Main Content

Accelerating Fixed-Point Models

If the model meets the code generation restrictions, you can use Simulink® acceleration modes with your fixed-point model. The acceleration modes can drastically increase the speed of some fixed-point models. This is especially true for models that execute a very large number of time steps. The time overhead to generate code for a fixed-point model is generally larger than the time overhead to set up a model for simulation. As the number of time steps increases, the relative importance of this overhead decreases.

Note

Rapid Accelerator mode does not support models with bus objects or 33+ bit fixed-point data types as parameters.

Every Simulink model is configured to have a start time and a stop time in the Configuration Parameters dialog box. Simulink simulations are usually configured for non-real-time execution, which means that the Simulink software tries to simulate the behavior from the specified start time to the stop time as quickly as possible. The time it takes to complete a simulation consists of two parts: overhead time and core simulation time, which is spent calculating changes from one time step to the next. For any model, the time it takes to simulate if the stop time is the same as the start time can be regarded as the overhead time. If the stop time is increased, the simulation takes longer. This additional time represents the core simulation time. Using an acceleration mode to simulate a model has an initially larger overhead time that is spent generating and compiling code. For any model, if the simulation stop time is sufficiently close to the start time, then Normal mode simulation is faster than an acceleration mode. But an acceleration mode can eliminate the overhead of code generation for subsequent simulations if structural changes to the model have not occurred.

In Normal mode, the Simulink software runs general code that can handle various situations. In an acceleration mode, code is generated that is tailored to the current usage. For fixed-point use, the tailored code is much leaner than the simulation code and executes much faster. The tailored code allows an acceleration mode to be much faster in the core simulation time. For any model, when the stop time is close to the start time, overhead dominates the overall simulation time. As the stop time is increased, there is a point at which the core simulation time dominates overall simulation time. Normal mode has less overhead compared to an acceleration mode when fresh code generation is necessary. Acceleration modes are faster in the core simulation portion. For any model, there is a stop time for which Normal mode and acceleration mode with fresh code generation have the same overall simulation time. If the stop time is decreased, then Normal mode is faster. If the stop time is increased, then an acceleration mode has an increasing speed advantage. Eventually, the acceleration mode speed advantage is drastic.

Normal mode generally uses more tailored code for floating-point calculations compared to fixed-point calculations. Normal mode is therefore generally much faster for floating-point models than for similar fixed-point models. For acceleration modes, the situation often reverses and fixed point becomes significantly faster than floating point. As noted above, the fixed-point code goes from being general to highly tailored and efficient. Depending on the hardware, the integer-based fixed-point code can gain speed advantages over similar floating-point code. Many processors can do integer calculations much faster than similar floating-point operations. In addition, if the data bus is narrow, there can also be speed advantages to moving around 1-, 2-, or 4-byte integer signals compared to 4- or 8-byte floating-point signals.