Problem restoring saved artirst states on Linux (ic5033)

Hi,

I have a problem restoring saved artist states for simulation setups in
icfb. When I try to restore a saved state, it does NOT display all the
saved states.

The version of cadence I'm using is given below. This is running on a
linux machine.

Program: @(#)$CDS: icfb.exe version 5.0.0 05/21/2005
17:35 (intelibm12) $
Sub version: sub-version 5.0.33.500.3.33 (32-bit addresses)

uname -a output:
Linux 2.6.11-1.1369_FC4smp #1 SMP Thu Jun 2 23:08:39 EDT 2005 i686 i686
i386 GNU/Linux

I saw an old post (feb 06) also about this issue ...but no answers. Is
this a bug on the Linux version? If yes, is there a patch? Or is there
a workaround?

Thanks!
Raghav

 

Any thoughts... Any one ??

Thanks!
Raghavendra

 

I checked my setup, but problem is that I run 5.1 and would then not
see the problem that you have in case it has been fixed between the
release of your version and my version.

I checked the other posting, but did not get any wiser. A year ago we
were also using 5.0 and subversions, but I cannot ever remember to have
seen a problem with incompletely written states. Is the directory where
artist saves the states on the same disk as the rest, or could it be
that it is on a different disk? Do a df -k . in both the .artist_states
directory and wherever your work directory is. If they are on different
disks, it /may/ happen that you have an NFS problem, but I emphasise
/may/ as this is just a wild guess. NFS is always a nice service to
blame when something goes wrong ...

 

Are you seeing "." and ".." showing up in the list of states? If so, the problem
may be one that showed up recently as certain file systems on Linux started
being used which happened not to return . and .. as the first two entries in
the directory. The code that showed the available states made the assumption
that . and .. were the first two entries, and stripped those - this was an
empirical assumption that worked for the last 10 years - but with the advent
of some different file systems on Linux has proved not to be true any more...

The code has been now changed to explicitly filter . and .. rather than assuming
where in the directory list they appeared. This was fixed in an IC5141 ISR - in
5.10.41.500.2.26

Andrew.

 

I believe that is not a filesystem problem (the filesystem doesn't sort the
filenames, it stores them in some internal order) but a sorting algorithm
change based on locale. The default locale on UNIX systems has been C for a
long time. The newer Linux systems are using en_US.UTF-8 so that
internationalized applications work correctly.

To get the default sort order back without setting LANG=C, please set
environment variable LC_COLLATE=C.

Satya

 

getDirFiles() doesn't return the files sorted - it just returns them in the same
order that the C readdir() function returns them.

In most OS versions this is the order that the files were added to the
directory, and so . and .. were always the first two entries. However, newer
systems don't store it this way (I guess they are using some kind of hash).

We started seeing this as people started using RHEL 4.0 and Fedora Core
releases of the same vintage, but my guess is that it is related to the file
system rather than the OS version, since it doesn't always happen on RHEL 4.0

Andrew.

 

Hello Andrew, Satya,

Thanks for your replies.

I tried this, but artist still shows . and .. as the first entries when
you try to load a saved artist states.

As noted, this does not always happen. Sometimes artist loads the
states without any problems but shows . and .. sometimes. Also, if I
can sometimes solve the problem if I do this:

$ cd ~/.artist_states/LIBRARY_NAME/CELL_NAME/spectre
$ cp -rp artist_saved_state new_state

I can see a new_state as an option below .. in my artist, load option.
This makes me wonder if it also has something to do with directory time
stamps.

A colleague of mine uses Mandrake Linux. Apparantely he has never seen
this problem restoring Artist states.

In summary, how would I obtain this fix? I am using 5.0.33.500.3.33.


Raghavendra

 

You would need to install IC5141 USR3 or a later ISR. There is no fix in IC5033
(IC5033 is no longer supported anyway).

Out of interest, what flavour of OS are you using, and do you know what type
of file system it is that the artist_states dir is on? (if it is nfs, what is
the OS and file system type on the server?)

Regards,

Andrew.

 

Can you please suggest a work around for this problem in IC5033?
Sometimes I loose all my saved artist states and have to create
everything back again

Fedora 4 (FC4) running on Athlon. NFS not used. Local ext3 filesystem.
Below is the info:
$ uname -a
Linux hostname.xx.xx 2.6.17-1.2142_FC4 #1 Tue Jul 11 22:41:14 EDT 2006
i686 athlon i386 GNU/Linux
$ pwd
/home/USERNAME/.artist_states/test_bench/test_telota_pss/spectre
$ df -k .
Filesystem 1K-blocks Used Available Use% Mounted on
/dev/mapper/VolGroup00-LogVol00
27869052 7433324 18997216 29% /
$ mount
/dev/mapper/VolGroup00-LogVol00 on / type ext3 (rw)

-Raghavendra

 

I sent you a hack by email to your posting address. If that's not a valid
address, please email me (you can figure it out from my posting address),
and I'll send it to you.

I don't want to post the hack in an uncontrolled forum - because I'd sooner
limit its use - the right thing to do is use the fixed version.

BTW, Fedora Core is not a supported distro for Cadence tools...
That means no testing is done on it.

Regards,

Andrew.

 

Hello Andrew,

I am having the same Problem here. Could you please send me this hack, too?

Another problem: I'm going to install the latest version, but I don't
like the new calculator is there a way to use the "classical version"
with IC5141?

Thanks,
Mario

 

Hi Andrew,

I am noticing the same problem using RHEL 4 with 5.0.33. Could you
send me the hack as well?

thanks,

 

Session->Options - you can choose AWD as the waveform tool to get the old one.

However, I'd recommend you give it a reasonable try (I suggest going to IC5141
USR4 at least). The old one will not be there in the next release (IC610), so
you need to get used to it!

Andrew.

 

Oh, oh, better get the helmet out. It was quite hot around here when
Wavescan turned out to be default viewer. That calmed down when the
option to switch was explained. I won't like to be around when the guys
discover that awd is gone completely. It is amazing how picky
analog designers are on their waveform viewers. On the other side,
inspecting waveforms is our business so one would like to have the
*best suited* tool available. It turns out that, for some reason, awd
is better suited than wavescan even if it misses a lot of the
interactive possibilities that wavescan has. Or is it just that analog
designers are more reluctant to learn anything new than other designers
(the digital ones.....)?

 

I think wavescan has created such a bad impression in earlier releases,
that it will be hard to convince people to use it. For some reason,
Cadence had decided to dump all their learnings on AWD over the years
to one side and get every single (well, almost) small detail wrong.
Performance is important, but not at the expense of correctness.
Wavescan was wrong way too often.

The one in USR3 (the officially approved version at my company) is
quite an improvement over the earlier releases. It still has a long way
to go though.

Satya

 

I have supervised a student and purpously told him to use wavescan
instead of awd just because I know awd will not be around when he is
finished with his studies. I really hope for him (and the rest of us)
that wavescan won't either. It is difficult to create a good waveform
viewer. Many companies have tried, but there are so many disiplines and
it is impossible to cover all of them with one tool. Not to forget all
the incompatible data storage formats. That Cadence obviuosly didn't
want to improve awd even if it had one of the more inviting printing
capabilities (hardcopy, creating plots for the digital archive was a
bit more hassle) must lie in the fact that they either don't know the
sourcecode and it is so spaghetti or they know the codebase and it is
so spaghetti that they had to drop it. The choice to move to Qt as GUI
toolkit probably solved a couple of problems for Cadence. We have seen
a lot of java sneaking in lately, and from my experience, java isn't a
language suited for EDA. I am just beginning to be a bit fed up with
waiting for Goudou when it comes to Cadence improving their front-end
tools (specially the waveform viewer and the simulation dashboard). But
there is nothing I/we can do since they keep their bread-and-butter
storage format proprietary. (psf-bin). There are some third party
providers of tools like veritools and sandworks who has waveform
viewers, but that is not an option for large companies who then have to
buy a lot of extra licenses and send designers to a lot of different
trainings. I have tried a lot of these "auxillary" waveform tools and
they all improve the field-of-interest which made the tool vendor
create a waveform tool, but they all fail in making the swiss army
knife.

The problem ist basically that large companies have a serious lag-time
in implementing new versions into their design-flows. For Cadence, in
my opinion, this clearly lies in the difficulty of integrating a lot of
designers on a project and still keep schematics up to date and
accessible to all. Yes, there are a lot of version management
available, but all of this require so much integration work that when a
new version of IC arrives at the market, I know it will take long time
before I see all those benefits at my desk. It looks like Cadence tools
grow to become the SAP of the EDA world where more millions are paid in
integration work than in the actual licenses. The loser is the engineer
who has to be on the edge of semiconductor development with tools that
are a couple of generations old just because the integration of the new
tools can't manage to keep up with the tool improvement.

The winner are those companies who can use the tools out of the
shrink-wrap. They can follow the tool improvement even up ahead with
pre-releases etc. This is typical for small start-up companies where
all engineers are experienced and there are some guys who know Cadence
in their sleep and can fix about any problem and at the same time do
design so that he knows where the shoe hurts. In large companies we
have to do roll-outs and roll-ins and double-ticket management and
whatever you do to organize a lot of information so that nothing get
lost.