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
I am new to the fpga world. I have some Altera flex10k devices that I want to use to implement a routing protocol. Does anyone know where I can get some (simple) example codes that use the ISA bus to interact with fpga/cpld devices. I spent all of yesterday, unsuccessfully, trying to write/read to/from ram on a device. Also, is there a web site that explains the ISA bus signals? Thanks, Gersham Jeffrey.Article: 11501
G. R. Jeffrey wrote: > > I am new to the fpga world. I have some Altera flex10k devices that I > want to use to implement a routing protocol. > > Does anyone know where I can get some (simple) example codes that use > the ISA bus to interact with fpga/cpld devices. I spent all of > yesterday, unsuccessfully, trying to write/read to/from ram on a device. > > Also, is there a web site that explains the ISA bus signals? The ISA bus is not very complicated, although it's an incredible mess. For the signals, have a look at http://www.gl.umbc.edu/~msokos1/isa.txt I'm afraid you won't find the timing specs on the net (spent a lot of time and never found any), but some 386-compatible microcontrollers implement an ISA-like bus. There are timings in the data-sheets. Try for example the AMD Elan SC300 Nicolas MATRINGE DotCom SA Conception électronique 16 rue du Moulin des Bruyères Tel: 00 33 1 46 67 51 00 92400 COURBEVOIE Fax: 00 33 1 46 67 51 01 mail reply : remove one dot from the address (guess which :-)Article: 11502
Reza Bohrani wrote: > Does anyone know of any VHDL implementation of the decoding of a > manchester-encoded bitstream. If no does anyone have any suggestion of > how > to attack teh problem? > I published a very simple circuit ( in schematic form, not in VHDL) in the Xilinx quarterly magazine XCELL, in 1995, issue number 17. You can get it on the Xilinx web site by either clicking on the search button and typing manchester decoder, or you go straight to http://www.xilinx.com/xcell/xl17/xl17-30.pdf Peter Alfke, Xilinx ApplicationsArticle: 11503
The UKs best embedded development tools site just got BETTER ------------------------------------------------------------ COMSOL has added over 30 new pages to its web site. We now cover 80 different microprocessor families. http://www.Computer-Solutions.co.uk New products include: Embedded C & C++ for Motorola micros Flash and ROM Emulators Super fast Ethernet BDM Emulators x86 and Power PC Real Time Executives Embedded TCP/IP, Web Server and Browser Embedded DOS & W95 Compatible file system along with all the old favorites In-circuit emulators Assemblers C & C++ compilers Real-time executives Software simulators 186-486 linkers Remote debuggers EPROM emulators EPROM-PAL-GAL-micro programmers GANG programmers CASE tools Forth systems from chips to Windows RS232 Debuggers PC instruments (Logic Analysers & DSOs) ----------------------------------------------------------------------- If you would like information on any of our current products, or to be emailed with info on new embedded products please email to the address below ------------------------------------------------------------------------ If you would like an entry ticket to the Embedded Systems Exhibition at Ascot, Berks on the 8th & 9th September send me an email Regards, Chris ----------------------------- Chris Stephens E-mail: sales@computer-solutions.co.uk Computer Solutions Ltd. Phone & Fax: +44 (0)1 932 829 460 1a New Haw Road, Addlestone, Surrey, KT15 2BZ England http://www.Computer-Solutions.co.uk For the largest range of embedded microprocessor development tools in the UKArticle: 11504
In article <35DA98AF.7104446@virgin.net>, tony.cooper@virgin.net wrote: > Does anyone know where I can source the following tools ALL running > under LINUX (Direct or under Xwindows). > > The TMS320C series of tools (ie. the C compiler, the assembler, linker > etc etc) > The Xilinx FPGA Implementation tools. > The XDS510 JTAG Emulator tools. > > I have all the above for Windoze, but I have just about had enough of > this awfull O/S... I would like some stability in my life, and I am > fairly sure that LINUX is the way to go. (I used to run Linux all the > time - it never once crashed on me). The WINE windoze emulator will likely run those tools on Linux, then you can use XEmacs and jump to the next compile error. cheers -- Rick -----== Posted via Deja News, The Leader in Internet Discussion ==----- http://www.dejanews.com/rg_mkgrp.xp Create Your Own Free Member ForumArticle: 11505
This is a multi-part message in MIME format. --------------6ABDE92354A7FF0401DC9D51 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit watm wrote: > I designed the blocks in the shematic, simulated them, checked that > everything was like i expected to be and .... when i loaded the code > into the FPGA, nothing happened the expected ! When you simulated, was that a functional simulation ( no device orrouting delay ), or a timing simulation ( speed back-annotated from the placed and routed device ) ? It makes a huge difference, particularly with high-speed synchronous designs and most asynchronous designs. --------------6ABDE92354A7FF0401DC9D51 Content-Type: text/x-vcard; charset=us-ascii; name="vcard.vcf" Content-Transfer-Encoding: 7bit Content-Description: Card for John L. Smith Content-Disposition: attachment; filename="vcard.vcf" begin: vcard fn: John L. Smith n: Smith;John L. org: Visicom Imaging Products adr: 1 Burlington Woods;;;Burlington;MA;01803;USA email;internet: jsmith@visicom.com title: Principal Engineer tel;work: 781-221-6700 tel;fax: 781-221-6777 x-mozilla-cpt: ;0 x-mozilla-html: FALSE version: 2.1 end: vcard --------------6ABDE92354A7FF0401DC9D51--Article: 11506
A reduced version of Altera's Max+Plus II software is available free for the downloading and it handles several parts, I believe including the FLEX 10K10, a 10,000+ gate SRAM-based device. You may have to make the PCB that uses it, but you can download the FPGA program to it over the PC's serial or parallel ports. You don't have to buy the BitBlaster or ByteBlaster download cables; just download the data sheet for them - you can make either yourself from the info in the datasheet. Lattice also has free tools but the devices supported for free are more toward their low-end parts (< 5K gates). Good luck mail83870@pop.net wrote in article <6qtirr$ees$1@nnrp1.dejanews.com>... > > > Can anyone recommend a little tinkertoy kit ( no less than oh, say, 10K > gates, tho' ) that I can slap into my intel box ( linux/win* ) and play > with. Bonus points for neat graphical interface. I am hoping to spend > darn little..... > > Thanks in advance for any help.... > > -----== Posted via Deja News, The Leader in Internet Discussion ==----- > http://www.dejanews.com/rg_mkgrp.xp Create Your Own Free Member Forum >Article: 11507
Has anyone out there had problems compiling a design in ver. 8.3 and found that some of the logic in the .gdf doesn't show up in the Floorplan Editor? I am new to Max+Plus II, so this could be a simple case of cockpit error. However, Altera's tech support had no explanation after reviewing the design files. I have the SIMPLEST thing you can make - a 16-bit input latch feeding data to a 16-bit bidirectional data bus in a FLEX part. The thing compiles fine, but when I look at the Floorplan editor, only the lower byte exists in logic. All input pins and bidir pins are allocated, but no fan-in shows up for the upper byte of the latch, and no logic utilization shows for the 8 FF's for the upper byte. Go figure. Also, any problems with Max+Plus II on NT4.0, or editing it briefly on a Win95 machine and going back to NT? Thanks for your feedback. Tom Burns tomburns@worldnet.att.netArticle: 11508
[ part 1 of 2: the conference itself & mostly front-end EDA tools ] !!! "It's not a BUG, jcooley@world.std.com /o o\ / it's a FEATURE!" (508) 429-4357 ( > ) \ - / The Fifth Annual ESNUG/DAC Awards: _] [_ "Mike Barnicle, Plagiarism, & DAC'98" - or - "106 Engineers Review DAC'98 in San Francisco, CA, June 15-19, 1998" by John Cooley Holliston Poor Farm, P.O. Box 6222, Holliston, MA 01746-6222 Legal Disclaimer: "As always, anything said here is only opinion." "It is a tale, told by an idiot, full of sound and fury, signifying nothing." - from Shakespeare's "MacBeth" Gentle Readers: Living under the Big-Brother-Is-Watching, Politically Correct shadow of the Boston Globe (as I do because I live in Eastern Massachussetts), a day now doesn't go by where you can't see on TV or read in the press some new preachy, self-rightous "journalistic integrity" rant concerning how the columnist Mike Barnicle plagiarised eight jokes from comedian George Carlin's book, "Brain Droppings". And you want to know what *I* think about this? I very sarcastically say: "Big deal. Big fucking deal. Who the hell cares?" And why I say this is because when I write, I haven't penned an original idea in months and I really don't intend to change that way of thinking any time soon! (In fact, I sometimes even go *out* *of* *my* *way* to avoid having new ideas!) Why? Because truely new ideas are a lot of work to cook up, and, most of the time, they turn out to be half-baked anyway. God! -- I can't tell you the number of times, as a man, where I've had some woman in my life (whether it be my girlfriend-du-jour, me mum, a friend, or a co-worker) tell me how my innovative approach to some issue was "in error", "wrong", "insensitive" or just plain "dumb-ass stupid". My male married friends tell me the three stock phases that keep them out of the divorce courts are: 1.) "Honey, I'm sorry, I was wrong." 2.) "Babes, I made a mistake; you were right." 3.) "No, Sweetie, you don't look fat in that dress." Get out of the personal relationship domain, and, say, try bringing up new ideas at work, and it gets worst! Now your female *and* male co-workers, bosses, associates, competitors, and even random-people-off-the-street-who- don't-even-know-you will all collectively gang up to *publically* tell you, in no uncertain terms, how *YOUR* new idea was so totally wrong, so totally NOT the solution, and so totally OUT-TO-LUNCH. I'm not kidding here. Human societies, on RARE occassion, will *sometimes* reward the crackpots who go against the herd *if* the herd happens to decide *at* *that* *very* *moment* it wants to go in their new direction -- otherwise the *punishment* for thinking differently will be swift and sure. And either way, they're still seen as crackpots. Some successful crackpots were: Martin Luther (the Protestant Reformation), George Washington (the American Revolution), and Bill Clinton (established that U.S. Presidents can do *anything* [ including chubby interns ] as long as the U.S. economy is thriving). Some examples of failed crackpots: the thousands of "heretics" burned at the stake for disagreeing with the Church; the millions of "terrorists" who fought on the wrong side of revolutions; and the millions of husbands and wives *immediately* divorced after being caught cheating. And there's even an "Honorable Mention" group -- crackpots who changed the world but paid for thinking differently anyway: Jesus Christ (formed new religion; died on cross), Joan of Arc (liberation of France; burned at stake), Socrates (founded Western philosophy; convicted pedophile forced to drink poison), and again, Bill Clinton (two term US President during sizzling ecomonic boom; forced to remain married to Hillary Clinton). So, the message here for anyone with half a brain is clearly: "Keep your yap shut and stop using that half of your brain that does work -- if you don't want to be taking some SERIOUS risks with your life, liberty, and pursuit of happiness!" The real wisdom *I've* found that works comes from the newspaper columns "Hints From Heloise", "Ann Landers", "Dear Abby", and "Miss Manners" -- and that wisdom is it's always best to print *OTHER* people's letters, their quotes, their problems with their solutions, and their new ideas _waaaaaay_ before you even think of toying with the idea of presenting your own original material. To me, it's obvious that Mike Barnicle knew this smart money approach to writing columns when he published George Carlin's jokes. I do it, too, with the 1,000 (or so) user letters I republish in ESNUG every year and even in this DAC'98 Trip Report. In this very DAC'98 Trip Report you'll be lucky if you'll get 2 percent original material from me; the remaining 98 percent of the ideas, opinions, and even the explainations of how the new EDA technologies supposedly work have all been unabashedly clipped/quoted/stolen by me from the 106 *other* DAC'98 Trip Reports *other* people wrote when they returned home. This particular DAC Trip Report doesn't just emit a faint odor of plagiarism -- it reeks of it! And I wish to personally thank Mike Barnicle and George Carlin for finally making this style of column writing now publically acceptable! Or at least that's the way I see it. I could be wrong. "Hypothetical Question: If married man speaks out loud in a forrest and there's no woman there to hear him, is he still wrong?" - joke overheard in local bar "It's a good thing Bill Clinton finally confessed. The investigation was getting tough when they DNA tested that dress... because they just found out that all the people from Arkansas have the same DNA." - another joke overheard in a local bar LIVING LIFE LARGE: Reflecting the economic boom going on in the overall U.S. economy, the total DAC'98 attendance grew by about 25 percent compared to DAC'97. That is, DAC'97 had 17,398 total attendance; DAC'98 had about 21,675 total attendees (+/- 50 people). But that tells only part of the story because total attendance includes EVERYONE -- professors, employees of EDA vendors, guests, *and* customers. An even larger growth of roughly 41 percent (from DAC'97 to DAC'98) happened if you look at the numbers for the DAC "Technical Attendees" -- i.e. the people actually interested in *buying* the EDA tools being exhibited. DAC'97 had 7,387 technical attendees; DAC'98 had 10,488 technical attendees. Comparing this year's and last year's exhibitor list yields DAC'98 grew by 21 percent with 38 new companies showing -- if you just counted names on the exhibitor list. Do a little digging and you'll find that 31 of those exhibitors were semiconductor houses -- so, looking at totals, the number of *EDA* exhibitors has actually remained at around a constant 190. But that 190 has had 40 vendors disappear and 44 new vendors replace them; making the yearly turnover of EDA companies of around 22 percent. Of those who have disappeared, some (like Cascade, Compapass, CCT, & Wellspring) were bought out by the bigger EDA fish; a few (like Speed Electronic, & KBS) simply went out of business; a couple did name changes (like Indus); and, for a good number of the MIAs (like DS Diagonal, Emultek, Nextwave, Neuw, & Technical Data Freeway) it's not widely known why they no showed. "We believe that genuine technology innovation in the EDA market emerges from focused startups rather than the major public players. One non-scientific measure is the presence of at least 40 new startups at DAC'98 compared with a year ago in DisneyLand." - from a financial analyst's report of DAC'98 "The best giveaways were Dilbert toys from a bogus company whose demo I only virtually observed (luckily). My daughter will be pleased!" - a well-known Silicon Valley ASIC designer MONEY _CAN_ BUY YOU LOVE: For the past two years, the DAC Plenary Sessions (otherwise commonly known as the "DAC CEO Panels") have been totally lame-o events. Prior to then, it was completely different. For example, at DAC'95, Joe Costello (then CEO of Cadence) blurted out: "We're stuck in a fixed-pie model. Have you seen three big dogs hovering over one bowl of dog food? It's not a pretty picture." Aart De Geus (then CEO of Synopsys) came back with: "If you think of yourself as a dog, you deserve dog food!" Alain Hanover (then CEO of ViewLogic) said about his rebellious Chronologic: "Our founding forefathers guaranteed us the right to life, liberty, the pursuit of happiness and the right to sue each other." Gerry Hsu (then CEO of ArcSys) took pot shots at Cadence by describing how bureaucratic it was when he used to work there. But amougst these rancorous exchanges we also got to openly discuss EDA bugs, interoperability issues, pricing models, EDA vendors becoming consulting houses, Verilog vs. VHDL, and much more. Yes, sometimes egos were bruised, and sometimes people laughed -- yet we all walked away knowing a bit more about the other guy's moccassins than before the event. But, for the past two years, the DAC CEO panels have become controlled, scripted affairs where more and more, less and less is being said. They've replaced actual EDA users on the panel with academics and CEO "users" who use bullshit management words like "paradigm shift", "resources", and "P&L statements" in their day-to-day vernacular instead of engineering TLAs like "ECO", "LRM", or "BFD". The audence is no longer allowed to ask questions. And the soft, safe, fluffy questions asked of the CEOs were asked weeks in advance so their staffs could craft long-winded soliloquies carefully showing how *their* company's approach would save the day. Even a problem child CEO like Gerry Hsu of Avant! (who has been embroilled in a long legal spat with Cadence) was nowhere to be seen -- lest he accidently cause a free exchange of ideas to happen. "Why are they doing this?", you ask. The answer is Wall Street. Never before has DAC seen so many investment bankers, analysts, and financial types running around in an *engineering* conference. They're all trying to get the inside scoop on chip design software. And the EDA execs are trying to parlay that interest into higher stock prices by now *appearing* to play nice with each other in public. Frank discussions with EDA CEOs are a thing of the past now that Wall Street is here. The Beatles were wrong with their song "Money Can't Buy Me Love". In the EDA industry, we all love each other. Honest. Money *can* buy you love these days. Or at least on DAC CEO panels it can. "One other panel was of note was Gary Smith's annual DataQuest briefing Sunday night where he reported DQ was forecasting a 10-15% downturn in EDA tool sales. Gary indicated that this was due to the users not believing the tools coming out could handle leading edge technologies and are slowing down their purchases." - an anonymous EDA executive "Thirty Five Years of Design Automation Panel: It was an interesting discussion but there were no real highlights. They showed some history and said their predictions of the future (from the past) were so bad that they did not want to do it again. They discussed tools/methods and said (manager version here): Tools have got to get better; so they will. Engineers have got to get smarter; so they will. IP happens. Any questions? - a Telecom designer TRYING TO MAKE FORMAL BIG: Hoping to capitalize on the severe need for verification engineers versus design engineers (with some estimates putting the need as high as 5 verifiers for every designer) this year's DAC was flooded with new Formal Verification tools all promising A Golden New Way To Debug Designs. There are basically three kinds of formal verification tools: equivalence checkers (EC), model checkers (MC), and pie-in-the-sky. EC is the most common type of formal verification tool. Two of the EDA big boys, Synopsys and Cadence, are even pitching their brand of EC to be used, with their brand simulation and static timing analyzers (of course) as "next generation sign-off". And it's widely rumored that Mentor R&D is in beta with an EC tool and a static timing tool to follow suit. "The demo of Prime Time and Formality was quite compelling. (Motive was not represented in the booth, verifying that Synopsys is trying to kill it ASAP in favor of Prime Time.) The interface for Prime Time is very user friendly, with wizards and tabbed dialog boxes and spreadsheets (cribbed from the interface of Synopsys's PC-based FPGA Compiler). All this generates scripts that are compatible with the commands that control Design Compiler. Prime Time also has statistical analysis tools that are graphically tied into the Design Analyzer generated schematics which helps you quickly locate the root cause of timing problems. Formality, a logic equivalency checker, also has quite a slick front end, and it is also tied into Design Analyzer. Their demo showed how Formality could be quickly used to verify a hand-modified netlist (the netlist was modified to correct a timing fault caught by Prime Time). As a corollary, Formality could be used to check blocks that have been implemented in schematics by hand (because of speed or efficiency problems with synthesized logic) against the RTL code." - a design engineer from a large semiconductor house Technically, an Equivalence Checker (EC) works by comparing a golden design (RTL or gates) against another design that's supposed to be functionaly the "same". EC can easily deal with whole projects (500k - 1M gates) overnight, but it depends heavily on both designs being structurally simular. Designs that are structurally different generally have to be broken into 10k gate modules for EC to work. Most EC tools support VHDL, Verilog and EDIF. Set-up and day-to-day usability are still issues, but debugging is really what separates the tools. Even though EC is used in parallel with simulation at most of the bigger semiconductor companies, EC tools are still overpriced for most engineers' desktops ($50k to $150k); they have to drop to around $20k to go to main stream... Assuming the $12.5 million Formal Verification market (1997 Dataquest est.) really does go main stream. "If you draw a big circle to represent verification, simulation as a weapon still fills 90% of the job. Formal verification maybe covers 5%-15% and again it is hard to measure how much it overlaps with the other tools. For what it costs, it doesn't do much." - A West Coast design engineer "ASIC flow breakdown: Design and block level verification: 27% Simulation: 46% Emulation: 15% Structural checking: 12% Staffing is 2:1 of verification engineers to design engineers." - Al Silbert of Nortel (on the Formal Verification Panel) "Chrysalis' Design Verifyer (EC) despite being hard to set up and expensive, has a nice Tcl interface for user programmability, and a good schematic based debugger. It has the advantage of being the oldest, most debugged EC tool on the market. They're also trying for MC, in a limited but pragmatic way. The great advantage Formality from Synopsys (EC) has is it integrates nicely into a Design Compiler based design flow. In addition, Formality has one of the best EC debugging interfaces. The criticism that Formality and Design Compiler may contain shared code misses the mark slightly, but that's a discussion too long to go into here. Cadence's Affirma (EC) is a "me, too" product. It integrates FV into their simulation environment, which makes for only one user interface to learn. Their use of Composer to display a schematic provides a familiar user interface, but doesn't allow the user to use the schematic to explore the design. Abstract (EC and MC) and Verysys (EC and MC) are both repackaged Siemens technology. These tools are not as easy to use, nor as easy to integrate into a flow as their major competitors. Verysys' debugging interface to their EC tool that is very competitive, though. It's rumored that Intel has bought stock in Abstract, but that hasn't made an impact on their tools yet. Avanti's Lynx (EC) is just VFormal from the Compass take-over, & hasn't been touched since the take-over. Verplex (EC), Formalized Design (EC & MC) and Fujitsu (EC & MC) are all very late newcomers this year to the market. Mentor is also rumored to be working on an EC tool, too. It's doubtful how any of these tools will fare in this already crowded market." - Simon Read, a Formal Verification consultant & ex-developer "Equivalence Checking is now so mature that they were almost selling it from hand carts out in the street like it was hotdogs. Not so sure who was buying though." - a European design engineer "a) Chrysalis: Oldest, overpriced. b) Synopsys Formality: Best interface with Synth and timing c) Verisys: Looks promising, but nothing stands out d) Abstract FV: Cool debug with auto simulation and waveform output e) Bell Labs: Formalcheck V: This looked the best last year f) Formal Verification: Small company in the back room Too many players for such a niche market. Hopefully one or two of them will "do an MTI" and have a useable product at a low price and cause a fallout and a launch into mainstream use." - Peet James of Qualis Design Model Checking (MC) answers "what if?" and "is it possible that?" questions about a single design. MC only works on small designs of about 300 flip-flops or less because it suffers from "state explosion" -- the time and/or memory needed doubles with every state bit you add. MC can detect and help fix subtle interacting FSM bugs -- which is why engineers like about MC -- because it catches bugs that would be hard to catch via most simulation runs. Market-wise, Bell Labs, 0-in, and the DAC'98 newbies Silicon Forrest and Formalized Design are all basically home grown MC tools. Abstract and Versys are reselling warmed-over MC technology they got from Siemens years earlier. (Oddly, it's rumored that Cadence's Ken MacMillan put a *free* model checker out on the Cadence web site.) "FormalCheck (MC) from Bell Labs (BLDA) can process larger designs than its competitors. Their graphical user interface has very well thought out features to help write & manage queries and novel, useful debugging features. They've added a batch oriented interface for automating regression tests. Let's see how they fare after Cadence buys them." - Simon Read, a Formal Verification consultant & ex-developer "The Bell Labs, 0-in, and Silicon Forrest (MC) tools are different in that they let the user query (or they do directed searches) for very esoteric bugs in designs which makes them technologically popular. Every engineer secretly frets over that one missed Pentium bug that haunts them for the rest of their careers." - an East Coast design engineer "Formal Verification is still sold as insurance. The sales pitch is basically a scare tactic. 'You need this $150K tool, or you may miss a bug'. Engineers love finding these hyper elusive bugs, and can get lost in the search for them. It wastes time and robs from mainstream target bug finders like simulation. Again, it is very hard to see if Formal Verification is cost-effective or not." - Peet James of Qualis Design "The best way to change methodologies is to start slowly, as Synopsys did when it came to market with gate-level optimization. Only later did Synopsys move people upstream to RTL synthesis. Contrast that to Silc, which tried to introduce behavioral synthesis in the mid-1980s, which turned out to be 10 years too early." - Richard Goering of EE Times SYNTHESIS GOES FORMAL? In ancient Greek mythology "centaurs" were creatures who were half-human & half-horse. Other than being novelties, "centaurs" never partook in anything of consequence in Greek myths. The EDA world's version of centaurs is "Formal Synthesis" -- a useless beastie that's half-formal and half-synthesis. Normal synthesis works by running step-by-step transforms on RTL code to eventually convert it to optimized gate level designs. A typical synthesis transform may be "initial mapping of RTL to gates" or "inserting extra buffers according to specific fan-out rules" or "doing specific boolean optimizations on logic cones". Formal Synthesis anal retentively ensures that each transform used is "proven" mathematically so no possible errors going from spec to gates ever happen. "Nobody makes designs with flawless specs. We sort of craft the spec and the design along the way with the verification guys looking over our sholders. We know 90 percent of what we're trying to design going in. It's that last 10 percent that makes designing interesting. I don't want errors in my designs, but the promises and limitations of formal synthesis are about as useful as tits on a bull." - a design engineer from Research Triangle Park, North Carolina "Abstract and Derivation Systems both market formal synthesis tools, both claim that their tools allow designers to 'explore the solution space'. To most designers, who are often unskilled in mathematical techniques, these tools will 'constrain the solution space' rather than simply 'prevent mistakes'. Formal synthesis is really only applicable where a lack of bugs is a paranoid, overriding concern." - Simon Read, a Formal Verification consultant & ex-developer "IP", YOU PEE, WE ALL PEE ON "IP": Remember how at last year's DAC each EDA vendor annoyingly somehow found a way to work the phrase "Deep Sub Micron" into their customer pitches? Well, this year's Most Tired Phrase used at DAC was "IP" -- Intellectual Property. Everyone and their grandmother somehow fenagled a way to work *their* product as being somehow absolutely *critical* for IP use and/or developement. "Are you making some IP?" Call Cadence, Mentor, Synopsys, blah, blah, blah, (the list goes on and on) for their tools. "Want to actually *buy* some IP?" Call Argonaut, Artisan, ARM, Aspec, Avanti, Cadence, CAST, CoreEl, CSELT, Denali, Design & Reuse, Duet, Eonic, Interra, Mentor, Phoenix, Sagantec, Sierra, Silicon Access, Synopsys (and whoever else I've missed.) And, to make matters worst, now all the God Damned foundries and semiconductor houses are getting into the same bloody IP game, too! (2nd try): "Want to buy some IP? " Call IBM, TSMC, Motorola, LSI,... And, not to be left out of the party, all the FPGA/CPLD vendors are sharing in the same terribly unoriginal idea, also! (3rd try): "Psst! Hey, Buddy, Wanna buy some IP?" Call Xilinx, Altera, Actel, Cypress,... Enough IP already! "Great parts produce great vices, as well as virtues." - from Plato's fourth letter to Dionysius II, the ruler of Syracuse "Parts is parts." - the catch phrase to a Wendy's Old Fashioned Hamburger restaurant ad campaign of a few years years ago "Do you remember a few years ago that televised Republican National Convention where the candidates kept strongly supporting 'family values' over and over, yet none of them could define what 'family values' were? IP was the 'family values' of [this year's] DAC." - a Texas design engineer "Personally, I don't get it -- all this hype around IP. Essentially the IP vendors want to be tax collectors. They want a nickle a chip for every chip their IP is in. We've seen dozens of IP vendors but haven't invested in any of them." - Mark Stevens of Sequoia Capital, a venture capital firm "Cadence had a special meeting for all their big time customers in which they were spreading their propaganda of the future of the IC industry. In this meeting they discussed how the future industry will require more system integrators than logic designers because of the IP industry. Their projection is that IC suppliers will incorporate more third party IP cores and less custom designs. After they were done, several Cadence marketing guys were seated at each table to discuss our reaction to their presentation. At that point several people at our table began to turn their theory into swiss cheese." - a Lucent engineer "Sand Microelectronics: I could not bare to talk to these guys in light of all the problems the [name deleted] team has had." - an anon design engineer IN LAWYERS WE TRUST: Very traumatically at this year's DAC, the IP business has run into its very first technological breakdown. The funny thing is that many engineers may have not picked up on this yet. The Great Divide that has separated IP has been how it has been delivered to the end customer; it's either soft core IP (like Verilog or VHDL source code or netlists) or hard core (like polygon oriented GDS-II files). Soft cores have traditionally had protection/reverse-engineering problems concerning how to keep questionable characters from simply copying it, tweaking it a bit, and then reselling it as their own unique IP. To fight this, LEDA's Kypton does confusing VHDL-to-VHDL scrambling; Chronologic's VMC converts Verilog into unreadable compiled code, and now Topdown offers something simular. But all these solutions each have some technical or business Achilles' heel leaving soft core IP vendors still nervously relying on legal threats to protect their family jewels. Conversely, hard cores, being essentially polygons, have been much, much harder to reverse-engineer, and, as a result, have been intrinsically more secure. That is, until now... Laughingly, the one EDA company most plagued with lawsuits from day one of its inception, Avanti, is now widely marketing a tool it got from the Compass aquisition called Lynx-LB that REVERSE COMPILES full-custom block implementations (i.e. GDS-II files) back into synthesizable Verilog or VHDL RTL! Formalized Design also offered a reverse compiler that reads in EDIF or SPICE files and regenerates RTL Verilog or VHDL. And it's rumored that Cadence and Sagantec might be working on something similar. Ostensively such tools are to make design migration of old IP into new applications doable -- but now what's to stop a Cadence or Synopsys Consulting Services or even your TSMC support engineer from using these very same tools to "migrate" your hot IP into their own cache of designs? Now the hard core IP vendors (and now even *anyone* submitting designs to foundries) have to do what the soft core vendors have been doing all along: nervously rely on legal threats to protect your IP family jewels. "Circuit Semantics has a tool called DynaBlock that can take an Hspice netlist (extracted from a full-custom layout using Arcadia or Star-RC), recognizes functional gates from the transistor connectivity, and characterizes the functional gates. The user gets a gate-level Verilog netlist and a Synopsys timing library (*.lib) for the gates in the netlist. This tool is useful to extract functionality and timing from legacy cores or hard IP's. They claim it can process 100k transistors in three hours. Compared to Avanti's Lynx-LB: Dynablock extracts a gate-level Verilog netlist, not an HDL boolean model like Lynx-LB. Dynablock extracts timing models that are acceptable to standard cell methodologies, such as Synopsys' Design Compiler. Avanti claims that SGS Thompson has used Lynx-LB successfully on designs of 900k transistors in about 20 minutes. Synopsys/Epic's Coremill is supposed to be similar to Lynx-LB and DynaBlock, but we never get a chance to see it." - a Silicon Valley chip designer "Lay not up for yourselves treasure upon earth, where moth and rust doth corrupt, and where thieves break through and steal." - Matthew 6:19 "YES, WE HAVE NO BANANAS" Take a college course in logic & sematics sometime and you'll find yourself in a final exam analyzing Venn diagrams about apparently contradictory statements like "Yes, we have no bananas". Yes, while in the full scan world, Mentor's Fastscan and Synopsys' Test Compiler still remain the biggest technology contenders; the biggest splash at DAC came from ATG Technology with its powerful NO SCAN ATPG tool. And again, yes, Sunrise is quite well known in the partial scan world, from some customer's reports it appears ATG Tech's NO SCAN ATPG tool seems to be getting very good results generating significant coverage for designs that have NO SCAN access whatsoever. Whoa! The only other no scan solution more popular was BIST, which has improved dramatically in the last few years. "Logicvision was by far the most impressive. While having all our buffers defined as BiDi's could be somewhat of a task for the tool, the ability to add BIST to an analog cell was very impressive. They demonstrated the tool's accuracy by adding BIST to a PLL. The algorithm would measure and characterize the PLL, including the very to difficult to measure jitter. While the simulation demo was enough to convince me, Logicvision went even farther by setting up a demo board with an off-the-shelf PLL which the tool characterized by downloading the BIST logic into an FPGA (sorry guys, it was "X"). While the BIST logic tested the PLL, an oscilloscope was setup to probe the various points on the PLL to show how closely the waveforms matched the simulated results. The accuracy looked to be within 5%. The advantage of this is that only one tester is needed to test both analog and digital." - an East Coast engineer (Lucent) "I'm curious about Duet's automatic scan chain insertion for IP blocks. I'll be over at their booth." - an anon engineer while on the DAC floor "TSSI presented their test insertion product and test vector preparation product in which their demo hung. The test insertion product was poor compared to Sunrise or Synopsys test compiler. The test vector product looked state of the art in terms of automation, capability, flexibility etc. Functional blocks are selected in a GUI, interconnected and the process kicked off from the GUI, although the most capable this is the demo that hung. Something bothers me when demos of test products hang." - an Arizona chip designer DARWIN GOES ESDA AWARD: Goes to Summit Design. With i-Logix gone, Speed dead, KBS gone, Vista gone, Mentor redoing Renior to have the *exact* look & feel of Summit's Visual HDL, and Escalade not capturing even a fraction of the ESDA market share they originally bragged they'd have, Summit design has moved on to expanding into code coverage, HW/SW co-design, waveform viewing, text2graphics, and is even flirting with behavioral synthesis by sharing its DAC bed with its new live-in girlfriend, Dasys. While some of the new ESDA players, like the awkwardly named mouthful "China Integrated Circuit Design Center" (offering it's Panda VDE visual design capture/debug/sim toolset) aren't being taken seriously because nobody wants to risk *their* career on EDA software of questionable support & reliability -- one new set of ESDA tools, Novas' DeBussy, surprisingly caught the fancy of many user's eyes. On top of all the usual ESDA FSM stuff, users were wowed at how Debussy recognizes synthesis elements early in the design process and checks for synthesis syntax errors; how nSchema annotates a schematic with SDF info plus clock tree analysis plus having asynchronous signal paths recognized; how nTrace displays memory contents plus driver and load tracing. "Novas' Debussy is a major leap forward in GUI's, compared to Virsim (which I sell) and SignalScan, which still looks like the old SunWindows. Debussy takes the best ideas from many tools and adds in synthesis to make a real winner. I don't know how it performs on large designs, but the demo looked great. The source code browser annotates the code with signal values, as does the schematic tool. The schematic for their simple circuit was made very readable by doing partial synthesis to recognize muxes and ALUs. Lastly, their state machine tool did a nice job of creating a diagram, but the example only had 5 states. How good was the tool? I saw the Virsim project manager watching a demo! - a Virsim salesman "The best niche tool? Debussy by Novasoft. It was a really cool Verilog debugging tool. You could even capture the contents of memory and view it through time. All the windows are very well integrated." - a design engineer from Texas "Escalade: I am going to bring Escalade in to demonstrate their DesignBook front end HDL design tool. This product competes with Renoir and VisualHDL. I am a little concerned because they have focused on building tools to protect Intellectual Property this year. The new tools are based on DesignBook but they refocused their target market this year. If they spread themselves too thin (like MicroSim did last year) they will be gone next year. In my opinion, DesignBook is one of the top three front end tools. Summit: When I checked out Summit's VisualHDL, I told them they had the best graphical front end design tool for Verilog. The only problem was that I couldn't afford it. They said to let them work on a way to solve that problem. (Don't hold your breath. <G>) Renoir: We had an in house demo of Renoir two weeks before DAC so I didn't look at it at DAC. Renoir is my first choice right now. It is less risky than Escalade's DesignBook and more affordable than Summit's VisualHDL." - a Telecom oriented designer "John, I just wanted to express my thanks for the jump-start your ESDA design shoot out of two years ago had on Mentor. Renoir, our high-level design capture tool, has now passed the 25% market share point and is rapidly displacing Summit. It wouldn't have happened without the wake-up call your contest gave to our "System Architect" product. It is now a multimillion dollar per year high profit generator for Mentor. My thanks." - Wally Rhines, CEO of Mentor Graphics, in ESNUG 278 #3 BIG IRON VS. LE TOUR DE FRANCE: Two years ago at DAC'96, at least a dozen companies (SpeedSim, Frontline, Fintronic, Synopsys, Vantage, Chronologic, Pendulum, Cadence, Cadence Alta, Synopsys, Mentor (rumored), and CAE Plus) had or were working on a cycle-based simulator. It was like the R&D staffs from these companies all called the same psychic hotline for advice. Now, two years later, Pendulum disappeared, Fintronic & CAE Plus are still around, and the rest of the cycle-based simulators have been gobbled up in the EDA food chain to end up either in Cadence, Synopsys, Avanti, or Zycad -- with the major players being Synopsys and Cadence in a weird way. Cycle-based simulators evaluate boolean equations -- which means they're great at simulating gate-level logic as long as you're ONLY interested in FUNCTION and NOT TIMING. Non-cycle-based simulators handle TIMING very well, but generally aren't nearly as fast as cycle-based simulators. As a result, there are very few EDA users of pure cycle-based simulators. (For example, Cadence's COBRA only has seven customers). But, the Chronologic whizkids found a way to merge Roadrunner cycle-based technology within their compiled Verilog VCS such that VCS runs cycle-based on the non-timing parts of simulations and compiled on the remaining parts. (It's rumored that Cadence is trying to do this with NC-Verilog but no one can confirm it.) Now let's look at the hardware emulation/acceleration world which is in essence a speeding up of the simulation of gates. Quickturn, Mentor, Aptix, IKOS, and Zycad all got into this business *before* cycle-based simulation was widely used. Now that EDA software has found a savvy way to speed up gate simulation, those old half-million dollar Big Iron solutions suddenly aren't so appealing to EDA users. (It's not that software simulation beats Big Iron in performance -- it's that software now has sped up simulation enough to make pricey Big Iron more of a luxary than necessity.) The Big Iron guys are fighting for their survival. That's why Quickturn sued Mentor to block them from selling Big Iron in the US. (And that lawsuit is why Mentor has recently put out a takeover bid on Quickturn.) Survivial is also why Quickturn bought out ARKOS from Synopsys last year (even after aquiring equivalent an ARKOS-like technology, IBM's COBALT) -- and it's why Quickturn did a If-You-Can't-Beat-Them-Join-Them purchase of little old 33 employee SpeedSim in Jan '97 for the inflated price of $55 million. It's also why Quickturn wrapped this all together to squeeze out its Mercury Design Verification system which is essentially an event-driven Verilog behavioural simulator with 74 FPGAs and 2 PowerPCs tacked on to speedily simulate a million gate design (w/ pricing of $395K for 500K gates and around $1 million for a 2 million gates.) Gotta fight to survive! Big Iron old timer IKOS and Big Iron DAC newbies Axis and VeriPOD have all, more or less, mimicked Quickturn technologically in trying have customers simulate RTL on some type of reconfigurable hardware. So far, financially Quickturn and IKOS have announced revenue drops. Nobody knows how the newbies Axis and VeriPOD will fare. "Subject: For Sale -- One QUICKTURN hardware emulator We have a Quickturn hardware emulator which we no longer wish to support in our flow. Anyone interested in this device should contact my boss, David Burleson at: David.Burleson@Sciatl.com" - Tom Coonan Scientific Atlanta (from ESNUG 295 #9) "Axis: combines debuggability and speed. Look, Act and Feel like a software simulator running at hardware speed. - Simulates up to 2M gates on 8 boards plugged into PCI bus on Sun. - Costs around $250K. Limitation is cost of high end Altera parts. - Speed 10K - 100K cycles/sec - No synthesis - directly executes behavioral RTL. - Start simulation with $rccon to run fast, then switch to $rccoff shortly before failure to do a pure software debug. Would be great to have this for $10K so that every designer could have one. Its value at $250K is a lot harder to justify." - a Silicon Valley chip designer "I think it stinks how easily EDA companies can chop and change their product line leaving customers high and dry. It has happened so many times in the last 10 years I am suprised anyone wants to buy CAD tools any more! - a European EDA user commenting on Synopsys selling ARKOS to Quickturn two weeks after DAC'97 "Mentor is OEMing QuickHDL-XLC from Topdown Design. Topdown calls it Cyclops. According to TD salesman, it's a Verilog pre-processor that levelizes even-driven simulations to make them cycle-based. Sells for $23,000 which is cheaper than NC-Verilog or VCS." - an anon engineer NICE HORSEY! In the early days of EDA, when the dinosaurs roamed the Earth and people knew who Carver Mead was, before synthesis, there used to be something sorta like proto-synthesis that was called "datapath compilation". In fact, there was even a company called Silicon Compilers that used to sell datapath compilers. And, it being the Age of EDA Dinosaurs and all, Mentor Graphics was around and Mentor did what it does best. It bought Silicon Compilers and destroyed it. Time went by. Synthesis was discovered. Silc and Synopsys and Synergy and Exemplar and Booledozer and Veribest and Synplicity and ACEO and Ambit came along. Some died, a few thrived, and a few are still struggling to survive even today. And datapath compilation suddenly reappeared out of nowhere. A very small company, Silerity, down in Los Angeles, offered a nascent tiling datapath compiler, PathBlazer, in '94. Viewlogic then bought Silerity in '95, and after a failed attempt at trying to market PathBlazer, View followed that time honored Mentor tradition and killed it. Time passed. Synthesis giant, Synopsys, bought SiArc and found it had accidentally acquired a SiArc in-house datapath compiler in the purchase. Secret microphones planted within Synopsys R&D at the time picked up the following: "Ewww... A datapath compiler... How disgusting! Let's erase the hard drive and pretend we never found it...", then there's a garbled disagreement, followed by the sounds of a physical struggle. Months later, Synopsys markets an equation-based/non-tiling-based SiArc datapath compiler under the name "Module Compiler". And around the same time, two other unrelated companies, Tera Systems and Arcadia, also introduce their own tiling-based datapath compilers, and a long lost EDA niche is reborn. And, although Module Compiler appears to be making a lot of money for Synopsys, it appears that a lot of users at this DAC were quite mesmerized by Arcadia's Mustang. Nice horsey! "Additionally, our pulse of field contacts on which new products are hot revealed some surprises: ... [ stuff deleted ] ... Module Compiler (datapath synthesis tool) is enabling SNPS sales reps to exceed quota. We were skeptical about SNPS' marketing claims when the product was announced 1-2 quarters ago, that Module Compiler could sell 1-for-1 with Design Compiler (control logic synthesis). Most SNPS' field reps agree that the 1:1 ratio is too generous, but the product is ramping exceedingly well and customers like it (we independently discussed with users)." - from a financial analyst's report of DAC'98 "The most interesting tool was Arcadia Mustang because the approach is so novel. It takes your regularly synthesized netlist and deduces the regularity from it to do datapath placement. I'm surprised it could work, but apparently they have success with it. IBM included the tool in their ASIC flow. There was also an interesting paper in the DAC Proceedings evaluating the results of commercial datapath compilers. They don't name names, but you can have fun trying to figure out which vendor they're talking about :-)" - a California engineer A BATTLE OF PROXIES: Whereas Synopsys likes to see itself as a "serious" player in the FPGA/CPLD synthesis game, the truth is that the real players in that market are Exemplar, Synplicity, the freeware FPGA synthesis tools that the FPGA vendors give out themselves (like Altera's MaxPlus II), and the it-might-as-well-be-freeware software that Minc/Synario/IST "gives" out. Last year's DAC was Synplicity's "coming out party" -- literally they had grown to the point where they were sponsoring parties big enough at DAC that other EDA rivals were attending uninvited -- Synplicity had finally arrived in EDA. This year's DAC was no different: Bob Russo, the Synopsys Senior VP of Worldwide Sales & Services, was unexpectedly seen at the Synplicity after hours Top Of The Marriott party! On the technical side, Synplicity announced an FPGA floorplanning tool that caught the interest of many engineers -- but no one reported on it in technical detail. Another company that has a well-known FPGA floorplanner is Morphologic with its "MorphMCFP". Targeting the big FPGA's (like the Lucent 2C, 2T, 3C, & 3T series plus the Xilinix XC4000E/EX/L/XL series), MorphMCFP lets engineers automatically translate/partition designs between FPGAs with complete compliance to each device's design rules. It does automatic I/O allocation and de-allocation (as needed) and the floorplanner instantly retimes the newly placed design as feedback to the engineer. Is Synplicity's new FPGA floorplanner this sophisticated? We don't known. On the business side, Synplicity also demostrated a wee bit of business acumen in cutting a deal to OEM ModelTech's combined Verilog-VHDL simulator. (Model Tech & Exemplar are both divisions of Mentor! Talk about each division being its *own* profit & loss center within Mentor! Whoa!) "I spent some time in Synplicity's booth looking at their new release. They have added a bunch of new debug features but nothing drastic. Symplicity and Exemplar keep pushing each other. Their fight for marketshare based on features and price is great for users. Exemplar still has the best value in Synthesis. The new Spectrum release looks great. I will have to play with it before I pass judgement on whether it works as well as it looks. It has orders of magnitude more control of the synthesis process than Synplicity at a fraction of the cost of Synopsys. The new constraint editor is the first usable one I have seen on any synthesis tool. I heard from an Exemplar wheel that Exemplar's growth last year was equal to Synplicity's total revenue. Both Exemplar and Synplicity have realized that the quality of their synthesis is often compromised by FPGA place and route tools. They are integrating the vendor specific P&R tools into their GUIs and linking the synthesis and floor planning tools. - a West Coast designer "Marketing and marketing clout will beat technology every time." - Mark Stevens of Sequoia Capital, a venture capital firm "THE MAN WITHOUT A COUNTRY" In American folklore there's a famous story about some poor sap who committed a horrible crime and flippantly told the judge that he wished he didn't have a country. The guy was sentanced to spending the rest of his life aboard sailing ships never setting foot in *any* country. ACEO is like this, but it's "A Company Without A Customer". For over seven years I've run a quasi-guerilla style mailing list (ESNUG) that has over 6,000 EDA users (worldwide) reading and responding to it weekly. In its entire seven year history I have, as of yet, to find anyone "on" or "off the record" ever talk about or even hint at having thought of trying ACEO's "Gatran" synthesis tool. Not one. Never. Nada. "Yes, we don't know what you're talking about." And yet ACEO claims to have had *hundreds* of customers over these past four years! And now ACEO annouced "Asyn" -- a "one pass hierarchical synthesis tool" -- and "Softwire" -- a multi-FPGA partitioning tool -- at this year's DAC. Hello, McFly! Is anyone home? Hello! "Again, this is a loaded question. Richard, I don't have an answer for all of that. Two year have passed. The last time I was on the CEO panel was two years ago. During these two years a lot things have happened. OK, many people think, uh, said things about me, about us. But one thing you not says is that I have not brought entertainment value to you, this industry." - Gerry Hsu, CEO of Avanti, while on the DAC'97 CEO Panel, responding to the question of why EDA start-ups seem so successful in developing new technologies so quickly. "You've had a lot of fun -- a *lot* of fun -- that you've had an opportunity to attack me. Just think how much you're going to be missing. You won't have Nixon to kick around any more, because, gentlemen, this is my last press conference." - Richard Nixon after he lost the 1962 election to be the governor of California Next Week: [ part 2 of 2: the back-end & the other new EDA tools/ideas ] =========================================================================== Trying to figure out a Synopsys bug? Want to hear how 6,000+ other users dealt with it? Then join the E-Mail Synopsys Users Group (ESNUG)! !!! "It's not a BUG, jcooley@world.std.com /o o\ / it's a FEATURE!" (508) 429-4357 ( > ) \ - / - John Cooley, EDA & ASIC Design Consultant in Synopsys, _] [_ Verilog, VHDL and numerous Design Methodologies. Holliston Poor Farm, P.O. Box 6222, Holliston, MA 01746-6222 Legal Disclaimer: "As always, anything said here is only opinion."Article: 11509
I need to find the algoritm or the block diagram of the 4PPM (Pulse Position Modulation). If someone has the algoritm inplemented in VHDL or ABEL, its perfect. If someone give a hand, i will appreciated. Thanks in advance, Rui PintoArticle: 11510
I would like to see if there are existing algorithms to solve the following minimization problem. Consider the expression: x1 + 2 * x2 + 3 * x3 + 4 * x4 + 5 * x5 + 6 * x6 + 7 * x7 + 8 * x8 If only multiplications by powers of two are allowed, how can this expression be re-written (using associativity, distributivity, etc.) such that the number of additions is minimized? You can use a subexpression, like a=x1+x2, in the final expression. I have several slightly different answers that use 11 additions, but none fewer. For example, one such set of expressions that uses 11 is: z=x6 + x7 y=x1 + x3 + x5 + x7 + 2*(x2 + x3 + z) + 4*(x4 + x5 + z + 2*x8) Can you get less than 11? More importantly, is there an algorithm that solves minimization problems of this nature? The purpose of it is for circuit synthesis, and I have come up with a simple algorithm which yeilds results that are linear in size. The closest method to it is Distributed Arithmetic, but the ROM size in DA grows exponentially. Anything like this out there? Thanks, Dan Benyamin benyamin@ulca.edu -----== Posted via Deja News, The Leader in Internet Discussion ==----- http://www.dejanews.com/rg_mkgrp.xp Create Your Own Free Member ForumArticle: 11511
Tim Forcer wrote: > > We have two Data I/O Chiplab 48 "project" programmers, which are nice > units. But they are unusable under Windows NT due to the dreaded port > access problem. Data I/O's UK agents state that no software upgrade > will allow function under NT, and we've tried various tricks involving > public-domain drivers such as giveio.sys without success. > > Anybody got any bright ideas? If not, we'll have to maintain a couple > of Win95 machines just for these units. > > Or you could purchase Partition Magic or System Commander and set up your machines with "dual-boot" options for Win NT and 95. -- Barry A. Brown Microwave Instruments Division Hewlett-Packard Company **** Remove the "nospam" from my email address *****Article: 11512
benyamin@my-dejanews.com says... > I would like to see if there are existing algorithms to solve the following > minimization problem. > > Consider the expression: > > x1 + 2 * x2 + 3 * x3 + 4 * x4 + 5 * x5 + 6 * x6 + 7 * x7 + 8 * x8 > > If only multiplications by powers of two are allowed, how can this expression > be re-written (using associativity, distributivity, etc.) such that the number > of additions is minimized? You can use a subexpression, like a=x1+x2, in the > final expression. I have several slightly different answers that use 11 > additions, but none fewer. For example, one such set of expressions that uses > 11 is: > > z=x6 + x7 > y=x1 + x3 + x5 + x7 + 2*(x2 + x3 + z) + 4*(x4 + x5 + z + 2*x8) > > Can you get less than 11? > More importantly, is there an algorithm that solves minimization problems of > this nature? Yes, there is an algorithm (as to whether or not you can beat 11, I have no idea). It takes some imagination to write it, but its not impossible. I have solved a couple very similar problems to this using purely algorithmic methods. The idea is to simply enumerate all operations possible, on some initial state of registers in an orderly fashion. After each operation, a new state is achieved as some incremental cost. Suppose these represents vertex's of a graph. Then it is a matter of reaching a certain node (or any node with a certain property) of your graph with the minimum cost. This reduces it to a graph theory/network flow problem. See: http://www.pobox.com/~qed/amult2.html for one example. Depending on how many auxilliary register you assume, your problem space is fairly large, so I'm not sure you could directly search through the space to find the answer, unless you have a super computer with a ton of memory at your disposal. It may be that there is a pattern for (x1 + 2*x2 + ... + n*xn) so you could solve the problem for small n first to try to extrapolate the solution for n=8. -- Paul Hsieh qed@pobox.com http://www.pobox.com/~qed/Article: 11513
Distributed arithmetic immediately came to mind, and as I read down, I see you also considered it. You can use a combination of distributed arithmetic, bit serial and scaling accumulators to minimize the look up table size. Your example below could be done with a 256 x 6 look-up table followed by a scaling accumulator (the feedback on the accumulator shifts one bit with respect to the input). The eight variables are serialized so that table address is one bit from each. So for each execution, you do n table lookups (n is the number of bits in your variables). The results from each lookup are added to the shifted sum of the previous partial results. After n cycles you have the complete result in the accumulator. If you require more speed (less clock cycles), you can duplicate the table as needed to do 2 or more bits at a time. Now, the table itself can be broken down too. Since each table entry is 1*a + 2*b + 3*c + ... + 8*g, you can break this into two smaller tables: 1*a + 3*c + 5*e + 7*h and 2*b + 4*d + 6*f + 8*g and then use an adder to combine the look up results. This results in two 16 x 5 tables and a 6 bit adder instead of one 256 * 6 table. This decomposition can be extended as needed to reduce the size of the tables to a convenient size. I hope this helps! benyamin@my-dejanews.com wrote: > I would like to see if there are existing algorithms to solve the following > minimization problem. > > Consider the expression: > > x1 + 2 * x2 + 3 * x3 + 4 * x4 + 5 * x5 + > 6 * x6 + 7 * x7 + 8 * x8 > > If only multiplications by powers of two are allowed, how can this expression > be re-written (using associativity, distributivity, etc.) such that the number > of additions is minimized? You can use a subexpression, like a=x1+x2, in the > final expression. I have several slightly different answers that use 11 > additions, but none fewer. For example, one such set of expressions that uses > 11 is: > > z=x6 + x7 > y=x1 + x3 + x5 + x7 + 2*(x2 + x3 + z) + 4*(x4 + x5 + z + 2*x8) > > Can you get less than 11? > More importantly, is there an algorithm that solves minimization problems of > this nature? > > The purpose of it is for circuit synthesis, and I have come up with a simple > algorithm which yeilds results that are linear in size. The closest method to > it is Distributed Arithmetic, but the ROM size in DA grows exponentially. > Anything like this out there? -- -Ray Andraka, P.E. President, the Andraka Consulting Group, Inc. 401/884-7930 Fax 401/884-7950 email randraka@ids.net http://users.ids.net/~randraka Andraka Consulting is a digital hardware design firm specializing in high performance FPGA designs for signal processing, computing and control applications. See us at WESCON'98 and the Military Applications for PLDs Conferences.Article: 11514
This solution can be implemented in a slightly different order to minimize the total delay to 4 adders. All solutions will require at least 4 adder delays. (assuming only full adders are allowed) C+2F E+G A+2B 4D+8H | | | | 3C+6F 5E+5G A+2B+2G | | | | | 3C+5E+6F+5G A+2B+4D+2G+8H | | A+2B+3C+4D+5E+6F+7G+8H Pat Kling <klingp@ct.picker.com> writes: > > Form A + 2B + 3C + 4D + 5E + 6F + 7G + 8H with 10 additions > > I = C + 2F > J = I + 2I = 3C + 6F > K = E + G > L = K + 4K = 5E + 5G > M = L + 2G = 5E + 7G > N = M + J = 3C + 5E + 6F + 7G > O = N + A = A + 3C + 5E + 6F + 7G > P = O + 2B = A + 2B + 3C + 5E + 6F + 7G > Q = P + 4D = A + 2B + 3C + 4D + 5E + 6F + 7G > R = Q + 8H = A + 2B + 3C + 4D + 5E + 6F + 7G + 8H > > I've got no idea if it can be done in less. Certainly it would require > a minimum of 8. > > Patrick Kling (klingp@ct.picker.com)Article: 11515
This is a multi-part message in MIME format. --------------B164B1F222E7F106329A2E81 Content-Type: multipart/alternative; boundary="------------85D0649013CC5DE10D78A794" --------------85D0649013CC5DE10D78A794 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit benyamin@my-dejanews.com wrote: > I would like to see if there are existing algorithms to solve the following > minimization problem. > > Consider the expression: > > x1 + 2 * x2 + 3 * x3 + 4 * x4 + 5 * x5 + > 6 * x6 + 7 * x7 + 8 * x8 > > If only multiplications by powers of two are allowed, how can this expression > be re-written (using associativity, distributivity, etc.) such that the number > of additions is minimized? You can use a subexpression, like a=x1+x2, in the > final expression. I have several slightly different answers that use 11 > additions, but none fewer. For example, one such set of expressions that uses > 11 is: > > z=x6 + x7 > y=x1 + x3 + x5 + x7 + 2*(x2 + x3 + z) + 4*(x4 + x5 + z + 2*x8) > > Can you get less than 11? I don't think so Dan, add up the number of 1-bits (or 0-bits +1 if this isless) in the co-efficients, then subtract 1 (because you always start with one): coef: 1 2 3 4 5 6 7 8 #of adds(or subs): 1 1 2 1 2 2 2 1 -> Total-1 = 11 > More importantly, is there an algorithm that solves minimization problems of > this nature? I'm sure there are, but don't have any ref's, hopefully someone else will. > The purpose of it is for circuit synthesis, and I have come up with a simple > algorithm which yeilds results that are linear in size. The closest method to > it is Distributed Arithmetic, but the ROM size in DA grows exponentially. > Anything like this out there? DA is nice, it is a regular design technique, easily implementedby program, and sometimes you want that regularity. But, I've been seeing through experience that most (all I've tried so far) fixed arithmetic circuits can be worried into a more compact format, if you have the time to pull the numbers apart and find patterns. A hand-crafted color-space converter I did saved ~30-35 CLBs (30%) over the DA equivalent circuit. I don't know if this is always the case. I'd be very interested if you get any ref's. --------------85D0649013CC5DE10D78A794 Content-Type: text/html; charset=us-ascii Content-Transfer-Encoding: 7bit <HTML> <P>benyamin@my-dejanews.com wrote: <BLOCKQUOTE TYPE=CITE>I would like to see if there are existing algorithms to solve the following <BR>minimization problem. <P>Consider the expression: <P>x1 + 2 * x2 + 3 * x3 + 4 * x4 + 5 * x5 + <BR>6 * x6 + 7 * x7 + 8 * x8 <P>If only multiplications by powers of two are allowed, how can this expression <BR>be re-written (using associativity, distributivity, etc.) such that the number <BR>of additions is minimized? You can use a subexpression, like a=x1+x2, in the <BR>final expression. I have several slightly different answers that use 11 <BR>additions, but none fewer. For example, one such set of expressions that uses <BR>11 is: <P>z=x6 + x7 <BR>y=x1 + x3 + x5 + x7 + 2*(x2 + x3 + z) + 4*(x4 + x5 + z + 2*x8) <P>Can you get less than 11?</BLOCKQUOTE> I don't think so Dan, add up the number of 1-bits (or 0-bits +1 if this isless) in the co-efficients, then subtract 1 (because you always start with <BR>one):<TT></TT> <P><TT>coef: 1 2 3 4 5 6 7 8</TT> <BR><TT>#of adds(or subs): 1 1 2 1 2 2 2 1 -> Total-1 = 11</TT> <BLOCKQUOTE TYPE=CITE> <P>More importantly, is there an algorithm that solves minimization problems of <BR>this nature?</BLOCKQUOTE> I'm sure there are, but don't have any ref's, hopefully someone else will. <BLOCKQUOTE TYPE=CITE>The purpose of it is for circuit synthesis, and I have come up with a simple <BR>algorithm which yeilds results that are linear in size. The closest method to <BR>it is Distributed Arithmetic, but the ROM size in DA grows exponentially. <BR>Anything like this out there?</BLOCKQUOTE> DA is nice, it is a regular design technique, easily implementedby program, and sometimes you want that regularity. <BR>But, I've been seeing through experience that <BR>most (all I've tried so far) fixed arithmetic circuits can be <BR>worried into a more compact format, if you have the time <BR>to pull the numbers apart and find patterns. A hand-crafted <BR>color-space converter I did saved ~30-35 CLBs (30%) <BR>over the DA equivalent circuit. I don't know if this is <BR>always the case. <P>I'd be very interested if you get any ref's. <BR> </HTML> --------------85D0649013CC5DE10D78A794-- --------------B164B1F222E7F106329A2E81 Content-Type: text/x-vcard; charset=us-ascii; name="vcard.vcf" Content-Transfer-Encoding: 7bit Content-Description: Card for John L. Smith Content-Disposition: attachment; filename="vcard.vcf" begin: vcard fn: John L. Smith n: Smith;John L. org: Visicom Imaging Products adr: 1 Burlington Woods;;;Burlington;MA;01803;USA email;internet: jsmith@visicom.com title: Principal Engineer tel;work: 781-221-6700 tel;fax: 781-221-6777 x-mozilla-cpt: ;0 x-mozilla-html: FALSE version: 2.1 end: vcard --------------B164B1F222E7F106329A2E81--Article: 11516
In article <6rhdeq$dh4$1@nnrp1.dejanews.com>, benyamin@my-dejanews.com wrote: | Consider the expression: | | x1 + 2 * x2 + 3 * x3 + 4 * x4 + 5 * x5 + | 6 * x6 + 7 * x7 + 8 * x8 | | If only multiplications by powers of two are allowed, how can this expression | be re-written (using associativity, distributivity, etc.) such that the number | of additions is minimized? You can use a subexpression, like a=x1+x2, in the | final expression. I have several slightly different answers that use 11 | additions, but none fewer. For example, one such set of expressions that uses | 11 is: | | z=x6 + x7 | y=x1 + x3 + x5 + x7 + 2*(x2 + x3 + z) + 4*(x4 + x5 + z + 2*x8) | | Can you get less than 11? Are subtractions also allowed (invert "b" operand and set carry in)? y = x1 - x3 + x5 - x7 + 2*(x2 + x6) + 4*(x3 + x4 + x5 + x6) + 8*x8 -- -- Tim OlsonArticle: 11517
Hello to everybody : - ) I am New on fpga programming and I try to implement simple video interface whitch I already do it with real componets, is there somebody can help me with some very simple examples. THANKS : )Article: 11518
Hello to everybody : - ) I am New on fpga programming and I try to implement simple video interface whitch I already do it with real componets, is there somebody can help me with some very simple examples. THANKS : )Article: 11519
Hello to everybody : - ) I am New on fpga programming and I try to implement simple video interface whitch I already do it with real componets, is there somebody can help me with some very simple examples. THANKS : )Article: 11520
Anybody out there got a good solution for a half full flag for a fifo? I was planning on using Xilinx sync dual port ram to get a 16 deep fifo. The write clk and read clk are app the same frequency (~53 mhz) but are from 2 different xtal sources so are async to each other. My fifo basics are rusty...I've always used a full u/d counter to do this in the past, but can't with this design due to the different clocks. I know that when the read pointer = the write pointer the fifo is either full or empty, but don't know what relationship indicates half full. Could someone help me out? The flag does not need to be exact, and I can re-sync to eliminate any glitches. Even a false half full indication would not cause a problem as long as it happens right around the half full point. Thanks DanArticle: 11521
Press Release / For Immediate Release Silicore Corporation Releases 8-Bit RISC Microprocessor - VHDL IP Core ---------------------------------------------------------------------- August 17, 1998 - Minneapolis, MN USA. Silicore Corporation announced today the release of its latest VHDL microprocessor IP core. Called the Silicore(TM) SLC1655, the core provides a complete 8-bit RISC processor solution for use on FPGA or ASIC devices. Typical applications include sensors, medical devices and consumer electronics. The core is unique because it allows the user to create a complete, single chip microcontroller right at their own desk. The Silicore(TM) SLC1655 requires only 10% - 25% of an FPGA part (including RAM and instruction ROM), leaving plenty of room for added peripherals. Furthermore, the core can be fabricated using a wide variety of package types, speeds and temperature ranges. For high volume production, the core can be moved into an ASIC device. Wade Peterson, President and CEO of Silicore, stated that: "This product opens up exciting new possibilities for electronic designers. It creates a whole new genre of solutions available to the end user. They can completely control the design, and add whatever peripherals and I/O's they want. With this product they no longer have to pay expensive NRE charges, or wait weeks or months for prototype microcontrollers." The Silicore(TM) SLC1655 core includes an 8-bit RISC processor, RAM, ROM, counter/timer and powerdown logic. A complete evaluation kit, using a Lucent Technologies ORCA(R) FPGA, is available. Typical operating speeds are 5-10 MIPS. It is delivered as a VHDL 'soft-core', and is portable across many brands of FPGA parts such as those available from Lucent, Xilinx and Altera. The Silicore(TM) SLC1655 is software compatible with the PIC(R) series of microcontrollers from Microchip Technology, Inc. [Over 600 million PIC(R) processors have been sold since 1990]. Many third party software tools such as assemblers, simulators and 'C' compilers are available for the Silicore(TM) SLC1655. Furthermore, the core runs at twice the speed of the Microchip parts. For more information see the Silicore web site at http://www.silicore.net, or contact: Jeff Pearson at 800-583-7484. Silicore is a service mark and trademark of Silicore Corporation. ORCA(R) is a registered trademark of Lucent Technologies Inc. PIC(R) is a registered trademark of Microchip Technology, Inc. Silicore Corporation is not affiliated with Microchip Technology, Inc.Article: 11522
Dj wrote: > I am New on fpga programming and I try to implement > simple video interface whitch I already do it with real componets, is there > somebody can help me with some > very simple examples. Soo, FPGA's aren't real components?? Shhh, don't tell my clients! As far as your problem, can you be more specific. I doubt anyone is going to post a complete video controller design as an example. -- -Ray Andraka, P.E. President, the Andraka Consulting Group, Inc. 401/884-7930 Fax 401/884-7950 email randraka@ids.net http://users.ids.net/~randrakaArticle: 11523
Dj wrote: > Hello to everybody : - ) > > I am New on fpga programming and I try to implement > simple video interface whitch I already do it with real componets, is there > somebody can help me with some > very simple examples. > > THANKS : ) > Hi,. Try this ... http://www.acte.no/freecore/vga_sync.htm I hope it can be helpful... If you have a more specific ideas on implementing VGA, let us know. Good luck to you! -- ˙Ĝ˙àArticle: 11524
Dan Kuechle wrote: > > Anybody out there got a good solution for a half full flag for a fifo? [snip] > ... I > know that when the read pointer = the write pointer the fifo is either full > or empty, but don't know what relationship indicates half full. I believe this occurs when write_ptr - read_ptr = fifo_size / 2. The subtraction has to be performed modulo fifo_size. --Jim
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