 FRANCE mail reply : remove one dot from my address (guess which :o)Article: 15336
Is possible to control the placement in Altera devices (8282 or 10K10 devices) In order to implement a logic circuit at an specific logic module of the FPGA ? Where can I find information about the format of the bitstream used to program the device ? Thanks SebastiaArticle: 15337
Hello Jeff, I designed a board with 2 spartan xcs05 (5V) devices configured by a xc9572 and a flash-prom. Due to an error in schematics, i removed the designed-in pulldown (it should be a pullup). So the mode pin is floating, and during configuration i can measure there a strong 5V-level. After configuration has finished there is no pullup any more; measuring 0V (maybe mode is driving GND ?). Everything works fine. So i am really interested in the behaviour of the xcs05xl-device modepin. (i am still waiting for availability of the xcs05xl in Germany). Jeff Hunsinger wrote: > > Thanks to all who responded. My problem turned out to be leaving the > MODE pin floating and counting on the weak pull-up to keep it high. > > I tied a 10K pullup to VCC and all my problems disappeared. > > Thanks again, > JeffArticle: 15338
On Fri, 05 Mar 1999 16:21:00 -0500, Brian Boorman <XZY.bboorman@harris.com> wrote: >Since I also scrapped the gui in favor of an automatic batch file, I am >beginning to wonder how many others have done the same thing. Since I >Jeff Hunsinger wrote: ><snip> >> .....I quickly >> gave up on it and just run the whole thing from the command line with a >> make file. >> I wish I knew enough about make to write a good, robust makefile. (Yes, as a matter of fact I *DO* live in the stone age!) Any chance of either of you sharing what you've created with other c.a.fpga readers? I wholeheartedly agree that the GUI can be a real pain, and the version control is a bit ...er... strange at times. -tom Tom LeMense TRW Automotive Electronics Farmington Hills, MI facility lemenst@fhsmtp.fh.trw.comArticle: 15339
On Thu, 18 Mar 1999 20:37:49 GMT, Lasse Langwadt Christensen <fuz@kom.auc.dk> wrote: >emanuel stiebler wrote: >> >> Hi, >> >> Brian Drummond <brian@shapes.demon.co.uk> wrote in article >> <36f57d3a.179649091@news.demon.co.uk>... >> > On Tue, 16 Mar 1999 01:08:18 GMT, tcoonan@mindspring.com (Thomas A. >> > Coonan) wrote: >> > >> > >Sounds like a school problem to me! >> > >Go look for carry-lookahead adders in XILINX app notes on their WWW. >> > >>I am currently trying to figure out how to improve an adder. An n-bit >> > >>adder can be constructed by cascading n full adders in series, with the >> > >>carry into stage i, Ci, coming from the output of stage i-1. >> > >> > Yup, looks like homework. >> > >> > But here's a goldmine. >> > >> > http://www.iis.ee.ethz.ch/~zimmi/comp_arith_notes.ps.gz >> >> Doesn't work. You sure this adress is right ? >> >> cheers, >> emanuel > >I you take the "front door" http://www.iis.ee.ethz.ch/~zimmi/ >and look around you sse the correct url is: > >http://www.iis.ee.ethz.ch/~zimmi/publications/comp_arith_notes.ps.gz > oops - finger trouble. Thanks for fixing it. - BrianArticle: 15340
michaellewis@my-dejanews.com wrote: > Hi: > > I have a concern about the synthesis of a state machine. I > am implementing a 16-bit mealy state machine using the > "One Hot" and "fastest & smallest" options for compliling > in FPGA Express. It turns out that one of the outputs of > the state machine is functionally set when one of two > (logical OR) of these states is set. However, the EDIF > netlist out of the tool gives 6 levels of logic to implement > this 2-input OR gate. In fact, it is a function of nearly all > (14 to be precise) state signals, as opposed to the 2 I would > expect. > > Even more curious is the fact that I see some of these > mutually exclusive One-Hot state signals entering LUTs > with AND-gate functions. But the output of the AND > gate can never be set if the inputs are truly mutually > exclusive. Maybe the compiler is providing a "safe" > representation, allowing for recovery from illegal > combinations of states, but I specifically opted for > "fastest & smallest" instead. > > Any ideas? > In the past I have run into the same problem when targetting a Xilinx device, and I think that it is a result of poor design of the synthesizer. As far as I can tell, when FPGA Express decodes the states it does not differentiate between a state machine that is fully encoded (as in a PLD) and a one-hot implementation (as in a FPGA). As a result, it uses all of the state bits to select between states which can lead to severe timing problems in FPGA designs. BTW, I don't know how Synplicity or Exemplar handle this case, but I think that it is a big problem with FPGA Express. My "solution" to this problem was to code my state machine as 3 processes: a combinatorial process, a registered process (for the state machine), and another registered process where I used additional flops that were asserted/deasserted during the states that I was interested in. I then used these additional flops to drive the rest of my logic, and got the expected circuits. Admittedly, this is far from an elegant solution, but it worked. Good luck. Elya Kapelyan > > Thanks, > Mike > > -----------== Posted via Deja News, The Discussion Network ==---------- > http://www.dejanews.com/ Search, Read, Discuss, or Start Your OwnArticle: 15341
This is a multi-part message in MIME format. --------------475081234C2C0861810B6BF4 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit I think that this subjet of power estimation is becoming so pasionated, that surpriseme, cause I thaught that only me was interested in this subjet. I'm agree with Peter, the K factor and the Altera's model has no good bases and it was madden only to provide an idea of the power consumption of Altera's products in some circunstances, after months of measurements, I can tell you that the model is wrong wrong and wrong. I'm agree with Bob also, xilinx didn't and don't make efforts to estimate and/or to optimize power, and reducin VCC is not a good solution, designers wants to run more and more fasters. I don't work with xilinx or Altera neither, I'm trying to analize both and to understand how, why, when and where they consumes. I'll take the Bob's contribution and I'll try to make a contribution bob elkind wrote: > Here's a designer's perspective: > > Suggestion: Provide a spreadsheet type of tool that > > a. groups power consumption "factors" (e.g. gross net capacitance, gross > gate capacitance) by clock net "driving" the power consuming components. > I divided the FPGA architecture in the following big pieces : -LUTs -DFFs -Clock Tree -Interconnect -I/Os -Memory blocks (or CLBs configured as memory). The interconnect can me divided in more pieces (half fast track, full fast track, columns, or simple, double, long lines) Here we have groups with similar (or the same) capacitance. > > b. provides parameters for each group (each group is driven by a different > clock!) for > > i. clock frequency > ii. average toggle factor (# clock cycles between state changes) > For each defined group we can provide parameters, like the average toggling rate of each one and the power supply of each group (some FPGAs uses 3.3 for internal logic and 5 volts for I/Os). Here it's important to defines what the toggling rate is. > > c. multiplies the right correction factors by the entered parameters and > the derived gate count/capacitance numbers, and adds the products to an > estimated power consumption number, taking into account the process > parameters (or range of parameters) for the fpga in use. > > It won't be perfect, its results won't be guaranteed, but it would > still be a *big* help. And my guess is that the data required to "code" > this is already lying around in the engineering dept. > > Does anyone agree that this would help ? Does anyone agree that this > is worth doing ? Does anyone have any better solutions, or improvements ? > Here we have a model that permits us to estimate power with a certain % of error, now the problem is, the toggling rate, first at all, we need REELS test vectors, to estimate the toggling rate, and second, a knlowledge of internal glitches in FPGAs, in asics power caused by glitches can represents the 15% of the total power consumption. For me, estimate power is important but it's not the MOST important, with a tool that allows to estimate it with a 5% or error is fine (maybe by now it's a dream). The most important thing of a knowledge of power consumption in FPGAs is to optimize algoritms using the existing FPGAs and to optimize FPGA's architectures for futures families. Anyway who has the truth in his hand? --------------475081234C2C0861810B6BF4 Content-Type: text/x-vcard; charset=us-ascii; name="garcia.vcf" Content-Transfer-Encoding: 7bit Content-Description: Card for Andres David Garcia Garcia Content-Disposition: attachment; filename="garcia.vcf" begin:vcard n:Garcia Garcia;Andres David tel;pager:http://www-elec.enst.fr/~garcia/index.html tel;fax:(33-1)-45-80-40-36 tel;home:(33-1)-44-16-18-90 tel;work:(33-1)-45-81-78-03 x-mozilla-html:TRUE org:Ecole Nationale Superieure des Telecommunications; Dept. Communications et Electronique version:2.1 email;internet:garcia@elec.enst.fr title:PhD Student on Electronics and Communications adr;quoted-printable:;;46, rue Barrault=0D=0A;Paris;;75634;France fn:Andres David Garcia Garcia end:vcard --------------475081234C2C0861810B6BF4--Article: 15342
This is a multi-part message in MIME format. --------------89EE418778AD5C8CD6B633BC Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit I need information about designs of beiterby encoders or theillis encoders using FPGA. Can anybody helps me? Thank you Andres David --------------89EE418778AD5C8CD6B633BC Content-Type: text/x-vcard; charset=us-ascii; name="garcia.vcf" Content-Transfer-Encoding: 7bit Content-Description: Card for Andres David Garcia Garcia Content-Disposition: attachment; filename="garcia.vcf" begin:vcard n:Garcia Garcia;Andres David tel;pager:http://www-elec.enst.fr/~garcia/index.html tel;fax:(33-1)-45-80-40-36 tel;home:(33-1)-44-16-18-90 tel;work:(33-1)-45-81-78-03 x-mozilla-html:TRUE org:Ecole Nationale Superieure des Telecommunications; Dept. Communications et Electronique version:2.1 email;internet:garcia@elec.enst.fr title:PhD Student on Electronics and Communications adr;quoted-printable:;;46, rue Barrault=0D=0A;Paris;;75634;France fn:Andres David Garcia Garcia end:vcard --------------89EE418778AD5C8CD6B633BC--Article: 15343
Hello, apologies in advance if this question has been asked a thousand times. I am looking for a "concise" comparison document of the current FPGAs on the market. The comparison would include such things as typical maximum internal speeds, volatility, size in blocks and/or gates, prices, strengths (i.e. fast I/Os, fast arithmetic, best for DSP, most routing resources, easy-to-use tools, etc.). It is probably impossible to say "in general, FPGA X is better than FPGA Y", but it is probably correct to say such a thing for a particular application. The question I am trying to answer is: given a particular application, which FPGA from which manufacturer should a designer use? Is there such a document on the Web? In a FAQ, maybe? If not, I would appreciate comments based on personal experience. =============================================== Pierre Langlois Département de mathématiques et informatique Collège militaire royal du Canada tél. (613)541-6000 x6860 B.P. 17000 Succ. "FORCES" fax. (613)541-6584 Kingston ON K7K 7B4 Canada langlois-p@rmc.ca ===============================================Article: 15344
In comp.arch.fpga Zik Saleeba <zik@zikzak.net> wrote: : As far as I'm aware all of the existing attempts at this are just : using conventional off-the-shelf FPGAs. My hardware architecture is : specifically designed to provide support for a full general-purpose : virtual-logic computing environment based around reconfigurable logic : elements. It's rather a big step from FPGAs :) Most currently FPGA boards plug into - at best - a PCI socket. The most optomistic estimate I've encountered for the progress of FPGA-based systems into more mainstream applications envisages that a corner of conventional CPUs gets devoted to programable logic activities, for the cases where a circuit can perform the task more efficiently than an instruction sequence - and that circuit is actually known. I'm not sure that "a full, general-purpose virtual logic computing enviroment" would be seen to be an evolutionary step from current systems, and I wonder if it will not be too much programmable logic too quickly - or, to put it another way, I wonder if it will be too expensive...? -- __________ |im |yler The Mandala Centre http://www.mandala.co.uk/ tt@cryogen.com Magnet: Windows cleaner.Article: 15345
This week's EDTN Tech Note is an article from Lattice Semiconductor that discusses performance and logic utilization tradeoffs between combinatorial and pipelined designs for CPLDs. The article can be found in the Tech Notes section of EDTN's PLD Design Center at www.edtn.com/pld Murray Disman Editor PLD Design CenterArticle: 15346
I think I finally have my line wrap problem solved - here goes: If you have node capacitance and voltage swing data (from the FPGA vendor), a simulator that can generate a node toggle report (ModelSim Elite or ???), and power test vectors (available only from an engineer with some insight into his/her design) it is in principle possible to predict power consumption as accurately as it can ever be predicted. Assuming good data from the FPGA vendor, the onus is then on the engineer to provide test vectors that exercise the design realistically enough to meet the accuracy required. I would hate to see FPGA vendors waste effort on providing glitzy software that just automates the crude estimation methods currently offered when they could (I think) more easily just provide the node data. Of course, if they do both, there is no problem. Your reminder about glitch power was good - I had been hoping that a simple functional simulator might be adequate (pretty easy to write), but glitches caused by delay skew on combinatorial inputs will consume significant power as you say, so timing is required, which means a bit more work for anyone with fantasies of writing (or having a student write) a crude simulator just for power estimation. Andres David Garcia Garcia wrote: > <snipped> > > Here we have a model that permits us to estimate power with a certain % of > error, > now the problem is, the toggling rate, first at all, we need REELS test > vectors, > to estimate the toggling rate, and second, a knlowledge of internal glitches in > FPGAs, > in asics power caused by glitches can represents the 15% of the total power > consumption. > > For me, estimate power is important but it's not the MOST important, with a > tool that allows > to estimate it with a 5% or error is fine (maybe by now it's a dream). The most > important > thing of a knowledge of power consumption in FPGAs is to optimize algoritms > using the > existing FPGAs and to optimize FPGA's architectures for futures families. > > Anyway who has the truth in his hand? > > Tom Burgess -- Digital Engineer National Research Council of Canada Herzberg Institute of Astrophysics Dominion Radio Astrophysical Observatory P.O. Box 248, Penticton, B.C. Canada V2A 6K3 Email: tom.burgess@hia.nrc.ca Office: (250) 490-4360 Switch Board: (250) 493-2277 Fax: (250) 493-7767Article: 15347
I am looking for information on building a small Bit Error Rate Tester inside a Xilinx. I have a Xilinx FPGA controlling a communications channel between a PCI card and an external cage. I am thinking it would be useful to place a Bit Error Rate Tester in the Xilinx to test the link and to determine its maximum transfer rate. It would be simple to generate a pseudo-random pattern to transmit down the link, but I cannot think of an easy way to sync back up to the received data, and count errors. I know this has been done before, I am just trying to avoid reinventing the wheel. Thanks DanArticle: 15348
On Wed, 17 Mar 1999 14:58:34 -0800, Tom Burgess <tom.burgess@hia.nrc.ca> wrote: >Hmm. Looks like this would do just fine. I assume that this >was from the ModelSim Elite ~$20,000 (U.S.) tool? Regrettably Indeed. EE is a great product, but for those who don't need such esoteric features, PE is truly excellent. Until the FPGA vendors come out with such figures and tools that can utilise such information and are accurate; there is little point (IMHO) in offering a feature to a market that cannot make use of it. Who knows what the future may bring? How about it Xilinx and Altera? Let's see a new slogan on those web sites: "Mines smaller than yours" (power consumption that is). >this exceeds my nonexistent tools budget by a couple of orders >of magnitude, but it does serve as an existence proof. My condolences. Cheers Stuart For Email remove "NOSPAM" from the addressArticle: 15349
Did you ever try the //synopsys full_case parallel_case pragmas? bruce Elya Kapelyan wrote in message <36F26991.7C8640DE@pentek.com>... > > >michaellewis@my-dejanews.com wrote: > >> Hi: >> >> I have a concern about the synthesis of a state machine. I >> am implementing a 16-bit mealy state machine using the >> "One Hot" and "fastest & smallest" options for compliling >> in FPGA Express. It turns out that one of the outputs of >> the state machine is functionally set when one of two >> (logical OR) of these states is set. However, the EDIF >> netlist out of the tool gives 6 levels of logic to implement >> this 2-input OR gate. In fact, it is a function of nearly all >> (14 to be precise) state signals, as opposed to the 2 I would >> expect. >> >> Even more curious is the fact that I see some of these >> mutually exclusive One-Hot state signals entering LUTs >> with AND-gate functions. But the output of the AND >> gate can never be set if the inputs are truly mutually >> exclusive. Maybe the compiler is providing a "safe" >> representation, allowing for recovery from illegal >> combinations of states, but I specifically opted for >> "fastest & smallest" instead. >> >> Any ideas? >> > > In the past I have run into the same problem when targetting a Xilinx >device, and I think that it is a result of poor design of the synthesizer. >As far as I can tell, when FPGA Express decodes the states it does not >differentiate between a state machine that is fully encoded (as in a PLD) >and a one-hot implementation (as in a FPGA). As a result, it uses all of >the state bits to select between states which can lead to severe timing >problems in FPGA designs. BTW, I don't know how Synplicity or Exemplar >handle this case, but I think that it is a big problem with FPGA Express. > > My "solution" to this problem was to code my state machine as 3 >processes: a combinatorial process, a registered process (for the state >machine), and another registered process where I used additional flops that >were asserted/deasserted during the states that I was interested in. I >then used these additional flops to drive the rest of my logic, and got the >expected circuits. > > Admittedly, this is far from an elegant solution, but it worked. > > > Good luck. > > Elya Kapelyan > > >> >> Thanks, >> Mike >> >> -----------== Posted via Deja News, The Discussion Network ==---------- >> http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own >
Site Home Archive Home FAQ Home How to search the Archive How to Navigate the Archive
Compare FPGA features and resources
Threads starting:
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