Using functions in Assembler global variables

Hi Brad,

First of all, this only stands any chance of working if the expressions are valid SKILL expressions and can be computed before simulation starts. It sounds as if ADE has sent the expressions to spectre, but that's going to fail because spectre does not have a function complex in its expression language.

Normally ADE will do some checks on expressions in the global variables to see whether it thinks they can be handled by spectre, and if so passes them to the simulator. Otherwise it tries to evaluate them in SKILL, and then sends the results to the netlist. If the SKILL evaluation fails for some reason, then it still passes the expression to the netlist on the off chance that spectre will be able to handle it anyway.

Can you please share the details of your expressions so that I can check? I may be able to figure out what's wrong and give you a workaround.

Regards,

Andrew

Hi Andrew,

I misspoke when I said I was using function complex(). Actually I'm creating a complex variable by using sqrt(-1) in an expression and this works fine in the CIW. The error returned in the spectre.out file includes the phase "Function 'complex' is not defined.." and that lead to my not so clear description (it was late in the evening).

I'm attaching the expressions now.

Regards,

Brad

Brad,

This is a tricky one. What's happening is that there is an issue where sqrt() was not recognised as a spectre function, and so was being calculated in SKILL always. Usually you want as many functions that could be evaluated in spectre to be passed to the netlist, as that then allows sweeps within the simulator and dependencies to be tracked there. We've just fixed the issue with sqrt not being passed to the simulator - this will be fixed in IC6.1.8 ISR26 (out in July if my memory is correct).

However, that's not going to help you. Spectre can't handle complex arithmetic, so in fact you really don't want the expressions to be passed to spectre. Unfortunately most of the expressions here appears to ADE to be the kind of things that spectre would be able to handle, and it doesn't know that the result of the calculation for j would prevent that. You're getting the complex(0, -1) in the netlist because ADE evaluated the sqrt(-1) but it is going to fail because nothing knows how to deal with complex(0, -1). Even if ADE passed it through, it would ultimately fail with something like:

  FATAL (SFE-1996): "testsqrt.scs" 3: Cannot run the simulation because the
expression of the parameter `i': Attempt to take the square root of a
negative number (-1).
, correct the expression and rerun the simulation.
 

The way to workaround this is to create some functions and load from your .cdsinit:

defMathConstants('CCFmath)
procedure(CCFcalcG(Gabs Gang)
  let(((j sqrt(-1)))
    Gabs*exp(j*Gang*CCFmath.PI/180)
  )
)

procedure(CCFcalcZ(Gabs Gang)
  let((G)
    G=CCFcalcG(Gabs Gang)
    (1+G)/(1-G)
  )
)

Leave the definitions of Ro, Xo, Gang, RLoss_dB, Gabs as you have them, and remove the j and G expressions. Then make:

Z=CCFcalcZ(Gabs Gang)

You'll still see the Z appearing in the netlist as complex(NNN, MMM) but that is OK - because nothing references it in the netlist.

Yes, it's a bit of a hack - but at the moment there's no clear way to force an expression to definitely be handled in ADE or in Spectre.

Regards,

Andrew

Andrew,

Thanks for the quick response and helpful work around.

A few related questions -

Can I preload an .il for a given assembler session without manually using load("my.il") in CIW or including in my .cdsinit? Just wondering what my options are.

I'm not familiar with CCFmath I assume it is an .il file? Where can I find the file and can I inspect the contents?

Is there a def file that includes matrix math expressions?
I'm somewhat new to SKILL but I'm already thinking about the possibility of overloaded operators to handle matrix math.

"it's a bit of a hack" [Andrew]

I don't mind working with custom functions in fact I like the possibilities that approach opens up. On the other hand it sure seems
the Cadence development guys really don't have RF designers in mind otherwise this basic RF 101 task that I attempted would have been easily handled in GUI from day 1.

Regards,

Brad

CCFmath isn't anything special - it's just a name. If you do:

defMathConstants('someName) 

then you can do:

someName.PI ; returns pi
someName.? ; shows available names
someName.TWO_OVER_PI

All that was for was to provide a convenient SKILL way of getting an accurate pi constant without needing to type 3.14159265358979323.

There's no standard way of loading a per-Assembler piece of SKILL code prior to netlisting. I can think of a way of possibly implementing this (so you could add the code into the documents folder of the maestro view and have it auto-loaded) but I've not written such a customisation.

So for now, the best thing is to load this code from your .cdsinit. Loading it in the CIW may not be sufficient if running multiple simulations as netlisting will be done in a background process and won't see what you've loaded in the CIW.

I'm not sure you can draw the conclusion that we don't have RF designers in mind from this. Yes, it's something that frequency-domain RF designers may want to do, but there are a lot of challenges that are caused by allowing arbitrary ideal frequency domain expressions/modelling by virtue of the fact that such frequency domain descriptions do not reliably translate into something causal and stable in the time domain. OK if you were only doing linear simulation in the frequency domain, but even with harmonic balance you need to ensure that any transient assist can handle the frequency domain part (so a complex port impedance is only supported in frequency domain analyses, for example). We have to handle the whole range of design types and not just RF, but there are continual efforts to improve the experience for RF designers within the Virtuoso platform.

Regards,

Andrew