Xilinx Methodology Questions : Unconstrained Paths and DLL o

[John Retta]

Hi
I have two questions regarding Xilinx designs.

[1] How do I identify which paths the static timing analyzer considers
to be unconstrained? This has been an ongoing, frustrating task for me.
I am an extreme advocate of synchronous design, and to discover that
I have 97.3% constraint coverage is disconcerting. That means 2.7% are
unconstrained. Clearly the tool has identified these paths, and if there
was a way to display them, this would be helplful in isolating problem
areas.

[2] Where does Xilinx post rules for treating the multiple clock outputs
of DLLs as synchronous? From various postings to the newsgroup, I
get the impression that an clk0 and clk divide by 2, should not be treated
as synchronous, but either there is a small phase offset which precludes
the assumption of edge alignment, or the PAR/TRC tools can not handle
setup/holdtime from the two domains. However, I have not read any
formal limitations placed on these outputs appearing in either data sheets
or application notes.

--
Regards,
John Retta

email : jretta@rtc-inc.com
web : www.rtc-inc.com

 

[Philip Freidin]

On Wed, 2 Jul 2003 08:24:33 -0600, "John Retta" <jretta@rtc-inc.com> wrote:

Hi
I have two questions regarding Xilinx designs.

[1] How do I identify which paths the static timing analyzer considers
to be unconstrained? This has been an ongoing, frustrating task for me.
I am an extreme advocate of synchronous design, and to discover that
I have 97.3% constraint coverage is disconcerting. That means 2.7% are
unconstrained. Clearly the tool has identified these paths, and if there
was a way to display them, this would be helplful in isolating problem
areas.

If you are using TRCE from the command line, it is the "-u xxx" option.
Also available in the GUI.

This adds a new section to the report (after all the normal timespec
sections), that lists unconstrained paths. You then play whack-a-mole
writing new time specs and re-running place/route/trce to make this
section of the report shorter and shorter, until hopefully you get it
to be empty.

Tragically you can get a situation where this section is empty, but
the coverage is still reported as not 100%. Xilinx waves its hands and
says that these are paths that can't/dont need to be constrained, such as
GND and VCC nets, and not to worry. ... I do ...

Sorry, I dont have an answer for your second question.

Philip



Philip Freidin
Fliptronics