The AI Engine performance is estimated on each batch processing run by reading the AI Engine profiling counter. This counter value \(C\) at the end of a batch run equals the number of cycles for MUSIC to process an entire batch of \(K\) snapshots. The average sweep time is then given as \(T=(C-B/8)/(K-1)\) where \(B\) is the number of bytes produced by the AI Engine for each snapshot (i.e., 256 cfloat
bins + 32 cfloat
tags = 2304B). Examples of some demo waveforms are given in the following table.
Test Case | Average Sweep Time (ns) | Note |
---|---|---|
Demo 3 (64 snapshots) | 993 | 0 sources |
Demo 4 (40 snapshots) | 972 | 1 source |
Demo 5 (40 snapshots) | 969 | 2 sources |
Demo 6 (40 shapshots) | 960 | 3 sources |