Multi-threaded DSP Buffering
Coordinating a long running DSP process to run on two DSP chips while performing other tasks on the CPU does require some planning. This is an example of that planning for the freeze-stretch driver. (In this case "long" means that it takes longer than a single buffer's amount of time)
| Step | Buffers A | Buffers B |
|---|---|---|
| 1 | Fill a complete input window A | Complete playback of output window B |
| 2 | Enqueue map DSP input buffer A for writing | |
| Copy input window A to DSP input buffer A | ||
| Enqueue unmap DSP input buffer A | ||
| Enqueue DSP task A with event A | ||
| 3 | DSP event B wait for completion | |
| Enqueue map DSP output buffers B for reading | ||
| Copy DSP output buffers B to output window A | ||
| Enqueue unmap DSP input buffers B | ||
| 4 | Begin playback of output window A | Begin recording data into input window B |
| 5 | Complete playback of output window A | Fill a complete input window B |
| 6 | Enqueue map DSP input buffer B for writing | |
| Copy input window B to DSP input buffer B | ||
| Enqueue unmap DSP input buffer B | ||
| Enqueue DSP task B with event B | ||
| 7 | DSP event A wait for completion | |
| Enqueue map DSP output buffers A for reading | ||
| Copy DSP output buffers A to output window B | ||
| Enqueue unmap DSP input buffers A | ||
| 8 | Begin recording data into input window A | Begin playback of output window B |