Custom Search
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
Paul, if you want the performance and asynchronous behavior that you mention, you must make a decision on a specific architecture. At the 50+ MHz performance level, you need a true dual-port RAM, you cannot fake it with multiplexed addresses. I wrote an app note on this subject, it's XAPP051, appropriately called "Synchronous and Asynchronous FIFO Designs". If you intend to use it, send me e-mail and I'll tell you a few new tricks. ( I was the driving force behind the Fairchild 3341, the industy's first FIFO in 1972, a quarter century ago ). In an asynchronous FIFO, there is no known phase or frequency relationship between the read and the write clock. The difficulty lies in the reliable generation of the control signals FULL and EMPTY ( the rest is trivial ). Slow designs you can do in a generic fashion, but for really high performance, you need to decide on a particular manufacturer's part and development system. Luckily the decision often narrows down to only a few contenders, sometimes only one. Near the end of the alphabet. :-) Peter Alfke, Xilinx ApplicationsArticle: 7351
Have anyone more than I had any problems with the Microchip 24LC164 (I2C EEPROM 2k*8)? As I have detected, it responds not only to the I2C address that I have choosen (1001xxxr) but also to 0110xxxr. I have tested the same type of device from Thomson, Xicor and Atmel and it is only the Microchip device that acts this way. I also changed the device address (by changing the A1 pin) to 1011xxxr but its still responds to address 0110xxxr too. I have had som contact with Microchip, but they neither confirms or denies. Is this a bug in the 24LC164? Normally it is of no problem, but if I want to use a device that actually uses the address 0110xxxr I would get a conflict on the address using the 24LC164 device. Has anyone seen the same behaviour?? /Anders SandgrenArticle: 7352
Joseph, > So which mode should we use delay or nodelay? I would guess that a card on > a bus should use nodelay (short setup time, more than 0ns hold time), but a > chip on a motherboard should use delay (large setup time, no hold time). According to the data book, DELAY is the mode you want to use. It supposedly requires 7ns setup and 0ns hold. > Why would XACT give more conservative numbers for this? What additional > assumptions do the "massaged" numbers make- there should be no variable > routing estimates for global clock to IOB FFs? Is a narrower temperature > window perhaps being assumed? I don't know what the exact (no pun intended) numbers are, but they e-mailed me a PERL script called 'model_io' that I had to run on the simulation .xnf file. This script changed the INFF setup and hold values to be 7setup and 0 hold so I could pass simulation.... What the claim was is as follows: "Run the model_io Perl script, (model_io version 2.01 or higher should be used). This corrects known conservative modeling values for the setup/ hold and clock-to-output timing of the I/O flip-flops. The Perl script provides the guaranteed data book values in the final .xnf file (see the XC4000E Technical Data data sheet). The Perl program version 5.00 or higher is required to execute the model_io script," So, no, it is not temperature related, as it shouldn't be. My understanding is that the numbers used in the speed files is 'conservative'....and does not match the data book values. Note the data book 'guaranteed' values require the use of a primary global clock, and it obviously must be direct. Alas, though I have this nice script, all I know now is my design passes a 7ns setup and 0ns hold time simulation....but still doesn't work in the system with the one bridge - 21152-AA. Argh! AustinArticle: 7353
I'm tempted to load my PCI controller FPGA configuration (Xilinx bit-stream)
over the PCI bus. One way is to broadcast the bitstream to all slots- a
simple descrete circuit looks for writes to read-only configuration space
registers and latches some of the address bits. The latched bits become the
cclk and din lines for the configuration data. Now this circuit never
drives the devsel# (or any other) line, so the write will eventually just
terminate after 6 cycles. This is so slow that the circuit can be seperated
from the bus with, say, 100 ohm resistors- so it won't load the bus and
the signals can also go to the FPGA without violating the spec.
This happens way after the PCI bios does memory mapping, so you have to do
the memory mapping on your own. (Actually the PCI bios should provide a
memory remapping call for just this purpose).
Now we just need a standard so that several people can have their FPGA
boards in the bus: maybe you have to broadcast the vendor and device IDs
before the bit-stream.
--
/* jhallen@world.std.com (192.74.137.5) */ /* Joseph H. Allen */
int a[1817];main(z,p,q,r){for(p=80;q+p-80;p-=2*a[p])for(z=9;z--;)q=3&(r=time(0)
+r*57)/7,q=q?q-1?q-2?1-p%79?-1:0:p%79-77?1:0:p<1659?79:0:p>158?-79:0,q?!a[p+q*2
]?a[p+=a[p+=q]=q]=q:0:0;for(;q++-1817;)printf(q%79?"%c":"%c\n"," #"[!a[q-1]]);}
Article: 7354
It would be really nice if PCI controller FPGAs could be configured over the
PCI bus. One way of doing it is to use configuration writes with the fpga
data encoded in the address. A simple circuit looks for these configuration
writes and latches the address. The circuit doesn't drive idsel# (or any
other line), but timeout occurs anyway after 6 cycles- the address is
guarenteed to get through, however. The circuit could be added onto the PCI
lines which already go to the FPGA through large resistors, so as not to
violate the spec. To configure the FPGA, you write the bitstream to each
slot which appears empty in succession until your card appears.
The main problem with this scheme is that your card won't appear until after
the PCI bios has assigned addresses. Thus you have to do the memory mapping
on your own (there should be a call for this in the PCI BIOS...).
Another problem is that a standard needs to be made to prevent you from
configuring other people's PCI cards: Two of the address bits are needed for
this: one bit enables shifting in of the other bit into a vendor ID shift
register. The configuration data is only accepted when the correct ID has
been loaded into the register. The shift register could be optionally
longer (the required vendor ID must be shifted in last for this to work), so
that the vendors can differentiate between cards using their own scheme,
without requiring a 40 bit register.
I want this circuit to be as simple as possible, so that it can be done
with descrete logic and you can make cards without any programmed devices.
So what do you think? Should we make a standard? I'm going to try this
circuit on a card I'm making (and leave space for the Xilinx prom in case
there's a problem :-). I know I can get it to work on a PC- I'm not sure
how easy it is to access configuration space in other systems though.
--
/* jhallen@world.std.com (192.74.137.5) */ /* Joseph H. Allen */
int a[1817];main(z,p,q,r){for(p=80;q+p-80;p-=2*a[p])for(z=9;z--;)q=3&(r=time(0)
+r*57)/7,q=q?q-1?q-2?1-p%79?-1:0:p%79-77?1:0:p<1659?79:0:p>158?-79:0,q?!a[p+q*2
]?a[p+=a[p+=q]=q]=q:0:0;for(;q++-1817;)printf(q%79?"%c":"%c\n"," #"[!a[q-1]]);}
Article: 7355In a previous article jhallen@world.std.com (Joseph H Allen) writes: : ;It would be really nice if PCI controller FPGAs could be configured over the :PCI bus. One way of doing it is to use configuration writes with the fpga ;data encoded in the address. Why not just do it through JTAG.Article: 7356
I'm currently using a serial PROM for FPGA loading . This device is programmed with a file in the MCS (intel86) format. As we're going to transfer to S/W dowload of this file, we need to know the exact format of this file in order to extract the respective data, to be tranformed into a S/W driver. Does anyone know this format , or where it could be found ?Article: 7357
In article <5u7v21$5d8@neptune.myri.com>,
Wen-King Su <wen-king@myri.com> wrote:
>In a previous article jhallen@world.std.com (Joseph H Allen) writes:
>:
>;It would be really nice if PCI controller FPGAs could be configured over the
>:PCI bus. One way of doing it is to use configuration writes with the fpga
>;data encoded in the address.
>Why not just do it through JTAG.
I don't know anything about JTAG or JTAG on the PC. Is it board by board
addressable? I'll have to learn more about it.
--
/* jhallen@world.std.com (192.74.137.5) */ /* Joseph H. Allen */
int a[1817];main(z,p,q,r){for(p=80;q+p-80;p-=2*a[p])for(z=9;z--;)q=3&(r=time(0)
+r*57)/7,q=q?q-1?q-2?1-p%79?-1:0:p%79-77?1:0:p<1659?79:0:p>158?-79:0,q?!a[p+q*2
]?a[p+=a[p+=q]=q]=q:0:0;for(;q++-1817;)printf(q%79?"%c":"%c\n"," #"[!a[q-1]]);}
Article: 7358> The circuit could be added onto the PCI > lines which already go to the FPGA through large resistors, so as not to > violate the spec. You will have more than one load in that particular slot, which is a violation of the spec. Just routing the signal through a resistor does not change the number of actual loads. There is also a maximum trace length spec, and a maximum capacitance spec. Also, routing the signal through a resistor changes the rise and fall times of that signal...and may cause setup and/or hold time violations.... Austin Franklin darkroom@ix.netcom.comArticle: 7359
AS OF 8/31/97, THERE ARE 436 COMPANIES ON THIS LIST!! -- IMPORTANT NEWS FOR USERS OF GRAY'S SEMICONDUCTOR URL PAGES the only URLs that now work for accessing my site are as follows: http://www.scruznet.com/~gcreager http://www.scruznet.com/~gcreager/brief.htm (non-frame brief listing) http://www.scruznet.com/~gcreager/verbose.htm (non-frame verbose listing) -- Now, just a reminder to everyone... I have been compiling a semiconductor manufacturer website listing (and making it available to other engineers on the web) for over 2 years now. I used to post it to sci.electronics every so often in the past, but it has gotten way too big for that. There are 436 companies on the list currently. These are chipmakers who market chips under their own names, so if you can't find the data sheet that you need because you don't know the manufacturer's website, try my listing. Bookmark it and tell your co-workers!! I try to keep it the most up-to-date and useful listing of chipmakers anywhere on the web (by constantly searching for new URLs before most others find them) and it is now widely used by a lot of people. The following categories exist among my "engineering" subpages. - Semiconductor URLs (Brief listing) - Semiconductor URLs (Verbose listing; includes product categories) - New Semiconductor URLs - Missing Semiconductor URLs - Motivations - 25 Best Semiconductor Websites - 25 Worst Semiconductor Websites - Engineering Humor try it out, I think you'll find it to be quite useful. Let me know what you think. Again, the URL is: http://www.scruznet.com/~gcreager The website is best viewed in a frames environment, but there are "back doors" for getting to these pages if your browser doesn't support frames. Thanks for all the support everyone has given me in this endeavor! :-) Gray Creager -- ******************** PLEASE NOTE ********************* To respond to me, remove "no.spam.to." from the e-mail address reported above. Sorry for this inconvenience. ****************************************************** Gray Creager http://www.scruznet.com/~gcreager ******************************************************Article: 7360
In article <01bcb632$b7b9d300$35a8b8cd@drt1>,
Austin Franklin <dark4room@ix.netcom.com> wrote:
>> The circuit could be added onto the PCI lines which already go to the
>> FPGA through large resistors, so as not to violate the spec.
>Just routing the signal through a resistor does not change the number of
>actual loads. There is also a maximum trace length spec, and a maximum
>capacitance spec.
The bus has a certain impedance: say 25 ohms (typical for a loaded bus). A
load seperated from the bus through a big resistor R (even if that load is a
transmission line) is going to have its effect reduced by 25/R. Suppose you
can model the line on the circuit side of the resistor as a distributed
capacitance of 15pF. If R is 100 ohms, the bus will see it as 3.75 pF.
Of course, you want the resistors to be as close as possible to the short
connection between the bus and the PCI controller (FPGA).
>Also, routing the signal through a resistor changes the rise and fall times
>of that signal...and may cause setup and/or hold time violations....
This is a more difficult problem. You want the resistors to be as small as
you can get away with, all of the resistors to be equal in value (especially
the one on the clock line), and the lines between the resistors and the
configuration circuit to be as equal in length (and in every other
characteristic) as possible. With a 100ohm resistor and 15pF load, the
risetime is going to be no faster than 5ns. Since the window is only 7ns,
the input capacitances have to be as matched as possible, since differences
will effect skew to a greater degree than usual.
You can actually pull off things like this. The best one I've seen so far
is a cheap Cirrus Logic PCI VGA card. It cost only $27! How many PCI cards
have you seen for only $27? They get it that cheap because it's only a two
layer board (with only four chips total). A two layer board for PCI?
Impossible and out of spec. you say! Well actually it is possible: the
reason the PCI bus wants you to have a ground plain on your cards is to
reduce inductive coupling. The return current part of an AC signal is the
lowest impedance ground path from the destination of the signal to the bus
(for DC it's the lowest resistance ground path). This incidentally is why
you are supposed to decouple the 3.3V lines even if you don't need 3.3V:
they provide more AC grounds. When you have a ground plain, this path is
usually the part of the ground plain right under the signal trace. Without
a ground plain, it's the loop made by the nearest ground trace and the
signal trace. Unfortunately these loops are physically big and loops from
other signals which share the common ground trace are right on top of each
other, so there is crosstalk in the form of inductive coupling (which is
additive if all the signal lines transition at the same time). So how do
they get away with only two-layers on the VGA card? Well the strength of
the inductive coupling also depends on the magnitude of the current through
the loop. So... you guessed it... they put 50ohm series resistors on most
of the signal lines (they did the 32 data bus lines).
I also have a slightly older Cirrus Logic PCI VGA card: it also is only 4
chips, but it's a four layer board. It was more than double the price of
this one (around $65 I think).
--
/* jhallen@world.std.com (192.74.137.5) */ /* Joseph H. Allen */
int a[1817];main(z,p,q,r){for(p=80;q+p-80;p-=2*a[p])for(z=9;z--;)q=3&(r=time(0)
+r*57)/7,q=q?q-1?q-2?1-p%79?-1:0:p%79-77?1:0:p<1659?79:0:p>158?-79:0,q?!a[p+q*2
]?a[p+=a[p+=q]=q]=q:0:0;for(;q++-1817;)printf(q%79?"%c":"%c\n"," #"[!a[q-1]]);}
Article: 7361
Ilan Ron <iron@radnet.co.il> wrote in article
<3408F0C3.5503@radnet.co.il>...
| I'm currently using a serial PROM for FPGA loading .
| This device is programmed with a file in the MCS (intel86) format.
| As we're going to transfer to S/W dowload of this file, we need to know
| the exact format of this file in order to extract the respective data,
| to be tranformed into a S/W driver.
|
| Does anyone know this format , or where it could be found ?
|
There is a utility available on the Xilinx web site that converts an Intel
MCS file into the file format that you specify. For example, you can
create data-byte statements for assembly code, C data structures, etc.
There is a UNIX version ('makesrc.tar.Z') and a PC version ('makesrc.zip').
The download also includes some simple documentation.
Also, for reference, you may want to look at the C source code provided on
their web site ('dn_ld.zip'). However, this is for downloading a Xilinx
bit file instead of an MCS file but the idea is the same.
--
Steven Knapp
OptiMagic(tm) Logic Design Solutions
E-mail: sknapp @ optimagic.com
Programmable Logic Jump Station: http://www.optimagic.com
Article: 7362
Sorry, I had drain bamage on the last posting. I forgot to include the
full path names. Here we go again:
There is a utility available on the Xilinx web site that converts an Intel
MCS file into the file format that you specify. For example, you can
create data-byte statements for assembly code, C data structures, etc.
There is a UNIX version
('ftp://ftp.xilinx.com/pub/utilities/prom/makesrc.tar.Z') and a PC version
('ftp://ftp.xilinx.com/pub/utilities/prom/makesrc.zip'). The download also
includes some simple documentation.
Also, for reference, you may want to look at the C source code provided on
their web site ('ftp://ftp.xilinx.com/pub/utilities/prom/dn_ld.zip').
However, this is for downloading a Xilinx bit file instead of an MCS file
but the idea is the same.
--
Steven Knapp
OptiMagic(tm) Logic Design Solutions
E-mail: sknapp @ optimagic.com
Programmable Logic Jump Station: http://www.optimagic.com
Article: 7363CELEBRATE A YEAR OF ACHIEVEMENT 1997 VSI FALL WORLDWIDE MEMBER MEETING October 3, 1997 (after the Embedded Systems Conference) Santa Clara, Calif. Free for members $150 for non-members JOIN US -- to see the progress of the working groups and begin to reap the benefits for system-level ICs. This meeting will begin the discussions of VSI specifications with members and with the industry. Keynote Speaker: Anant Agrawal Vice president and general manager of the performance products group, Sun Microsystems Microelectronics Division. The program runs from 9:00 until 4:00pm when parallel sessions for each of the six DWGs will begin, ending at 5:30pm. The YEAR OF ACHIEVEMENT party begins at 5:30pm. Don't miss it. You are invited to send as many representatives as you wish. The DWG break-out sessions at 4:00pm will be: --Design Implementation and Verification --High-level System Design --IP Protection --Mixed-Signal --On-chip Buses --Manufacturing Related Test. Please register in advance, by fax or email. The fax registration form can be downloaded here. Contact the VSI Alliance at: 15495 Los Gatos Boulevard, Los Gatos, CA 95032, USA phone: 1-408-356-8800 fax: 1-408-356-9018 email: nancy@vsi.org. Venue Santa Clara Marriott 2700 Mission College Blvd., Santa Clara, Calif. By special arrangement, a limited number of hotel rooms are available at the Marriott at a rate of $149, single or double occupancy per night. For reservations, call 1-800-228-9290 or 1-408-327-6808. Mention the "VSI Meeting";. The special priced rooms are available on the nights of October 2nd and 3rd. You must make reservations by September 11 to get this rate. Individuals are responsible for their own billing arrangements with the hotel. please register as soon as possible and indicate which DWG breakout session you expect to attend -- so we can make plans for a good meeting. Best regards, Stan Baker Executive Director, VSI AllianceArticle: 7364
Anyone have a clue, what is the simplest and most reliable means for exporting post-compile pinouts from Altera's MAX+II to OrCad or Protel, for generating a schematic symbol for board design? There must be a simple solution, since this is otherwise a horribly tedious, error-prone, and repetitive task. Thank you for your help! -- Bob Elkind **************************************************************** Bob Elkind mailto:eteam@aracnet.com 7118 SW Lee Road part-time fax number:503.357.9001 Gaston, OR 97119 cell:503.709.1985 home:503.359.4903 ****** Video processing, R&D, ASIC, FPGA design consulting *****Article: 7365
<snip> > You can actually pull off things like this. The best one I've seen so far > is a cheap Cirrus Logic PCI VGA card. It cost only $27! How many PCI cards > have you seen for only $27? <snip> > I also have a slightly older Cirrus Logic PCI VGA card: it also is only 4 > chips, but it's a four layer board. It was more than double the price of > this one (around $65 I think). Intel (who was a major player in the creation of the PCI spec) violates the spec. all the time. One prime example is their PCI Ethernet cards....they remove the unused pins from the PCI connector...which is clearly a violation of the spec. Look at pictures of their PCI NICs in any CDW, Insight, PC Complete etc. catalog... Many PCI cards on the market violate the spec, yet they still 'appear' to work. Working is not a very good test of PCI compliance. I 'believe' the scheme you have described does not meet the letter of the spec. I do not have the time to do a detailed analysis of it....but I would suggest you do before you decide to do this in a production board. For a prototype, it may work fine. Austin Franklin darkroom@ix.netcom.comArticle: 7366
I have read a few articles and papers about asynchronous design styles. All approaches make assumptions about delays (e.g. wire delays are significantly shorter than gate delays or a forked wire will have equal delays on both ends or even stronger requirements) or add too much overhead (ack/req-protocols for every register and function) or are not efficiently to implement on a typical FPGA architecture like the Xilinx family of FPGAs (e.g. the C-Element seem to be expensive). So I fear that there is no way to specify an asynchronous circuit with a high-level language like VHDL which is fed into optimizing tools like Synopsys and XACT. Instead the designer or the CAD tool has to control the placement and routing to make an asynchronous design. I would appreciate any hints on literature about asynchronous high-level design with FPGAs, if this is possible at all. The most promising approach seems to me "burst-mode asynchronous controllers" by Steven Mark Nowick, but I'm not yet sure if this is applicable to my problem or to FPGAs at all. Christian. -- PGP: pgp-public-keys@keys.de.pgp.net Subj: GET 0x73036211 Fingerprint: 33 20 99 54 66 C1 1C 32 D9 C1 BC F8 53 AF E4 C6Article: 7367
In article <01bcb7a4$227517f0$35a8b8cd@drt1>,
Austin Franklin <darkroo4m@ix.netcom.com> wrote:
>Many PCI cards on the market violate the spec, yet they still 'appear' to
>work. Working is not a very good test of PCI compliance. I 'believe' the
>scheme you have described does not meet the letter of the spec. I do not
>have the time to do a detailed analysis of it....but I would suggest you do
>before you decide to do this in a production board. For a prototype, it
>may work fine.
Well of course I'm still going to leave pads for the Xilinx prom. Also the
resistors are basically going to be right on top of the lines between the
bus and the FPGA, so if they're not in, there are no stubs. This is a low
volume vertical market product (as most FPGA designs are) anyway.
One the positive side, 74ACT chips only have 4pF input capacitance and much
better setup time requirements than the FPGA, so the resistors can be pretty
big. Without the vendor ID detector, the circuit is only four standard
chips. The clock line has to go to three of them though, so it needs a
smaller resistor or has to be buffered.
--
/* jhallen@world.std.com (192.74.137.5) */ /* Joseph H. Allen */
int a[1817];main(z,p,q,r){for(p=80;q+p-80;p-=2*a[p])for(z=9;z--;)q=3&(r=time(0)
+r*57)/7,q=q?q-1?q-2?1-p%79?-1:0:p%79-77?1:0:p<1659?79:0:p>158?-79:0,q?!a[p+q*2
]?a[p+=a[p+=q]=q]=q:0:0;for(;q++-1817;)printf(q%79?"%c":"%c\n"," #"[!a[q-1]]);}
Article: 7368Hello, I'm looking for some info/parts to do the job of interpolating and decimating a US/Euro HDTV digital stream. In the past I've used the Raytheon TMC2242 to do this on CCIR streams. any suggestions? chad.Article: 7369
I have previously mentioned a Xilinx App Note on synchronous and
asynchronous FIFOs, available on the Xilinx web page as XAPP051.
Here is an improvement, better called a fix, that makes the coding of
the address counters truly Grey, and thus avoids any glitches. I
apologize for the format, the logic looks ugly in its ASCII form, but in
the 4-input look-up-tables it is very simple and compact.
If anybody has a problem, send me e-mail.
If you don't use XAPP051, ignore this long posting.
Peter Alfke, Xilinx Applications
=================================
Asynchronous FIFO
Here is an improvement to XAPP051:
“Synchronous and asynchronous FIFO designs”.
The decoding of Full and Empty, shown in Figure 4, can generate
glitches, because the 4-bit address counters have more than one
simultaneous transition. ( The two upper and the two lower address bits
are individually Grey-coded, but over the whole address, there are two
bit changes at every quadrant border.)
This can easily be changed as shown below.
This change affects only Figure 3. Since each counter bit is driven by a
4-input function generator, the count sequence can be freely adjusted.
New count sequence for read and write counters:
3210
0000
0010
0011
0001
1001
1000
1010
1011
1111
1101
1100
1110
0110
0111
0101
0100
0000
etc
The true hardware cost of this improved design is identical to the
original design: Each counter fits into two CLBs.
D0= ( Q1 •Q0 ) + ( Q1•Q2bar ) + ( Q1•Q3bar) + ( Q0•Q2bar•Q3bar )
D1 =( Q0bar •Q3 ) + ( Q0bar•Q2bar ) + ( Q0bar•Q1) + ( Q1•Q2bar•Q3 )
D2 = ( Q2 •Q3 ) + ( Q2•Q1 ) + ( Q2•Q0) + ( Q0•Q1•Q3 )
D3 = ( Q3 •Q1bar ) + ( Q3•Q2bar ) + ( Q3•Q0 ) + ( Q0•Q1bar•Q2bar )
=========================================
Figure 7: 32--deep FIFO, Read Counter or Write Counter
43210
00000
00010
00011
00001=A, an auxiliary signal, used in the equations below
00101
00100
00110
00111=A
10111
10101
10100
10110=A
10010
10011
10001
10000=A
11000
11010
11011
11001=A
11101
11100
11110
11111=A
01111
01101
01100
01110=A
01010
01011
01001
01000=A
A = ( Q1 XNOR Q2 ) AND ( Q0 XOR ( Q3 XOR Q4 ))
D0 = ( Q1 AND Abar ) OR ( Q0 AND A )
D1 = ( Q0bar AND Abar ) OR ( Q1 AND A )
D2 = [( Q3 XNOR Q4 ) AND A ] OR (Q2 AND Abar)
D3 = {[( Q2 AND Q3 ) OR ( Q2bar AND Q4 ) ] AND A } OR ( Q3 AND Abar)
D4 ={[(Q2 AND Q3bar ) OR ( Q2bar AND Q4 )] AND A } OR (Q4 AND Abar )
=======================
Figure 10: 64--deep FIFO, Read Counter or Write Counter
543 210
000 000
000 001
000 101
000 100
000 110
000 111
000 011
000 010=A, an auxiliary signal used in the equations below
001 010
001 000
001 001
001 101
001 100
001 110
001 111
001 011=A
101 011
101 010
101 000
101 001
101 101
101 100
101 110
101 111=A
100 111
100 011
100 010
100 000
100 001
100 101
100 100
100 110=A
110 110
110 111
110 011
110 010
110 000
110 001
110 101
110 100=A
111 100
111 110
111 111
111 011
111 010
111 000
111 001
111 101=A
011 101
011 100
011 110
011 111
011 011
011 010
011 000
011 001=A
010 001
010 101
010 100
010 110
010 111
010 011
010 010
010 000=A
A= ( Q1 XOR Q4 ) AND ( Q0 XNOR Q3 ) AND ( Q2 XNOR Q5 )
D0 = [( Q1 XNOR Q2 ) AND Abar ] OR ( Q0 AND A )
D1 ={ [(Q0bar AND Q2) OR ( Q0 AND Q1 )] AND Abar} OR ( Q1 AND A)
D2 = {[(Q0 AND Q1bar)OR( Q0bar AND Q2)] AND Abar}OR(Q2 AND A)
D3 = [( Q4 XNOR Q5 ) AND A ] OR ( Q3 AND Abar )
D4 = {[(Q3bar AND Q5) OR ( Q3 AND Q4 )] AND A} OR ( Q4 AND Abar)
D5 = {[(Q3 AND Q4bar)OR( Q3bar AND Q5)] AND A}OR(Q5 AND Abar)
=================================
Summary:
This improves the address count sequence for 16, 32, and 64--deep FIFOs,
so that only one counter bit changes at any transition. This avoids all
spurious decoding glitches that might affect the asynchronous detection
of FULL and EMPTY. This version increase the CLB count only for the
64-address counters by one CLB, and it does not affect any of the
control structures.
For any questions, please send e-mail to: peter@xilinx.com
Article: 7370
Are you sick of all the Home Business hype? Most of them are either too expensive or so complicated that you can't explain them to other people. Are you ready for a simple Home Business that makes sense?
A business that doesn't monopolize all of your time. That requires No: Stocking, Selling or Delivery. No:
billing or collections. No: complicated paperwork. No: employees to pay. No: contracts to sign, you can quit any time. There is a one time $29 enrollment. There is never any new money spent after that. If you decide that this business is not for you simply return the business kit within 60 days for a full refund. Tens of thousands of people have been able to make this their full time business in a reasonably short amount of time. This is not "get rich quick." This is a highly reputable company that opened in 1985. I will send you details on this company when you e-mail me for information.
God forbid, but if something should happen where you couldn't work or a loved one became ill and you had to stay home to care for them, would you be able to? Would your company keep paying you?
Nobody likes to think about these things, but they do happen. And unless you have great insurance you may want to consider this business. You've nothing to lose by asking for information. I'll be happy to send it to you. The fact is, if you are trading time for money your income is temporary, ( even if you own the company.) The average person works from paycheck to paycheck with little or nothing left over to save or invest. This business can help you break the paycheck to paycheck cycle.
For information email me at rramsey36@aol.com or rgr@freedomnet.com
Article: 7371Peter Alfke <peter@xilinx.com> wrote in article <3406BCBE.42F0169C@xilinx.com>... > > Slow designs you can do in a generic fashion, but for really high > performance, you need to decide on a particular manufacturer's part and > development system. Luckily the decision often narrows down to only a > few contenders, sometimes only one. Near the end of the alphabet. > :-) > > Peter Alfke, Xilinx Applications > > interesting to note that there a lot of parts at each of the two ends of the alphabet. then again, i guess you said 'near' the end not 'at' the end ;) rkArticle: 7372
I know, I know, your supposed to instantiate memgen components
but ...
Exemplar's Leonardo claims to be able to infer single port RAM
in 4000e devices. Has anyone been able to make this work. I
sent my test code in to exemplar and they couldn't make it work
with version 4.03 (what I have) but claim lots of people do it
with this version. Is this true ? Anybody out there make this
work?
I'll attach my test code which they don't understand why it
doesn't work. (I'm currently working the support issue, we
buy our licenses from Mentor so Exemplar doesn't want to help.)
Thanks,
--
______________________________________________________________
Allan Redenbaugh allan@ti.com
Raytheon TI Systems voice : (972) 575-7764
P.O. Box 869305 M/S 8454 fax : (972) 575-6965
Plano, TX 75086
______________________________________________________________
LIBRARY ieee;
USE ieee.std_logic_1164.ALL;
USE ieee.std_logic_arith.all;
PACKAGE vdm IS
subtype Vid_Type is signed (11 DOWNTO 0);
END vdm;
LIBRARY ieee;
USE ieee.std_logic_1164.ALL;
USE work.vdm.ALL;
ENTITY test IS
PORT (clk : IN std_logic;
addr : IN integer RANGE 3 DOWNTO 0;
data_A : IN Vid_Type;
data_B : IN Vid_Type;
data_C : IN Vid_Type;
data_D : IN Vid_Type;
dout_A : OUT Vid_Type;
dout_B : OUT Vid_Type;
dout_C : OUT Vid_Type;
dout_D : OUT Vid_Type
);
ATTRIBUTE clock_node : boolean;
ATTRIBUTE clock_node OF clk : SIGNAL IS true;
END test;
ARCHITECTURE synth OF test IS
TYPE Mem_Word IS ARRAY (3 DOWNTO 0) OF Vid_Type;
SIGNAL mem_A, mem_B, mem_C, mem_D : Mem_Word;
SIGNAL we : bit;
BEGIN
we <= '1';
write_proc: PROCESS(clk)
BEGIN
IF clk'event AND clk = '1' THEN
IF we = '1' THEN
mem_A(addr) <= data_A;
mem_B(addr) <= data_B;
mem_C(addr) <= data_C;
mem_D(addr) <= data_D;
END IF;
END IF;
END PROCESS;
dout_A <= mem_A(addr);
dout_B <= mem_B(addr);
dout_C <= mem_C(addr);
dout_D <= mem_D(addr);
END;
Article: 7373Uwe Bonnes wrote: > > In comp.lang.vhdl Kevin Bush <kevin.bush@minc.com> wrote: > : ANNOUNCEMENT: > > : MINC is now offering a full-featured VHDL synthesis engine for $495. > : VHDL EASY(tm) is VHDL-93 and IEEE 1076 compliant and uses a combination > : of > : proprietary and industry standardsynthesis and optimization > : methodologies > : including: > > Is it available for Linux? > > -- > Uwe Bonnes bon@elektron.ikp.physik.th-darmstadt.de > > Institut fuer Kernphysik Schlossgartenstrasse 9 64289 Darmstadt > --------- Tel. 06151 162516 -------- Fax. 06151 164321 ---------- VHDL EASY is currently available for Windows 95 and Windows NT, although we are considering Linux. We still need to look at the impact on our GUI, etc. What other EDA type tools are available on Linux? Kevin Bush, MINC kevin.bush@minc.comArticle: 7374
SNUG '98 Call For Papers
------------------------------
WHAT: Eighth Annual Synopsys Users Group (SNUG '98)
WHEN: March 11th-13th
WHERE: Double Tree Inn, San Jose, CA (formerly the Red Lion Inn)
An Invitation to Contribute
SHARE Your Experiences!
The success of a user's group depends on the active participation of users
who are willing to share their experiences with others. If you have
information on High-Level Design methodology or experiences with Synopsys
tools that would be of interest to other users, you are encouraged to
present in one of the sessions described below.
Preliminary User Breakout Sessions
These sessions are always the hit of the conference. Hear Synopsys users'
experience on specific topics. Each user breakout session will consist of
three presentations, twenty-five minutes each, with another five minutes for
questions and answers. Preliminary topics include:
*Synthesis/Design Productivity
Strategies, experiences, and best practices for design productivity
with an emphasis on synthesis. Automation and Optimization
techniques for synthesis.
*High-Level Verification
Verification strategies covering design for test and system-level
verification. Users share experiences in developing a test bed to
verify the combined hardware and software systems. Design for
test strategies for complex, large designs.
*Higher Levels of Abstraction / Behavioral Synthesis
This sessions covers real-world experiences in making the transition
to high-level design. Topics include the methodology for top-down
design, behavioral synthesis, & high-level techniques for DSP design.
*Deep Submicron / Large Designs
Concentrating on the unique challenges of submicron and large
designs. Sessions provide experience with automating scripts for
submicron, special techniques for managing wireloading,
floorplanning, and non-linear delay modeling.
*Makefiles/Synthesis Scripting
This is a popular session that was added to SNUG last year to
address the increased effort to automate and extend the synthesis
process through scripting. The session includes case studies by
users that have taken advantage of the power of Make and Perl
drive synthesis iterations and to extend DC Shell.
*Configuration Management
Management of files and directory structures is a big key
toward design success. This session will explore approaches
or methods like MAKE used to manage the complex designs of today.
*Design Reuse
This session includes a practical methodology for design reuse
based on real-world experience. Issues and guidelines are explored.
*FPGA Synthesis
This year a new session will be added that will focus directly on
synthesis of FPGAs. Tricks and techniques used for designing and
synthesizing FPGAs will be presented. Experiences with FPGA Express
will also be included.
To present your experiences with a contribution in an user session, please
forward a summary or brief description of your idea to the conference
Technical Committee, (snug_tech@synopsys.com), by October 1, 1997. You
will be notified of your acceptance shortly afterwards. A Technical
Committee member will work with each author to develop and review the paper
and presentation. Please review the Author's kit posted on the Synopsys
web site for details on paper format and structure.
Final drafts are due by January 5, 1998.
Final papers are due for publication in the SNUG proceedings by
February 2, 1998.
A Preliminary SNUG '98 Schedule
Wednesday, March 11th
Morning 1/2 day Tutorial Sessions
Afternoon Industry related Panel Discussion
Evening R&D Cocktail Party /Synopsys New
Product Demos
Thursday, March 12th
Morning Executive Status
Morning/Afternoon User Breakout Sessions
Evening Cocktail Party / Vendor Fair
Friday, March 13th
Morning 1/2 day Tutorial Sessions
Who to Contact
Should you wish to discuss your potential contribution, please feel free to
contact your local Synopsys applications engineering manager or the SNUG'98
Technical Committee via email at:
snug_tech@synopsys.com
All email sent to this alias will be reflected to User Group Technical
Chairperson and the Technical Committee. These addresses are NOT for basic
information on attending the conference itself.
Don Mills, Renae Cunningham,
SNUG '98 User Chairman SNUG '98 Synopsys Program Manager
640 North 2200 West 700 E. Middlefield Road
MS F1-J12 Mtn. View, CA. 94043
Salt Lake City, UT Fax: 415-965-2539
Phone: 801-594-3270 renae@synopsys.com
donald.r.mills@lmco.com
===========================================================================
Trapped trying to figure out a Synopsys bug? Want to hear how 5459 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."
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