Build an example application using CMSIS-DSP

The CMSIS-DSP software library is an open-source suite of common compute processing functions optimized for Arm Cortex-A and Cortex-M processors.

The source code is available in this GitHub repository .

The library is easiest to use with Development tool IDEs such as Keil MDK and Arm Development Studio. The library is installed as a CMSIS-Pack . A number of example projects are also provided.

Keil MDK will be used for this learning path.

Build example with Keil MDK

Install and launch Keil MDK .

Open the Pack Installer, select ARM as the device, and locate DSP_Lib DotProduct example from the Examples tab.

Click Copy, and save the project to an appropriate location on your host.

The project contains build targets for different Cortex-M processors.

Open the Project > Batch Setup pane from the menu bar, and select all targets. Click Rebuild to build all.

Run example on Cortex-M3 FVP

From the Select Target pulldown, highlight the ARMCM3 target (build for Cortex-M3).

The project is already configured for the supplied Fixed Virtual Platforms (FVPs). See the Options > Debug pane.

Start a debug session (Ctrl+F5) to load the image to the Cortex-M3 FVP, and run to main.

Note the number of instructions executed to this point, as shown in the FVP display (184 in the example below):

If not already set, place a breakpoint at the while(1); statement at the end of the main function.

Click Run (or press F5) to run the example to this breakpoint, and stop.

Note the number of instructions executed, and subtract the initialization cycle count.

Terminate the debug session.

Run example on Cortex-M55 FVP

From the Select Target pulldown, highlight the ARMCM55_FP_MVE target (build for Cortex-M55). This build makes use of the Helium vectorizing extensions that Cortex-M55 supports.

Repeat the steps above to run on the Cortex-M55 FVP, and note the number of instructions executed, which should be a dramatic improvement over Cortex-M3.

Understand the example

To make any function from the CMSIS-DSP library available to your code, simply include the arm_math.h header file in any relevant source:

    

        
        
            #include "arm_math.h"
        
    

This file resides in the Include folder of the library installation, and so you should add this to the Include path (-I) for your compiler.

For a Keil MDK project, this is done by selecting CMSIS > DSP in the Manage Run-Time Environment pane.

The example project makes use of two functions from the library:

• arm_mult_f32()
• arm_add_f32()

The appropriate implementation is selected at build time, based on the processor. Hence the Helium optimized implementations can greatly accelerate the algorithm.