I did some performance testing.

The Pro Micro runs an Atmel 32u4 at 8MHz.
If the code does nothing else,
I was able to poll all pins around 13 times every 1ms,
or 13kHz.


Sample Rate
-----------

The number of samples taken since the last HID report
is sent as the second Y axis.
If you use the 
[included gamepad tester](gamepad.html),
it will show the approximate number of samples as an integer,
for Y axis 2.

This is approximate,
because the browser encodes the value as a real number between -1 and 1.
We convert it back, but may lose a little precision.
It's close enough for me,
hopefully it's close enough for you.


Debouncing
----------

Using `millis()` time to debounce the switch
roughly halved my sample frequency.
So instead, I do some preprocessor arithmetic
to calculate how many samples to take after an edge,
in order to debounce switches.

The drum controller was a partcular pain:
in addition to the switch bouncing,
the stick was bouncing on the rubber pad.
I settled on a 40ms silence window as feeling pretty good.
You can adjust this if you want to.