I noticed that the loop over 65535 possible denominators took a long time,
causing up to 100ms wait for a sound sample to start playing!
This algorithm, adapted from an algorithm shown in Python's fractions.py,
is guaranteed to find the best denominator in a small number of steps
(I think log2-many steps but I'm not sure). In practice, it means the time
between samples playing is just 10ms, and some of that is recreating the
sine wave sample in Python each time.
It often finds the same solution as the old code, but sometimes it finds
one a bit better since it compares the ratios using float point instead
of integer arithmetic.
There is a race between when we run background tasks and when we
sleep. If an interrupt happens between the two, then we may delay
executing the background task. On some ports we checked this for
TinyUSB already. On iMX RT, we didn't which caused USB issues.
This PR makes it more generic for all background tasks including
USB.
Fixes#5086 and maybe others.
RP2040 and SAMD51:
- Detect when DMA has finished, and stop DMA audio explicitly.
- Do not accidentally reuse `first_buffer` supplied by WaveFile or RawSample. Always realloc `first_buffer` and `second_buffer`
RP2040:
- When audio playing is stopped, write a final zero to the output register. This prevents residual PWM tones.
- Handle buffer size for 8-bit samples properly for 16-bit output.
- Fail on some edge cases (which may not be possible at the moment).
Passing in a PWMOut still works but is deprecated. It will be
removed in CircuitPython 8.0.0
This also switches STM32 timer indices and channel indices to
0-based in our pin data rather than `- 1` everywhere. The latter is
more bug prone.
Most of the way for #3264
Tested on Metro M0, Metro M4, Feather S2, Feather nRF52840, Feather
STM32F4 and Arduino RP2040.
For RP2040 boards, we now change the default flash size based on
the configured flash. We will still try to read the size from the
flash first.
Fixes#4874
