Accelerating Model Simulations and Analyses
What Is Acceleration?
Normally, when simulating or analyzing a model in SimBiology®, the model is expressed in MATLAB® code. You can accelerate the simulation by converting the model to compiled C code, which executes faster. Because this compilation step has a small time overhead, acceleration is not recommended for individual simulations of small models. However, for large models, or for repeated simulations during analysis, acceleration can provide a significant speed increase that outweighs the small time overhead.
When to Accelerate
The functionality to accelerate simulations performs optimally under the following conditions:
Running repeated simulations with different initial conditions
Running very long simulations (for example, simulations that take longer than a minute to run)
Prerequisites for Accelerating Simulations and Analyses
To prepare your models for accelerated simulations, install and set up a compiler:
Install a C compiler (if one is not already installed on your system). For a current list of supported compilers, see Supported and Compatible Compilers.
Ensure that any user-defined functions in your model can be used for code generation from MATLAB, so they can convert to compiled C. For more information, see Language, Function, and Object support for C and C++ code generation (MATLAB Coder) or contact MathWorks Technical Support.
On Windows®, if you have not installed another compiler, SimBiology uses the lcc-win64 compiler for model accelerations. If you have installed another supported compiler, it will be selected automatically. For better performance of the acceleration functionality, you may want to install a supported compiler other than lcc-win64, and it will be selected automatically.
Accelerate Simulations Programmatically
sbioaccelerate if you are
accelerating a SimBiology model. For a
SimFunction object and an exported
SimBiology.export.Model), use the
Follow the two-step process for acceleration.
sbioaccelerateto prepare your model for accelerated simulations. Use the same input arguments that you plan to use with
sbiosimulatein the next step. For example:
For a very large model, this step may take a minute or longer to complete.
sbiosimulatewith the same input arguments that you used with
sbioaccelerate. For example:
If you pass in an array of doses to
can simulate the model using any subset of these doses and do not need to run
For illustrated examples, see the following.
SimFunction object is automatically accelerated at the
first function execution. Hence it is not necessary to accelerate the model
before you create the object. However, manually accelerate using the
accelerate method of the object
if you want it accelerated in your deployment applications.
For exported model, see
When to Rerun Acceleration
If you make any modifications to the model, such as changes to reactions or adding events, you need to rerun the acceleration, before running simulations.
However, there are exceptions. You do not need to accelerate again if you are making the changes to:
InitialAmountproperty of species
Capacityproperty of compartments
Valueproperty of parameters
StopTimeproperty of configset
OutputTimes property of SolverOptions
UserDataproperties of any applicable objects
Accelerate Simulations using SimBiology Model Analyzer
You can enable the model acceleration in the SimBiology Model Analyzer app by checking the Prepare the model for accelerated simulation box in the Model step of the program.
Troubleshooting Accelerated Simulations
If you have custom functions, use persistent variables only for those (constant) variables that you do not want to recalculate or reload every function call. The reason is that during the acceleration process, SimBiology converts the model and custom functions to compiled C code. If you try to use a persistent variable to share data across generated (or compiled) C functions, you may have different results. For instance, if you use a persistent variable to count how many times a function is called, each compiled function will have a separate count. Those persistent variables in the corresponding compiled functions will be different from the one used in the MATLAB function that you defined.
If you specify custom functions in SimBiology expressions, you might see the following warning if your code is not compatible with code generation from MATLAB:
The SimBiology Expression and any user-defined functions could not be accelerated. Please check that these expressions and any user-defined functions are supported for code generation as described in the Code Generation from MATLAB documentation.
where Expression is any of the following:
Reaction rate/rule expression
Initial assignment rule expression
Repeated assignment rule expression
Event trigger expression
Event function expression
For more information, see Language, Function, and Object support for C and C++ code generation (MATLAB Coder) or contact MathWorks Technical Support.
SimFunction object |