Refactoring non-linear solvers

[AnHeuermann]

Purpose

Make solveNLS more generic, so it can work for any given nonlinear systems, not only for those saved in data->simulationInfo->nonlinearSystemData[sysNumber].

Progress

• KINSOL
• Newton
• Hybrid
• Homotopy
• Mixed
• solveNLS accepting NONLINEAR_SYSTEM_DATA* nonlinsys

Changes

Unified user data structs for all non-linear solvers:

typedef struct NLS_USERDATA {
  DATA *data;
  threadData_t *threadData;

  int sysNumber;                        /* System index, for print messages only */
  NONLINEAR_SYSTEM_DATA* nlsData;       /* Pointer to nonlinear system data */
  ANALYTIC_JACOBIAN* analyticJacobian;  /* Pointer to analytic Jacobian */
} NLS_USERDATA;

Contains pointer to nlsData that contains this user data. Used to hide omc internal data from 3rdParty function calls.
userData will be passed to residual and Jacobian functions.

Use initNlsUserData and freeNlsUserData for memory management.

Simplify NLS functions interface

• int allocateNewtonData(int size, void** voiddata) --> DATA_NEWTON* allocateNewtonData(int size, NLS_USERDATA* userData)
• int nlsKinsolFree(void **solverData) --> void freeNewtonData(DATA_NEWTON* newtonData)
• int solveNLS(DATA *data, threadData_t *threadData, int sysNumber) --> modelica_boolean solveNLS(DATA *data, threadData_t *threadData, NONLINEAR_SYSTEM_DATA* nonlinsys)
• int _omc_newton(int(*f)(int*, double*, double*, void*, int), DATA_NEWTON* solverData, void* userdata); --> int _omc_newton(genericResidualFunc f, DATA_NEWTON* solverData, void* userData);
• void printNonLinearSystemSolvingStatistics(DATA *data, int sysNumber, int logLevel) --> void printNonLinearSystemSolvingStatistics(NONLINEAR_SYSTEM_DATA* nonlinsys, enum LOG_STREAM stream)

Other

• Added more documentation.
• Fixed memory leaks while freeing NLS data.
• Removed unused currentNonlinearSystemIndex from NONLINEAR_SYSTEM_DATA.
• Added TRUE and FALSE to openmodelcia_types.h, next to definition of type modelica_boolean.
• typedef genericResidualFunc function pointer.

[AnHeuermann]

@bernhardbachmann KINSOL and Newton are already refactored.

We can try to merge the current state into a copy of https://github.com/bernhardbachmann/OpenModelica/tree/01-ESDIRK_integrator and see if this is working for your gbode solver.

If that is working I'll finish this for the other non-linear solvers and we can merge this into the master.

[AnHeuermann]

Merging is not straight forward.
I will cherry-pick some commits from bernhardbachmann/01-ESDIRK_integrator that I made a while ago to improve the NLS solver.

[AnHeuermann]

716f148 broke some tests. Now I need to figure out why.

[bernhardbachmann]

The gbode development branch has been renamed to 01-GBODE_development. Please use this for merging your NLS changes

[phannebohm]

LGTM, except some small typos.