( ESNUG 456 Item 8 ) -------------------------------------------- [07/17/06]
Subject: ( ELSE 06 #13 ) 2 more users on their Virtex/Precision experiences
> What we saw during an aggressive Xilinx Virtex-4 project 18 months ago
> was that Precision was unusable and we had to use Synplify. Since then
> Precision caught up significantly and today we use both in parallel.
>
> One interesting observation we made is that Synplify does slightly better
> area optimization, but is slightly more likely to lead to routability
> problems later in Xilinx ISE.
>
> - Alfred Fuchs of Siemens
From: Tsvi Slonim <tsvi.slonim=user domain=matissenetworks spot calm>
Hi, John,
We started using Mentor's Precision RTL more than 2 years ago. At the time,
we evaluated both Synplicity and Mentor and found similar performance
results for both tools. I compared the two products again a year and a half
ago, and Precision still slightly outperformed Synplicity for our design.
Synplicity's plusses were its user interface and particularly the quality of
the diagnostic messages, but we went with Mentor.
Most of our designs are mid-size Xilinx Virtex II Pro 40 and 50. I have
written scripts for Precision, so it is fairly straightforward to use it on
new designs -- it only takes an hour or two for set up. Precision RTL's
runtime varied with the design complexity; for one design it took perhaps
10 minutes, while another design, with multiple resulting netlists, took
30 minutes. However, synthesis speed is not a big consideration relative
to our design process.
Throughout most of 2005, I had many open issues with bugs in Precision
related to our particular design. I had good interactions with Mentor, but
it took a long time to resolve the issues. As of the last revision of
Precision 6 months ago, the bugs have been fixed. I would recommend that in
the future Mentor institute some form of patch or pre-release for a more
timely response to bugs. The main issue that still remains is the quality
of the diagnostic messages. For example there are certain warnings that are
missing. Under certain conditions, constant conditionals were not flagged
and recorded, so the tool would silently optimize away pieces of code,
making it hard to debug. Mentor claims they are planning to correct this.
Precision's ability to meet timing is fine... usually our timing issues
aren't synthesis related but rather are a Xilinx tool issue. At one time,
when we tried Xilinx's XST native synthesis tool, we found that the XST
results were much worse than Precision, and caused timing violations. This
was particularly evident in state-machine optimizations. We did notice a
few cases where Precision did not implement logic ideally. For example,
conditionally incrementing counters would have the clock enable signal
routed into the carry-in input rather than the clock-enable flop input. This
is functionally equivalent but much worse in timing. I believe Mentor has
fixed this. But again, usually our timing issues are a P&R problem, not
synthesis.
Precision's biggest plus is its quality of results. I don't have a lot of
experience with other products, but our designs have yet to perform better
on another product. I have done a lot of synthesizing in the last 6
months... Our Precision flow is very streamlined -- I run it, I get a
result, and I am happy with it. Since most of what I do involves
scripting, I don't use Precision's GUI very much.
- Tsvi Slonim
Matisse Networks, Inc. Mountain View, CA
---- ---- ---- ---- ---- ---- ----
From: Awanish Verma <awanish=user domain=cisco spot calm>
Hi, John,
I began using Precision Physical in August 2005. So far, I have used it on
one project -- a packet forwarding/rewrite engine. It is a high performance
buffer-oriented design implemented in Virtex-4 FPGA. The fastest clock
requirement for the design is 156 MHz, and needed to be margined at 166 MHz.
I chose Precision for its physical synthesis for the design, which had a
lot of big buffers implemented in an FPGA by stitching in the block RAM
(in Xilinx). Placement of memory and the control logic was an important
criterion.
Setting up the environment was easy, plus I had assistance by the Mentor
FAE. I used it through the CLI in LINUX environment (no GUI).
Precision's biggest advantages are:
1. Multi-pass physical synthesis. Precision performance was similar to
other synthesis tools for a single pass, but its multi-pass results
are better. I got better results for reiterative synthesis by using
Precision physical synthesis for multi-pass place and route.
2. It takes the input from place and route results so timing modification
on critical paths are realistic as compared to rough estimates.
3. Interactive timing analysis with cross probing to see the result of
modified placement.
Improvements needed:
1. We had to keep it updated with the releases of the newest versions of
the Xilinx place and route tools.
2. Stability. When I used it in August 2005 I faced a lot of crashes in
UNIX/Linux environment, which Mentor acknowledged was a bug. Also
there was a bug with the use of Xilinx DCM (frequency divider), which
might have been fixed there were couple of other bugs with Virtex-IV.
Mentor says the UNIX/Linux issue will be fixed by Q306.
3. Make it more 'industry standard'
Mentor's support for Synopsys Design Constraints (SDC) was good, as well as
generating the netlist and scripts which can be analyzed in PrimeTime. I
have only had limited use of Precision's cross-probing and interactive
timing analysis but my first impression is that they were very useful.
On a scale of 1 to 10, I would rate Precision Physical an 8, provided it is
stable. I would surely recommend the more stable version of the tool.
- Awanish Verma
Cisco San Jose, CA
Index
Next->Item
|
|
