assura RCX - find nets by net base name

M

Marcel Preda

Guest
Hi all,

I have an assura extctracted view, with parasitic info.
The nets and associated shapes were fragmented.
I need a function to get all the nets derivative from a baseName net.
because I need to do something with their respective shapes.

e.g. "GND" has become a list of nets:
"GND", "1:GND:, ...."45465:GND".
I need function to pass "GND" as argument and to get all those nets.

For now I'm using a code like:
~~~~~~~~~~~~~~~~~~~
baseName = "GND"
rexCompile(strcat("[0-9]*\\:*" baseNet "$"))
setof(cvNet cv->nets rexExecute(cvNet->name))
~~~~~~~~~~~~~~~~~~~~~
It works but it is slowly, for ~ 300 such nets have to wait about 15
minutes.
I've used skill profiler, and less than 1/4 of total time is used by
rexExecute,
about 75% of time is used to scan cv->nets .

Is there any faster solution ?

I would like to have a function dbGetNetsByBaseName(s_baseName cv).
If such a native function exists I'm sure that is faster than what I
have now.

Best Regards,
Marcel
 
Hi Marcel,

Rather strange, you do not have a loop around this, do you ?
The above should very quick and I am surprised it took 15 mins for a
small database as yours. In fact, 300 nets is nothing.
You might post your whole function may be.

For the other bit, there is a function called dbFindNetByName

Cheers,
Riad.
 
Hi Riad,

thank you for reply.
May be I did not tell exactly what I wanted to tell :)

The total number of nets in my av_extracted view is 828 717 (including
parasitic nets).
And I have to call in a loop the code from my 1st exactly 322 times.
There are 322 net names which satisfy some criterion and I have to do
some check against them.

E.g. I run it for:
"GND", "VCC:, "OUT0", "OUT1", "BLOCK1|net111", ... etc (in total are
322 nets)
And I get the results:
myProc("GND" cv) returns: "GND", "1:GND", "2:GND" .. there are 17536
nets in this list.
myProc("VCC" cv) returns a list of 11303 nets

and so on...

I already have an idea: to scan the cv->nets list only once and using
some regexp function to build an associative array,
each key will be the net base names (e.g. "GND:, "VCC", ...) and each
value will be a list with "derived" nets:
"GND" => "GND", "1:GND", "2:GND" ...
"VCC" => "VCC", "1:VCC", "2:VCC" ...
I'm sure it will be faster, but it will lead to a huge memory ussage.
I still hope that there is some other solution.

Best Regards,
Marcel

On Oct 29, 2:56 pm, Riad KACED <riad.ka...@gmail.com> wrote:
Hi Marcel,

Rather strange, you do not have a loop around this, do you ?
The above should very quick and I am surprised it took 15 mins for a
small database as yours. In fact, 300 nets is nothing.
You might post your whole function may be.

For the other bit, there is a function called dbFindNetByName

Cheers,
Riad.
Click to expand...
 
marcel wrote, on 10/29/09 14:51:
Hi Riad,

thank you for reply.
May be I did not tell exactly what I wanted to tell :)

The total number of nets in my av_extracted view is 828 717 (including
parasitic nets).
And I have to call in a loop the code from my 1st exactly 322 times.
There are 322 net names which satisfy some criterion and I have to do
some check against them.

E.g. I run it for:
"GND", "VCC:, "OUT0", "OUT1", "BLOCK1|net111", ... etc (in total are
322 nets)
And I get the results:
myProc("GND" cv) returns: "GND", "1:GND", "2:GND" .. there are 17536
nets in this list.
myProc("VCC" cv) returns a list of 11303 nets

and so on...

I already have an idea: to scan the cv->nets list only once and using
some regexp function to build an associative array,
each key will be the net base names (e.g. "GND:, "VCC", ...) and each
value will be a list with "derived" nets:
"GND" => "GND", "1:GND", "2:GND" ...
"VCC" => "VCC", "1:VCC", "2:VCC" ...
I'm sure it will be faster, but it will lead to a huge memory ussage.
I still hope that there is some other solution.

Best Regards,
Marcel
Click to expand...
Hi Marcel,

A few suggestions:

1. Create the list once:
nets=cv~>nets

and then do the setof. This avoids re-building the list every time you do
cv~>nets which should really help (the database is not really stored in a
list, so each cv~>nets is constructing your 800K list cells each time)

2. You could also try:
netNames=nets~>name ; again, do this just once
matchedNames=rexMatchList(pattern netNames)
; assuming you actually need the net objects...
matchedNets=foreach(mapcar netName matchedNames dbFindNetByName(cv netName))

This might be quicker, because the number of matches is relatively small, and
transforming to build a list of net names is likely to be small.

3. Try pre-allocating space with needNCells :
needNCells('list 1000000)
needNCells('dbobject 1000000)

This is worthwhile if you find it spending a lot of time in gc() when you
profile the code.

Sorting the net names into a table that you suggest probably won't use that much
memory actually (compared with the extracted view), but the work of traversing
the list of net names once and then building the sublists could end up being no
better than suggestion 1 or 2 above.

Regards,

Andrew.
 
On Nov 2, 10:06 pm, Andrew Beckett <andr...@DcEaLdEeTnEcTe.HcIoSm>
wrote:
marcel wrote, on 10/29/09 14:51:



Hi Riad,

thank you for reply.
May be I did not tell exactly what I wanted to tell :)

The total number of nets in my av_extracted view is 828 717 (including
parasitic nets).
And I have to call in a loop the code from my 1st exactly 322 times.
There are 322 net names which satisfy some criterion and I have to do
some check against them.

E.g. I run it for:
"GND", "VCC:, "OUT0", "OUT1", "BLOCK1|net111",  ... etc (in total are
322 nets)
And I get the results:
myProc("GND" cv) returns: "GND", "1:GND", "2:GND" .. there are 17536
nets in this list.
myProc("VCC" cv) returns a list of 11303 nets

and so on...

I already have an idea: to scan the cv->nets list only once and using
some regexp function to build an associative array,
each key will be the net base names (e.g. "GND:, "VCC", ...) and each
value will be a list with "derived" nets:
"GND" => "GND", "1:GND", "2:GND" ...
"VCC" => "VCC", "1:VCC", "2:VCC" ...
I'm sure it will be faster, but it will lead to a huge memory ussage.
I still hope that there is some other solution.

Best Regards,
Marcel

Hi Marcel,

A few suggestions:

1. Create the list once:
    nets=cv~>nets

    and then do the setof. This avoids re-building the list every time you do
    cv~>nets which should really help (the database is not really stored in a
    list, so each cv~>nets is constructing your 800K list cells each time)

2. You could also try:
    netNames=nets~>name ; again, do this just once
    matchedNames=rexMatchList(pattern netNames)
    ; assuming you actually need the net objects...
    matchedNets=foreach(mapcar netName matchedNames dbFindNetByName(cv netName))

    This might be quicker, because the number of matches is relatively small, and
    transforming to build a list of net names is likely to be small.

3. Try pre-allocating space with needNCells :
    needNCells('list 1000000)
    needNCells('dbobject 1000000)

    This is worthwhile if you find it spending a lot of time in gc() when you
    profile the code.

Sorting the net names into a table that you suggest probably won't use that much
memory actually (compared with the extracted view), but the work of traversing
the list of net names once and then building the sublists could end up being no
better than suggestion 1 or 2 above.

Regards,

Andrew.
Click to expand...
Hi Andrew,

Thanks for the hints.
Now the entire loop is 5-6 x faster .

Best Regards,
Marcel
 
You must log in or register to reply here.

Welcome to EDABoard.com

Sponsor

Back
Top