| Description |
This paper walks you through the process of profiling code running on a Texas Instruments DM6437 DSP. To demonstrate the process we profiled many common DSP algorithms including Auto-Correlation, the Dot Product, the FFT, an FIR filter, an IIR Filter, a Matrix Multiply, and an LMS Filter. These algorithms were profiled under different conditions. We examined the effects of data type, arithmetic, compiler optimization levels, and the use of DSP intrinsics. For each DSP algorithm profiled, we compared automatically generated C code to TI hand-optimized assembly. The bulk of this paper uses a profiling technique integrated with MATLAB, Simulink, the Real-Time Workshop, and Code Composer Studio by Texas Instruments (TI). This technique uses TI STS Objects strategically placed into the code serving as time markers. We also compare and contrast this technique with more traditional profiling approaches which are independent of any 3rd party software or IDE. |
