Code generation for non-linear systems

Index
» Developer
» OpenModelica development
» Code generation for non-linear systems

Page Start Prev 1 Next End

- Subscribe

Code generation for non-linear systems

#1984 Permalink

Code generation for non-linear systems

Feb-10-12 13:33:05

Hi,

I am not sure wheter I am correct, but it appears as OpenModelica generates inccorect code for the solving of non-linear systems.

In the generated C code a function is called from "functionAlgebraics" which calls a macro (start_nonlinear_system(68)) leading to the definition of the variable: double nls_fvec[68].
Later on a solver macro (solve_nonlinear_system) is called with a function pointer to some function which apperas to calculate some residual values. In this functions the results are written to a pointer pointing to nls_fvec. Here seems to be the problem, as some values are written beyond the allocated space (in my case up to index 74). I am not sure wheter I am correct here, as in some models the simulation runs correctly. However, I have a model where this seems to cause a segmentation fault crashing the simulation program.

Im am using OpenModelica (r11077).

Regards,

Michael

Edited by: mburisch - Feb-10-12 13:34:00

mburisch: 19 Posts; Thank you [ 0 ]

#1985 Permalink

Re: Code generation for non-linear systems

Feb-10-12 13:37:58

Could you provide a model that shows this behaviour? I have not seen this in any code myself and need a starting point if I'm going to fix it.
Writing to a variable beyond this limit is pointless since the nonlinear solver will not consider this and you will either get wrong result or a crash.

sjoelund.se: 1700 Posts; Thank you [ 159 ]

#1986 Permalink

Re: Code generation for non-linear systems

Feb-10-12 14:04:43

This only happends in one of our larger model under certain conditions. I will try to make a small model showing this behaviour.

Regards,
Michael

mburisch: 19 Posts; Thank you [ 0 ]

#1987 Permalink

Re: Code generation for non-linear systems

Feb-10-12 14:10:42

Note: I have compiled the model using the gcc option "-fstack-protector-all" to check for buffer overflows and it reported an error (stack smashing) during execution.

mburisch: 19 Posts; Thank you [ 0 ]

#1996 Permalink

Re: Code generation for non-linear systems

Feb-13-12 17:27:50

I was able to create a small model which reproduces the error. In my case the generated function "eqFunction_6" declares a non-linear system with with an array of size 12. In "residualFunc6", however, this array is accessed at index 15. Although for me this does not give a run-time error, it does corrupts the stack.

As uploading does not work for me, I post the model here:

Code:


package Medium

  function h_pt

    input Real p;

    input Real t;

    output Real h;

  algorithm

    h := t;

  end h_pt;

  

  function d_ph

    input Real p;

    input Real h;

    output Real d;

  algorithm

    d := h;

  end d_ph;  

end Medium;



connector Port

  flow Real mf;

  Real p;

  stream Real h;

end Port;





model Pipe

  Port port_a, port_b;

  Real loopPressure;

    

protected

  Real density;

  Real outputPressure;

  

  function func

    input Real inputPressure;

    input Real density;

    output Real outputPressure;

    output Real loopPressure;

  algorithm

    loopPressure := inputPressure;

    outputPressure := inputPressure;

  end func;



equation

  density = Medium.d_ph(port_a.p, port_a.h);

  (outputPressure, loopPressure) = func(port_a.p, density);



  port_a.mf + port_b.mf = 0;

  port_b.p = outputPressure;

  port_a.h = inStream(port_b.h);

  port_b.h = inStream(port_a.h);

end Pipe;



model Split

  Port port_a, port_b, port_c;

equation

  connect(port_a, port_b);

  connect(port_a, port_c);  

end Split;



model Join

  Port port_a, port_b, port_c;

  final parameter Real epsFlow = 1e-15;

equation

  port_a.mf + port_b.mf + port_c.mf = 0;

  port_b.p = min(port_a.p, port_c.p);

  port_a.h = (max(port_b.mf, epsFlow) * inStream(port_b.h) + max(port_c.mf, epsFlow) * inStream(port_c.h)) / (max(port_b.mf, epsFlow) + max(port_c.mf, epsFlow));

  port_b.h = (max(port_a.mf, epsFlow) * inStream(port_a.h) + max(port_c.mf, epsFlow) * inStream(port_c.h)) / (max(port_a.mf, epsFlow) + max(port_c.mf, epsFlow));

  port_c.h = (max(port_a.mf, epsFlow) * inStream(port_a.h) + max(port_b.mf, epsFlow) * inStream(port_b.h)) / (max(port_a.mf, epsFlow) + max(port_b.mf, epsFlow));

end Join;  



model Boundary

  Port port;

  Real t;

  Real mf;

  Real p;

equation

  port.mf = mf;

  port.p = p; 

  port.h = Medium.h_pt(p, t);

end Boundary;



model Test

  Split[numberOfSplits - 1] split;

  Join[numberOfSplits - 1] join;                                                                 

  Pipe[numberOfSplits] pipe;

    

  Boundary source;

  Boundary sink;

    

  parameter Integer numberOfSplits = 2;    

  parameter Integer numberOfLoops = 10;

    

equation  

  source.mf = -10;

  source.p = 100000;

  source.t = 300;

  sink.t = 300;

  

  // Grid:

  //  split

  // --+--+---

  //   |  |  | pipe

  // --+--+---

  //  join

  

  connect(source.port, split[1].port_a);

  

  for i in 1:numberOfSplits - 1 loop

    connect(split[i].port_c, pipe[i].port_a);

    connect(pipe[i].port_b, join[i].port_c);



    split[i].port_c.mf = -source.port.mf / numberOfSplits;    

  end for;

  

  

  for i in 1:numberOfSplits - 2 loop

    connect(split[i].port_b, split[i + 1].port_a);

    connect(join[i + 1].port_b, join[i].port_a);

  end for;  

    

  connect(split[numberOfSplits - 1].port_b, pipe[numberOfSplits].port_a);

  connect(pipe[numberOfSplits].port_b, join[numberOfSplits - 1].port_a);

  

  connect(join[1].port_b, sink.port);

end Test;