( ESNUG 426 Item 6 ) -------------------------------------------- [03/31/04]
Subject: Ten More Letters On Mentor/Exemplar/Leonardo/Precision
> Scenario 1: Wally's Revenge
>
> Synplicity was founded by ex-Mentor synthesis gurus Ken McElvain and Andy
> Dauman. For Wally Rhines, the CEO of Mentor, it's a wee bit embarrassing
> to have his ex-employees best him in a market that Mentor used to own.
> If it turns out that Synplicity sales really did drop 31% while Mentor
> sales remained steady in FPGA synthesis, Wally gets to laugh as he watches
> Bernie Aronson, the CEO of Synplicity, do his best imitation of the
> Help-I've-Fallen-And-I-Can't-Get-Up lady. Sweet, sweet revenge....
>
> - from http://www.deepchip.com/gadfly/gad031204.html
From: Dan Cohen <dan.cohen=user domain=andrew spot balm>
Hi John,
We have been using Precision synthesis for a digitally pre-distorted
amplifier design that uses 3 large Vertex-2 parts. We evaluated Synplicity
against Leonardo and slightly preferred Synplicity but have found Precision
a step improvement over both. Precision reduced the 2.5 hour synthesis
time to a little under 10 minutes and the results were slightly better
than with Leonardo.
- Dan Cohen
---- ---- ---- ---- ---- ---- ----
From: [ An Anon Engineer ]
Hi John,
We use both Synplify and Precision for our FPGA's Designs.
We're currently using Precision 2003c and Synplify 7.3, on Altera
Stratix and Xilinx Spartan3/Virtex/Virtex2 technologies.
Last year we performed an evaluation using some REAL designs on Solaris
WS and we got:
- Synplify was always much faster
- The area and timing results, considering post P&R values, were
similar: Synplify was sligthly better
For currently under development designs we have decided to use Precision
because of:
- "strange" problems during Synplify Mapping phase (in one design)
- incorrect inference control of MULT18x18 Virtex/Virtex2 multipliers
with Synplify
- Precision "ASIC flavour" in scripting methodologies better suits our
synthesis methodology (e.g. user friendliness in setting timing
exception constraints).
Please, keep us anonymous.
- [ An Anon Engineer ]
---- ---- ---- ---- ---- ---- ----
From: Florian Jutisz <florian.jutisz=user domain=motorola spot balm>
Hello John,
I have been using Leonardo Spectrum for four years. I have never really
chosen this tool - it was the standard tool where I started to work. Since
then, my responsibilities have changed and I could now decide to change
the tools we use if I am willing to. However, I have always been pretty
satisfied with Leonardo: synthesis results are good, the tool runs fast
(we work on multi-million gates designs) and a very nice feature is that
you can run the tool in batch mode. You just launch a shortcut and your
synthesis script is executed in background.
In addition, the graphical display is nice, and features like incremental
synthesis (incorporate already synthesized design without re-synthesizing
it) run pretty well.
Main trouble we have with Leonardo today is that Mentor will not support it
anymore in the next years, replacing it by Precision Synthesis. Thus, they
do not implement any real new features in it, for example Leonardo does not
support Verilog 2001, while other Mentor Graphics tools in the flow support
it (HDL Designer, ModelSim). Another drawback of this transition is that
Precision is not compatible with Leonardo Spectrum scripts: we have like
35 scripts that will become unusable!
This means the day Mentor stops supporting Leonardo to force their customer
transition to Precision (and buying new licenses), I will have a look at
all existing synthesis tools (like Synplicity) and make a real comparison
to choose the most appropriate for our needs.
- Florian Jutisz
Motorola Toulouse, France
---- ---- ---- ---- ---- ---- ----
From: Larry Kerr <lkerr=user domain=analogic spot balm>
Hi John,
I am an FPGA design engineer at Analogic.
I have been using Leonardo Spectrum for about two years. Analogic owns
several site licenses for Leonardo, as well as Precision Synthesis. All
of my designs target Xilinx FPGAs, exclusively. I have designed with
Virtex, VirtexE, VirtexII-1000, VirtexII-6000, VitrexII-Pro-20, and
VirtexII-Pro-30. I have Leonardo integrated into Xilinx Project Navigator
and it work great for me.
- Larry Kerr
Analogic Corporation Peabody, MA
---- ---- ---- ---- ---- ---- ----
From: [ An Anon Engineer ]
Hi, John,
We are Leonardo/Precision users. We purchased Leonardo in 1999 as part of
a Mentor HDL Designer/Modelsim/Xilinx/PCB flow. At the time we didn't
see significant differences between Leonardo and FPGA Express. In actual
use, we found Leonardo somewhat difficult (lots of obscure options), and
somewhat buggy (crashes with no message for certain errors).
More recently, we had an outside design house do a project, they used FPGA
Express. I started converting their design to Leonardo, and was having
some difficulty. Then the Precision Synthesis upgrade was released, I
found it very easy to use, and much less buggy, and will use it on all
future designs.
One thing I find puzzling about most of these tools. They seem to be
promoted as if everyone does everything only once, using the GUI. In
reality, you have to create scripts so you can run the tool the same way
the next time you do a revision (much more common in the FPGA world).
- [ An Anon Engineer ]
---- ---- ---- ---- ---- ---- ----
From: Mike Karlinsky <mkarlin1=user domain=motorola spot balm>
John,
These are my answers regarding my experience of using Precision tools:
1. I believe Motorola bought Leonardo, Precision and Modelsim as a
package but I am not sure. I used Leonardo before. Now I switched
for Precision. I use Modelsim for about 8 years.
2. I use mostly Xilinx Virtex2 and Spartan3 families.
3. Last time I used Synplicity was about 4 years ago. Compared with
Leonardo I liked Synplicity better. Between Precision and Synplicity
they both are about the same.
Yes, I would definitely buy Precision Synthesis tools and Modelsim tools.
If to choose between Precision and Leonardo I would choose Precision.
- Mike Karlinsky
Motorola Schaumburg, IL
---- ---- ---- ---- ---- ---- ----
From: Steve Setzer <steve.setzer=user domain=flir spot balm>
Hi, John,
We purchased Precision Synthesis as part of the FPGA Advantage package
last year. I am currently using these tools for a pair of Xilinx
Virtex II 2V1500ff896 devices. These designs are mathematical and data
flow intensive on video streams. I am using Precision with the register
retiming feature in one case at 80 MHz. I have taken the flow through
the integrated Xilinx ISE place and route tool. We will be out in the
lab debugging these designs in two weeks.
- Steve Setzer
FLIR Systems Billerica, MA
---- ---- ---- ---- ---- ---- ----
From: Israel Cruz <israel.cruz=user domain=honeywell spot balm>
Hi, John,
I started off doing synthesis with Leonardo. I have used Synplify Pro as
well. I generally agree with most people that Synplify Pro gives good
results and is faster and easier to use. However, I like the many options
given to you with the Leonardo tool. If you are really concerned about
ease of use and the quickness of the run, you can always use the Leonardo
Wizard option. But in past designs, I have had to go into the Leonardo
menu options to change certain things, and I noticed recently that I could
not do this easily with the Synplify Pro tool.
- Israel Cruz
Honeywell Clearwater, FL
---- ---- ---- ---- ---- ---- ----
From: [ An Anon Engineer ]
John,
I started using Leonardo in 1998 and continued to use it until Precision
came out. I now use Precision. I've seen and tried Synplify and Synplify
Pro- I just don't care for it. There's something about that big "Run" button
that makes it look like a toy. But, I'll admit that Synplify results use
fewer resources compared to Leonardo. And as for performance claims, I never
believe any synthesis tool. I let the place-and-route tell me the truth.
Targets: Xilinx Virtex, VirtexE, Virtex 2; Actel SX, AX families.
The reason I preferred to stay with Leonardo was the fact that I could write
simple scripts. In the scripts I could include all my files, set attributes,
and write out the .edf file. I didn't need to create or save a project and
all the miscellaneous file it that came with a project. I only wanted an
.edf output. Additionally, I could run these scripts at the command line.
Also important was that I could set attributes in the scripts. I prefer to
put the attributes in the scripts as compared to littering my code with
attributes. Synplify only gave me the choice of adding attributes to my
code. With Leonardo I could add them to my code or the script. I chose the
script. That meant I could easily port my code to another tool such as
Design Complier to target an ASIC library later.
With Precision came better synthesis results over Leonardo. I would say the
results are similar to what I see with Synplify Pro. Unfortunately,
Precision became a little more like Synplify in terms of feel. However, I
can still create scripts and set some attributes in them- attributes (such
as preserve) that I would need to set in code in Synplify. In general I
would say that I am happy with Precision. I say "in general" because I can
always think of improvements no matter how much I like the tool.
I have always been satisfied with Mentor support. They have some great
staff- including someone I have gotten help from that I think in located in
the UK. I also have to say the Technical Marketing Engineer was able to help
me get answers and early versions of a particular release of Precision when
I would find a problem with the version I had.
I plan to continue to use Precision as my primary FPGA synthesis tool. As a
designer, I believe Precision gives me more options to solve problems than
Synplify does. And that's what I want from a tool. Options.
As a side note, Synopsys DC-FPGA looks like interesting. I'm willing to
try it when it's available. Though as we all know, previous Synopsys FPGA
tools have been terrible. DC-FPGA seems like a good way to use similar
scripts for both FPGA synthesis and later an ASIC port. I do think that
if I knew I was going to stay with an FPGA for the end product I would
stay with Precision though.
I would prefer if you kept my comments anonymous.
- [ An Anon Engineer ]
---- ---- ---- ---- ---- ---- ----
From: Lars-Goran Davidsson <lars-goran.davidsson=user domain=ericsson spot balm>
Hi, John,
In our normal FPGA flow we use Synplify Pro for synthesis before we do
place and route with Xilinx tools and often have to use the '-timing'
option when we do the mapping.
In our projects we had two FPGAs who did not meet timing. One because of
the cell placement and one because the FPGA was nearly full.
We decided that we should try to solve this with Precision Physical and
in both cases was this succesfully solved by Precision Physical.
The first case:
Target Xilinx Virtex26000-6, System frequency 123 Mhz, 100 of 144 Block
ram used, 60% of logic slices used. After P&R we missed the goal with
slack of 1.0 - 1.5 ns. We test a lot of different timing setups and
seeds for the synthesis but was not able to meet the timing.
Then we started using Precision Physical. We compiled and did a first
physical placement and a new P&R, and then we missed the target by
200ps. We compared this slack between the Xilinx TWR and Precision
Physical timing result and over constrained the clock by 5%. A new
physical placement run was done with a P&R and then we met the timing by
10ps. What Precision Physical did was to move and swap some cells and
some cells were duplicated. Note that we only used press button no
manual placement or swapping of cells.
The second case:
Target was a Xilinx2 1000 -5, System frequency was 123 Mhz. The FPGA was
nearly full. The Xilinx mapper repported nearly 20% of unrelated logic,
65% of the MULT18X18s and 60% of the RAMB16s.
The main problem was bad placement of memories and mults inside the
FPGA. This was solved by manual swapping of a few memories and mults.
After swapping the blocks we started the automated flow in Precision
Physical to solve the remaining timing violations in this the FPGA. The
improvment of timing was about 5-10%.
- Lars-Goran Davidsson
Ericsson
|
|
