Is there any way to make XL functions (those with prefix "lx") run faster in the Layout L environment, for example by loading the necessary context manually? Here's the speed difference (in CPU time) as measured by the profiler for the lxComputeViaParams function used in a script (this function is called multiple times):
Layout L: took 0.71 seconds to run
Layout XL: took 0.06 second to run
Any suggestion? Thanks!
What I see (when running in L) is that the first time around it takes longer (in my case 0.4 seconds) but the second time it takes 0.02 seconds. It seems the first time that a via is referenced takes longer - probably as it has to build the structures for it the first time around?
If I'm running in XL, it's fast straight away.
I think the most likely explanation is that XL does the Via initialisation as part of its startup, whereas L does it when the vias are used.
And by the way, in my case it was 0.4 seconds for looking at 10 different viaDefs; once I'd looked at each once, it was much faster. 0.04 seconds for the first time to access each viaDef doesn't sound "extremely" slow to me.
Thanks for the feedback. Unfortunately I am not able to reproduce the cache structure like you did. It doesn't matter how many times I call the function, the speed is always about the same i.e. slow. FYI, I'm using IC 6.1.3.
This is what I observed (and before I get to this conclusion, I was scratching my head for half a day, wondering why is my script suddenly 10x slower). First I ran my script in L environment, it takes about 8 seconds to run. Next I switch to XL environment for the same layout, the script takes less than 0.5 second to run. Then, I switch back to L environment, and now the script takes 8 seconds to run again. Undo the script, run it again in L, still 8 seconds, it doesn't speed up at all as what I would expect from the cache structure you mention.
btw, as long as one of the layout windows is in XL mode, the script takes less than 0.5 second to run. Once all of them are in L, the script slows down again.
I'm not sure if this has anything to do with some of the context optimization done in XL to speed some of the stuff up, but definitely there's something different that is causing the huge difference in function execution speed.
The best thing to do is to contact Cadence Customer Support and create a service request, with some data and representative SKILL code which shows the problem. It may well be that my testing was just way too simple compared with what you're doing (since I could only spend a few minutes on it).
If it's consistently repeatable, and you're using a recent hotfix and it still happens, I think we need to get it into R&D to get an improvement done so that L doesn't suffer from this. I can't see why it should be consistently slower with L.
Hi Andrew, as suggested I filed a service request and it turns out
this issue is related to license checkout. The workaround provided by
Bhupendra, the support I worked with, is to use lmCheckOut() and
lmCheckIn() to manually checkout the license before the script is ran
and check it back in once it's done.