Re: Compilation question

Muranyi, Arpad wrote:

>Thank you to all of you who responded to my question
>even though it is debatable whether I picked the right
>forum to ask it...  I did hesitate when I wrote it, but
>decided to do it anyways, because I see compile time
>and runtime mentioned in the LRM quite a bit.
>
>Regarding the sweep simulations, I feel I need to clarify
>what I meant.  I wasn't referring to DC sweeps.  I had
>transient (time domain) or frequency domain simulations
>in mind, which is being rerun multiple times while changing
>parameter values for each iteration.  For example, a
>termination resistor would be simulated with values of
>40 to 60 Ohms at 1 Ohm increments, a total of 21 simulation
>runs.
>
>Now, if this termination resistor is part of a more complicated
>behavioral model of a driver, which is inside a huge library
>with many such modules (to make things worse), it would be
>important for me to know whether this driver model and
>consequently the library file will be recompiled or not
>(meaning converted to a DLL or some sort of an executable
>binary) before each such simulation begins.
>
>Take this further one step, I wonder what happens if this
>"resistor" is described as a non-linear I-V curve, which
>is passed in as an array parameter to the behavioral buffer
>model, in which the size of the array is not predetermined
>inside the module, i.e. it is resizable.
>
>Contrast this question to the case when I am making changes
>to the equations in my behavioral model with my text editor.
>It goes without any doubt that it will be recompiled before
>the simulation can start.
>
>Does a resizable array result in such recompilations when
>its size is changing due to passing a different array length
>into it when the module is instantiated?  (This question may
>apply to multiple runs as described above, as well as to a
>single run having multiple instances of the same module with
>different array sizes passed into it).
>
>The reason I wonder is because array sizes can't be changed
>during runtime according to the LRM.  It must be known at
>compile time.  Does this imply that changing the array 
>parameter size requires a re-compilation?
>  
>
No. The code (C/C++) generated for a library component can use varibles 
that will be set at elaboration, i.e. constant items in the Verilog-A 
are not necessarily constant in the C/C++.

This is a fairly standard trade-off, the more flexible the compiled code 
is the less efficiently it will run. If you want the fastest simulation 
you may want to recompile on a per-instance basis post-elaboration 
(making everything possible a run-time constant), but that slows down 
your simulator start-up considerably, expands the memory footprint and 
is probably deterimental to i-cache usage.

As before: this is mostly an implementation issue, not a language 
definition issue.

Kev.

>Thanks,
>
>Arpad
>============================================================
>
>
>  
>