SDAA248 October   2026 AM2752-Q1 , AM2754-Q1 , AM62D-Q1

 

  1.   1
  2.   Abstract
  3.   Trademarks
  4. 1Profiling Existing Algorithms
    1. 1.1 Measuring Cycles With the CPU Cycle Counter
  5. 2Compiler Optimizations
    1. 2.1 Verifying Compiler Optimization with Generated Assembly
  6. 3Double Check Memory Cache Settings
  7. 4Use Existing Optimized Libraries
  8. 5Optimizing Custom Kernels/Modules
  9. 6Example
    1. 6.1 Example System Results Summary

Example

Using the steps outlined earlier, we can look at the provided example algorithm (see example code provided on github) and go through the process. We will use a simple example algorithm, which utilizes a few kernels.

The example algorithm is a simple audio algorithm. The following are the parameters of the system:

  1. 8 channels of audio (48kHz), 16-bit samples (Q15), interleaved data
  2. Block size: 128 samples per channel (1,024 samples total)
  3. Input and Output samples are to be signed 16-bit samples (Q15)

In this imaginary system example, there will be a simple algorithm used, which will do the following

  1. Convert the 16-bit samples to a float
  2. 3 stage cascade biquad low pass filter
  3. Global gain of 0.75 is applied
  4. Convert back to a 16-bit sample for use with DAC/amplifiers

This example system will start with the most basic "Natural C" code, which does not have any optimized code for a DSP, and we will go through each of the steps outlined above and see what the performance lift is on a C7000, using the AM275 EVM as the test platform.