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
The IEEE International Workshop on MEMORY Technology, Design, and Testing will be held August 11-12 in San Jose. Probably it is a 5 minute walk for you :) Won't you come? Details can be found at <ftp://ftp.cs.tamu.edu/pub/fmeyer/service/mtdt97/abstracts.html>Article: 6401
University of Aberdeen Department of Engineering Parallel Processing Research Group Research Studentship (EPSRC) =========================== An EPSRC funded research studentship is available at the Department of Engineering, University of Aberdeen, from October 1997. Supervision is offered in the Parallel Processing Research Group which includes work on DSP and imaging applications, Artificial Neural Networks, Programmable Hardware. The studentship provides UK/EU student fees, and standard maintenance payments (UKP 5190) for UK students. If you have, or expect, a good honours degree (1st or upper 2nd) in electronic engineering, computer science, physics, or related discipline, and would like more information, please get in touch with: Dr Alastair R Allen Dept of Engineering University of Aberdeen Aberdeen AB24 3UE Tel: 01224 272501 Fax: 01224 272497 Email: a.allen@abdn.ac.ukArticle: 6402
No messageArticle: 6403
Does it support XC3010-XC3090, for which Lucent/AT&T were a 2nd source? If it does, then it is good value. Peter. Return address is invalid to help stop junk mail. E-mail replies to z80@digiserve.com.Article: 6404
Hi, The recent posts on gate counting has been very interesting (and timely). I just received a proposal to review and would like some opinions on the claims that are made. The device of interest is the Atmel AT6010. Here's the information provided that I need to assess (actually, I have investigated this but would like some additional opinions on this). . Larger than Act 1, comparable to Act 2. - equivalent to 20,000 gates + 6400 flip flops + RAM buffers . Combination of Random Logic with SRAM blocks - RAM buffers allow implementation of FIFO's and data buffers without adding chips. That's it! Feel free to confirm or to flame [I didn't write this I don't care ;)]. But, I would just like an honest assessment and evaluation. And for anyone who has designed with this chip, actual gate counts vs. 'theoretical' gate counts would be interesting, as in the ability to change the internal design without changing either any pin outs or any critical path timing. For anyone who either works for or against Atmel, please identify your affiliation or e-mail your response to me. If you wish to publicly post and identify your affiliation via e-mail, that is fine, and I will hold your affiliation in confidence. Thanks for any information youse can provide, rk rich.katz@gsfc.nasa.gov (301) 286-9705 kArticle: 6405
In article <3380c5f6.14644618@nntp.netcruiser>, Stuart Clubb <s_clubb@netcomuk.co.uk> wrote: I'm doing a Xilinx PCI design, so this is all very interesting. >Xilinx's PCI solution: >I couldn't find any publish VI characteristics data for Xilinx. Is it >available? Does it meet the PCI spec? They're on pages 14-9 for the XC4000E series and 14-17 for the XC3000 series in the '96 databook. Both are VI-curve complient, but I suspect that the XC3000 series might not have enough GND pins. >Document pcids.pdf page 4 indicates that in burst mode when supplying >data, the Initiator and targets both supply data to the PCI bus at 50% >data rate as a wait state insertion is required on every cycle. Does anyone have any idea why this would be (for a target side PCI using Xilinx)? A target only PCI interface with pipelining should be able to do this. Also I have a question about the PCI CLK line... on PCI motherboards, can this line be bussed or must it be driven point-to-point from a clock buffer to each individual board? It seems to me that if it is bussed, PCI can not guarentee that ringing will not occur at the threshold voltage (with its reflected wave switching).Article: 6406
A few more criterias (that Janick forgot?) are: * Performance of tools, such as: Speed Memory requirements Turn around time * Quality(Robustness) * Sign-off janick@qualis.qualis.com (Janick Bergeron) wrote: >It's a business decision. Look at every factor EXCEPT language >features. Their value can only be evaluated subjectively and is the >source of much religious conflicts. And the feature that makes a >language better for your next project may not be that important for >the next one. > >Here are a few things to look at, in no particular order >(except the order in which they came to mind): > >- Cost of tools >- Relationship with tool vendor >- Integration of HDL tool with existing tool set >- Knowledge and experience of engineers >- Engineer's subjective preference >- Support from target and potential FPGA/ASIC vendors >- Quality of models from vendor >- Relationship with FPGA/ASIC vendor >- Availability of 3rd party models for board/system simulation >- Availability of IP blocks >- Ease of use >- Sign-off kit requirements >- Customer requirements (if they purchase the final design) > > >Better yet: edge your bet by picking bilingual tools! > >If you pick Verilog, I personally think Verilint is a must have. > >-- >Janick Bergeron Qualis Design Corporation Ph.: (503) 350-3663 >Director of PO Box 4444 Fax: (503) 643-1583 >Technology Beaverton, OR, USA, 97075-4444 janick@qualis.com > VHDL - Verilog - Synthesis - Modelling - Verification - TrainingArticle: 6407
Dear folks, during the timing simulation with vhdlsim of a system composed of 3 Xilinx FPGA's I observe spurious glitches on an output signal with a duration of the order of 10 ns, while the clock has a 80 ns period. I'm wondering why I get this, as synopsys is supposed to generate glitch free logic, right ? Are maybe the Xilinx cells poorly modeled ? The three FPGA's are instantiated in a top level test bench entity, and the signals generated by the test bench are carefully timed in order to avoid hold/setup violations on the input flip-flops. Any idea? thanks in advance -Arrigo -- Arrigo Benedetti e-mail: arrigo@vision.caltech.edu Caltech, MS 136-93 phone: (818) 395-3695 Pasadena, CA 91125 fax: (818) 795-8649Article: 6408
snip.... > > - LogiBLOX components are compiled when they are created, so there is an > underlying EDIF, VHDL or Verilog model for each component that is > built. There is no need for programs like XSIMMAKE with M1; they are > ready to simulate when they are built. > > A conversion guide is shipped with the M1 software to assist users > upgrading exsiting XBLOX designs to use LogiBLOX. > > Watch the Xilinx Web Site (http://www.xilinx.com) for more information > in the near future. > > -- > David Dye > Software Applications Engineer > Xilinx Inc., San Jose, CA Sorry if I blow my own trumpet, but...... Sounds a lot like Atmel's FPGAs macro generators which have been shipping for about 2 years and create simulation models/schematics/symbols/hard layouts/deterministic timing all from one simple UI with about 50 different parameterisable macros availble. Now we have linked all this good stuff in with synthesis too (IDS4.0); for automatic macro instanciation from VHDL or Verilog based synthesis tools. If you would like more information on these tools e-mail fpga@atmel.com. Martin Mason. FPGA Apps. Atmel Corp.Article: 6409
Matthias Sauer wrote: > > >>>>> "Brad" == Brad Taylor <blt@emf.net> writes: > > Brad> I thought the readers of this newsgroup might want to see the > Brad> following article from the June97 issue of Scientific American. > Brad> It's kind of nice to see FPGAs (and associated researchers) getting > Brad> some recognition in the mainstream press. My guess is that this > Brad> won't be the last article. Notice that they call it "configurable > Brad> computing". > > That's interesting, indeed. For those who might be interested in > "Re"-configurable computing (the next generation, maybe?) I suggest to have a > look at > > http://www.comlab.ox.ac.uk/oucl/hwcomp.html > > (To pick just one random URL :) > > Brad> http://www.sciam.com/0697issue/0697villasenor.html > > Brad> I'm interested in any comments about the article. - Brad > > I haven't read it thoroughly yet, but it seems that the authors completely > forget about what is going on in Europe (see the above URL) > > Cheers, > > Matthias > -- > Matthias Sauer Tel +44-1865-283549 > Oxford University Computing Laboratory Fax +44-1865-273839 > Wolfson Bldg., Parks Road email masa@comlab.ox.ac.uk > Oxford OX1 3QD, U.K. > URL: http://www.comlab.ox.ac.uk/oucl/users/matthias.sauer/ I also didn't see any mention of PAM or the fine work in Toronto or Splash or any commerical work. I believe that all the people mentioned have DARPA contracts. When I read in the artical "the first demonstrations did not occur untill a few years ago" I have to question what "a few" means. I think the PAM project did quite a bit work in 89 - 92 which is over 5 years ago. And algotronix had a board out in 91 I think. Just picking some nits there. Anyway its great to see some exposure. In fact we advertized in the issue. Checkout page 133 for our ad. To see what the ad looks like visit http://www.marshall.com Steve CasselmanArticle: 6410
Has anyone any ideas on how to do a 8-bit (say) numeric inversion inside a FPGA? Ie 4-bit lookup plus 4-bit inperpolation. TIA, Keith.Article: 6411
Brad Taylor wrote: > > I thought the readers of this newsgroup might want to see the following > article from the June97 issue of Scientific American. > It's kind of nice to see FPGAs (and associated researchers) getting some > recognition in the mainstream press. My guess is that this won't be the > last article. Notice that they call it "configurable computing". > > http://www.sciam.com/0697issue/0697villasenor.html > > I'm interested in any comments about the article. > - > Brad For reconfigurable computing under a machine paradigm you may like to LOOK AT OUR WEB PAGES: http://xputers.informatik.uni-kl.de ReinerArticle: 6412
Philip Freidin <fliptron@netcom.com> wrote in article <fliptronEAJ09r.Ipp@netcom.com>... > > I've seen some pretty stupid B*L@ S#I& in my reading of posts in various > news groups, but this one belongs in alt.conspiracy.not.thinking.clearly uh, perhaps a bit strong. > > Kevintsmith (Of Quicklogic) responds to jwbrooks (Of Xilinx) > <snip> > > > There is > >sufficient routing wire and interconnect to easily hook up the logic > >cells and I/Os in a device that has every logic cell used and every I/O > >pre-placed. The low resistance and capacitance of this type of > >interconnect give 1,000 times less impedance (RC = 30ohms for QuickLogic > >vs. RC=40kohms for Xilinx's SRAM interconnect). > > Normally I would have just left this post go by, but the above is what > really set me off. ( As those who have seen me post before, you are > probably aware that I am not a highly opinionated person, and am only too > happy to let bygones be bygones ) well, Quicklogic compares the resistance of different switches in a White Paper on their WWW site. Here they state: Programming Technology ViaLink Antifuse Dielectric Antifuse SRAM Interconnect Silicon Area Required 1.0X 2.5X 12.5X Typical ON Resistance 30ohms 200-500 ohms 600-800 ohms Typical OFF Capacitance 1 femtofarad 5 femtofarads 50 femtofarads Available Routing per Gate 1.0X 0.8X 0.24X Table 3. Size and RC Characteristics of Various Programmable Technologies End direct pull from Q-Logic WWW site. The 40 kohms sounds kind of high and it is; otherwise, parts just wouldn't run at all through the routing network and they would have to be pretty bad transistors. The 600-800 ohms sounds more reasonable; perhaps some of the sram vendors can comment on this; phil mentioned values in the 100-200 ohm range. One interesting parameter not included on this chart, which I feel should be for a good trade-off, is leakage current @ maximum temperature and voltage. The antifuse based products have tons of antifuses, on the order of 1 million for a 10,000 gate device, which gives them the routing power. On the other hand, most of the antifuses are not programmed; that is the key point made here I believe by Q-Logic, with their low OFF capacitance, to distinguish their Vialink from the Actel PLICE antifuse and contributes to greater speed. On the other hand, increasing the number of antifuses will increase the leakage current for the amorphous silicon antifuses, which may be a limiting factor for very large devices (esp. at MIL temps). Perhaps the Q-Logic people can discuss this. <snip> > > The pass transistor of the SRAM FPGA is 40Kohms (according to Kevin), and > is therefore 1000 times worse. If this were true, then it would be > reasonable to expect the SRAM devices to be 1000 times slower. Last time > I looked, muxes were often built out of pass transistors, just like the > the muxes that make up the logic section of the QL devices. Are these also > 40K ohms? i think there's some good points here, as i commented on above. but i don't think that the speed of the device is wholly dependent on the speed of the switch. that is why the architecture of say the xilinx and orca's differ from say the actel and quicklogic ones. the goal of the architectures with the slower switches is to minimize the impact of their switch technology, which does take up more room (routability) and does have a higher resistance (speed). i did some interesting comparisons with one of the orca guys, comparing, if my memory is correct, a matching circuit (like 12 bits or so) between the orca and 3200dx family; i believe that the orca's were a bit faster. in any event, they were both pretty good and roughly the same. of course, it seems that the sram-based devices, because they use standard cmos structures (no antifuses, high voltage transistors, etc.) have the ability to move to new, smaller processes quicker, giving them a technology edge over their antifuse counterparts. yes, muxes are often built out of pass transistors but the resistance of the mux transistor (do these devices use a pass transistor or a transmission gate like the old cd40xxb stuff?) is less critical than the resistance of the routing switch since obviously in a logic cell the capacitance would be far less than a large routing track with a bunch of loads hanging on it. comments? rkArticle: 6413
Hi! Have anyone tried a timing simulation in Mentors QVPRO environment? I'm trying to do a mixed simulation with a synthesized XC3000 model together with a VHDL testbench. Using backannotated timing information doesn't seem to work. I can't see any delays at all. I'm using Mentor version A3F and Xact version 5.2.1. John john@gaia.physto.seArticle: 6414
Richard Nuth <nuth@col.hp.com> wrote: >I have Simucad's SILOS III installed on my PC at home and find it to be >quite suitable for Verilog simulations. It runs fine under NT 3.51 and >NT 4.0. The simulation environment is better integrated than many I >have seen, and you can also run "batch" mode simulations with it. Are there simulators out there that do NOT let you run a batch mode simulation? If so, I'd like to know which they are so I can avoid them. SteveArticle: 6415
> >Document pcids.pdf page 4 indicates that in burst mode when supplying > >data, the Initiator and targets both supply data to the PCI bus at 50% > >data rate as a wait state insertion is required on every cycle. > > Does anyone have any idea why this would be (for a target side PCI using > Xilinx)? A target only PCI interface with pipelining should be able to do > this. Target does not have this problem...or at least shouldn't in a 'well designed' PCI interface. I believe the problem is with master only. It has to do with the input timing of TRDY. If TRDY gets pulled, the Logicore design cannot respond in time to stop the transaction. The design I 'use' does not have this problem for target or master... ;-) Austin Franklin darkroom@ix.netcom.comArticle: 6416
Candidates interested in pursuing a post-doctoral fellowship in the areas of digital system design and design automation at Princeton University should contact me via email. I am attaching a notice of the program that this is part of. The teaching component of this program makes this an attractive opportunity for candidates intending a subsequent academic career. PRINCETON UNIVERSITY POSTDOCTORAL TEACHING FELLOWS PROGRAM Princeton University is seeking applicants for its Postdoctoral Teaching Fellows Program. The program is designed to provide up to three years of stipend support to postdoctoral fellows who wish to obtain further training in both research and teaching. The Fellows will spend the majority of their time conducting research in the laboratory of a Princeton University science or engineering faculty member. Fellows will also gain experience in curriculum development, instructional laboratory design and lecturing by working closely with an experienced Master Teacher. The goal of the Program is to prepare candidates for careers in both research and teaching. Eligible candidates should have a PhD or equivalent post-baccalaureate degree in one of the physical and natural sciences, or engineering. Before applying, applicants must first identify a research mentor from among the Princeton University science and engineering faculty, and obtain his or her agreement to act as research sponsor. A faculty member, who may or may not be the research mentor, will be identified for successful candidates to coordinate the teaching and research activities of the Fellow. The stipend is $35,000 per annum. Applications are currently being accepted for Fellows to begin at any time during the 1997-1998 academic year. Review of applications will begin June 1, 1997. Applications can be obtained by writing to: The Council on Science and Technology 123 Lewis Thomas Laboratory Princeton University, Princeton, NJ 08544 tel: 609-258-4316 FAX: 609-258-3345 email: cprevost@pucc.princeton.edu -- Sharad Malik sharad@ee.princeton.edu Associate Professor 609-258-4625 Dept. of Electrical Engineering 609-258-3745 Fax Princeton University http://www.ee.princeton.edu/~sharadArticle: 6417
Stuart Clubb wrote: > Lucent Technologies have announced a special "Summer deal pricing" (at > least in Europe). Is there an ACADEMIC pricing of this available in the U.S., for Win95 or Linux platform? I'd be interested. Thanks in advance. Vitit Kantabutra vkantabu@howland.isu.eduArticle: 6418
Mark, I believe you need to use the "exemplar.lmf" file located in the Exemplar tree ($EXEMPLAR/data/exemplar.lmf). Specify this file in the EDIF netlister setup in Maxplus2. Altera is aware that the exemplar.lmf file provided in the Maxplus2 version 7.x is obsolete, so for now specify the one shipped by Exemplar. Troy Garrett Mark Sandstrom wrote: > > Hi! > > I'm synthesizing by Leonardo v4.03 to Altera FLEX10K CPLDs. > If I load and resolve the FLEX10K modgens, the edif produced > by the write altera -command contains cell referencies to > "BUF":s. > > The Altera's place and route tool Maxplus2 (v7.2) doesn't > accept this edif and issues an error message: > "Error: Can't find design file 'buf'". > > I've been told that FLEX10K libraries that Maxplus2 uses > don't contain an element called "BUF". > > There is no problem with cell referencies to "AND2", "OR2" etc. > > If I don't load and resolve the modgens in Leonardo, > the Maxplus2 accepts the Leonardo's edif output, but the > synthesis result is not good enough. > > Is there any known solution to this problem? > > I'd appreciate any information concerning this or other possible > problems associated with Leonardo -> Altera synthesis flow. > > Best regards, > > Mark H. SandstromArticle: 6419
Hi, I was wondering if there are a class of circuits/systems that I should not implement using FPGAs. I am told that floating point units dont perform well when implemented using FPGAs. In fact a multiplier that I had implemented required 200ns (16 bit). Present day CPUs have much better FPUs/ALUs. There was some discussion abt implementing comparator using FPGAs, and some suggested use CPLDs for combinational logic. Are there any other thumb rules as to when to go for an FPGA and when not to. This becomes important when you are considering co-design and you want to decide if a function is to be executed in h/w or s/w. I have seen systems where this is specified as a part of the input code. Thanks a lot. Phani ------------------------------------------------------------------------ Phani K Putrevu 384 Probasco Street #14 MS - Comp Engg Cincinnati OH 45220 University of Cincinnati Ph : 513-281 1154 (res) email: pputrevu@ececs.uc.edu 513-556-0904 (lab) URL : http://www.ececs.uc.edu/~pputrevu ------------------------------------------------------------------------Article: 6420
Yes this is pretty much the same package. We could not quite call the package Trianus for marketing reasons:) We are fixing the interface to Lola(Trianus) so it will have the look and feel of windows on the pc or mac. You know there are certain times in history when an effort by a small group can make a big difference. What is need is a language that goes from operating systems and high level languages right down to programming hardware. There are currently attempts to have java and C++ be these languages but they are frozen by standards and red tape. Today as we speak Lola is the only language that is designed just for programming FPGAs. VHDL and verilog where designed for different purposes (none include placement for example). Also Lola compiles very fast because of the common data structure. A language that is designed to program both configurable hardware and normal CPUs will help change the world. Lola is not quite the language but it is a damm go start. Steve Casselman, President Virtual Computer Corporation See our distributors page http://www.marshall.com mcintosh@vima.austin.tx.us wrote: > > Recent articles in Comp.Lang.Oberon have asked variously > what Prof. Dr. Wirth is doing, why no high visibility Oberon projects > exist, and whether advertising would help the Oberon community. > > The June '97 Scientific American has a Xilinx FPGA on the cover and > an article about Configurable Computing on p. 66. In addition, in a > full page ad on p. 133, the Xilinx XC6200 development package is > mentioned, along with the Lola language and Niklaus Wirth. > > A prof. John Koza, Stanford, is quoted as a user. > > The Web index on p. 61 does not mention any link to any apparant > connection to the product, an unfortunate omission. > > The WWW page given in the ad does not lead to a page describing > the product. However, some searching on altavista gives > http://www.vcc.com/ as a url for the Virtual Computer Corporation. > > It is tempting to assume that the Trianus package announced on > approximately Dec 23, '96 is identically the software being distributed > with the H.O.T. Works product. Some reference is made to a Xilinx > reference design, which may imply that Xilinx or another party does > or could second source the board itself.Article: 6421
testArticle: 6422
There is a Cypress product, on CD, entitled "Cypress WARP Version 4.0". The current Cypress databook does not to describe a product by this exact name; the products listed there are "Warp2", "Warp2+", "Warp3", and "Warp3 Pro Series Built-in". Does "WARP Version 4.0" correspond to any of these? Thanks. SteveArticle: 6423
In <q0raf0gjxx.fsf@vision.caltech.edu> Arrigo Benedetti <arrigo@vision.caltech.edu> writes: > >Dear folks, > >during the timing simulation with vhdlsim of a system composed of 3 >Xilinx FPGA's I observe spurious glitches on an output signal with >a duration of the order of 10 ns, while the clock has a 80 ns period. >I'm wondering why I get this, as synopsys is supposed to generate >glitch free logic, right ? >Are maybe the Xilinx cells poorly modeled ? The three FPGA's are >instantiated in a top level test bench entity, and the signals generated >by the test bench are carefully timed in order to avoid hold/setup >violations on the input flip-flops. >Any idea? > >thanks in advance > >-Arrigo >-- >Arrigo Benedetti e-mail: arrigo@vision.caltech.edu >Caltech, MS 136-93 phone: (818) 395-3695 >Pasadena, CA 91125 fax: (818) 795-8649 Maybe it's difficult to say without look at your design, but I would say to register your outputs may fix the glithes. Alfred Chang aca@ix.netcom.comArticle: 6424
Daniel Roganti (ragman@euronet.nl) wrote: : Anyone with some advice about using leonardo? : Any pros/cons regarding this development software ? : thanks : : +_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_ : Name: Daniel Roganti : Email work : daniel@betronic.nl : Email home : ragman@euronet.nl : Smail: Oud Wulvenlaan 35-2, 3523XS Utrecht, Netherlands,Europe : WWW: http://www.euronet.nl/users/ragman : Back home: Margate, Florida Hometown : Elmont, New York : ...exploring cyberspace before it runs out of space... : +_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_ : We have recently bought Exemplars Leonardo 4.0.3 for Linux. We are doing FPGA synthesis (mainly Xilinx) with it. We considered: Synplify from synplicity FGGA Express from Synopsis Galileo and Leonardo from Exemplar Synplify and FGPA express are Windows applications in its ugly sense. Solely PushButton Based. Solely Interactive. Unable to be integrated into more complex Design Flows. No Command-Line at all. But they are fast. (Till now we have no speed problems, even with leonardo, even with a 133 Pentium) Synplify has the advantage that you get both VHDL and Verilog together. For Exemplar and FPGA Express you have to decide for VHDL or Verilog. Or you have to pay twice price. For Synplify,FPGA Express and Galileo you have to decide which target Devices you want. The more Devices the more price. For Leonardo you always get every Device. So you have the possibility to consider another FPGA manufacturer without to much initial overhead. Leonardo cant be compared with Synplify Express and Galileo. It is a optimisation system which gives you access to its functionality via 1) a synthesis tool which is controlled by a built in tcl interpreter. The so called program "elsyn". 2) a tcl/tk interactive GUI interface. The so called program "leonardo". this GUI starts a graphical tk session, and feeds commands to a parallel running "elsyn" This is very effective, since you can learn interactively from the GUI what commands have to be fed to elsyn to do a specific job. Once you have learned what has to be done, you can write it together to a tcl script and send it to elsyn directly. The functionality which elsyn exposes to you gives you very detailed control of the synthesis. Contrary for Synplify, FPGA-express and galileo the job is pretty fixed. leonardo is the most expensive of these tools (aprox 16000$). But in my opinion its capabilities go far beyond what you can do with the other tools. Due to Price, and due to our inexperience concerning HDL tools our decision around Jan 97 was to go for synplicity. At the very last moment Exemplar announced Linux support for Leonardo, and I investigated a demo installation of Leonardo. I was such impressed that we have bought it around April. Note: The leonardo Price for Linux is the same as for Windows. The Price for Sun and HPUX is aprox twice as high. So with Linux you can use this tool on a full fledged Unix (on a cheap intel based PC) for the same price as the Windows version. Larice
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