Issues with large parameter arrays

[Adrian Pop]

See forum post:
​https://www.openmodelica.org/forum/default-topic/2734-ram-problem-with-very-large-arrays-used-as-parameters-in-a-model

I have observed this myself as well for the old front-end.
With -d=newInst is much better for the front-end but the
back-end uses a lot of memory and time to generate the code.

model test 
  public 
    parameter Integer Nmax = 100; 
  protected 
    parameter Real x[Nmax](each fixed=false) "abscissa"; 
    parameter Real y[Nmax](each fixed=false) "y=f(x)"; 
  // any variable for my calculation 
initial algorithm 
  x := linspace(0.0,14.5,Nmax); 
  // In this example f (x) = -1 and there would be no need for an array, 
  // obviously I fill y with a more complicated algorithm, not a simple formula... 
  y := fill(-1.0,Nmax); 
equation 
  // any equation for my calculation 
end test; 

[Adrian Pop]

For example with Nmax=3000 it takes 74 seconds and 5GB with the master and flags:
-d=newInst,initialization,execstat.

It seems most of the time (70s) is spent into: simCode: created initialization part.

I remember I encountered this before, but with the current front-end and mostly because constant evaluation and simplify took forever and consumed all memory (basically trying to find over and over again if the array is constant).

[Francesco Casella]

I'm not 100% sure, but isn't this the same as #1451, or at least related?

[Adrian Pop]

@casella: not sure if is the same as #1451, but it does seem so.

Running valgrind to profile now with Nmax=2000.
For execstat I get:

Notification: Performance of Backend phase and start with SimCode phase: time 0.01732/91.98, allocations: 53.55 kB / 123.6 MB, free: 2.516 MB / 31.91 MB
Notification: Performance of simCode: created initialization part: time 2908/3000, allocations: 17 GB / 17.12 GB, free: 351.9 MB / 1.821 GB

[Adrian Pop]

While making the incidence matrix we seem to generate a new one with absolute values for no good reason, do the people working on the backend have any idea why?
This is the actual problem for this ticket.

I disabled the generation via the PR: ​https://github.com/OpenModelica/OpenModelica/pull/295
and about 99 models are failing.

I guess we could just do intAbs on the values in the incidence matrix when is needed (when we read them) and not generate a new one with absolute values all the time.