Send the newly written half of the buffer from the PRU to the host

[Scrashdown]

This would be cleaner way which would avoid total desync.

[Scrashdown]

To reduce the number of cycles spent with SBBO, we can output this value to the PRU local memory which will then be read by the host. This is because the PRU local memory will probably never get saturated by stores while the external memory (the one we write the data to) may be. I will have to measure the time SBBO instructions take with the oscilloscope.

[Scrashdown]

This feature has been partially implemented by using 2 interrupt signals from the PRU (one for each buffer half) and waiting alternatively for each of them on the host side. This ensures the host will not get desynchronized in case it misses some samples. However, using this method, missed buffer halves cannot be detected.

A cleaner solution which would allow for better detection of skipped buffer halves would be to keep track of a counter of the number of buffer halves written to the host on the PRU, and send the counter to the host along with each buffer half. However, with 6 channels and a microphone sample rate of about 1MHz, the timing constraints are very tight.

[dinuxbg]

For fully deterministic operation SBBO should be avoided. You may check the BeagleLogic design where one PRU core handles I/O, passes data using PRU Register Scratch area, and another PRU core communicates with the host RAM. BeagleLogic reaches the impressive 100 MSPS. I've implemented that idea here: https://github.com/dinuxbg/beaglemic/tree/master/pru