T-DeckLoader/lib/AceButton/examples/AutoBenchmark

AutoBenchmark

This sketch measures the amount of time consumed by the AceButton::check() method when processing various button events. It uses a special ProfilingButtonConfig object that allows the program to inject button events into the library. The profiling numbers come from activating the TimingStats object that has been instrumented into the AceButton::check() method.

Note that ProfilingButtonConfig class generates synthetic button events, bypassing the actual digitalRead() function. The digitalRead() function on an Arduino AVR platform (UNO, Nano, etc) is known to be slow which will add to the timing values in actual usage. The digitalWriteFast library might be an alternative if speed is critical.

Benchmark Results

In all of the tests, the min time for the "idle" case is larger than any of the other button events. This is because when a button event occurs, the AceButton::checkDebounced() method returns immediately until the debouncing time is over which brings down the minimum time. No debouncing is done in the "idle" case so the minimum code path takes more CPU cycles.

All times are in microseconds. The "samples" column is the number of TimingStats::update() calls that were made.

Arduino Nano

• 16MHz ATmega328P

------------------------+-------------+---------+
button event            | min/avg/max | samples |
------------------------+-------------+---------+
idle                    |  12/ 13/ 20 | 1934    |
press/release           |   8/ 14/ 20 | 1925    |
click                   |   8/ 14/ 24 | 1925    |
double click            |   8/ 13/ 24 | 1925    |
long press/repeat press |   8/ 15/ 24 | 1927    |
------------------------+-------------+---------+

Arduino Pro Micro

• 16MHz ATmega32U4

------------------------+-------------+---------+
button event            | min/avg/max | samples |
------------------------+-------------+---------+
idle                    |  12/ 13/ 24 | 1935    |
press/release           |   8/ 14/ 24 | 1928    |
click                   |   8/ 13/ 24 | 1928    |
double click            |   8/ 13/ 24 | 1926    |
long press/repeat press |   8/ 15/ 28 | 1928    |
------------------------+-------------+---------+

Teensy 3.2

• 96 MHz ARM Cortex-M4

------------------------+-------------+---------+
button event            | min/avg/max | samples |
------------------------+-------------+---------+
idle                    |   3/  3/  5 | 1985    |
press/release           |   1/  3/  6 | 1983    |
click                   |   1/  3/  6 | 1984    |
double click            |   1/  3/  6 | 1984    |
long press/repeat press |   1/  3/  6 | 1983    |
------------------------+-------------+---------+

NodeMCU 1.0 clone

• 80MHz ESP8266

------------------------+-------------+---------+
button event            | min/avg/max | samples |
------------------------+-------------+---------+
idle                    |   7/  8/ 24 | 1922    |
press/release           |   6/  8/ 53 | 1919    |
click                   |   6/  8/ 50 | 1920    |
double click            |   6/  8/ 67 | 1910    |
long press/repeat press |   6/  9/ 60 | 1894    |
------------------------+-------------+---------+

The large max times for "double click" and "long press" seem to be reproducible. I have not researched this but my speculation is that the system WiFi code interrupts the AceButton::check() method right when the "double click" and "long press" samples are taken, causing the extra latency.

ESP32-01 Dev Board

• 240 MHz Tensilica LX6

------------------------+-------------+---------+
button event            | min/avg/max | samples |
------------------------+-------------+---------+
idle                    |   3/  3/  3 | 2002    |
press/release           |   2/  2/  8 | 2002    |
click                   |   2/  2/  7 | 2002    |
double click            |   2/  2/  4 | 2002    |
long press/repeat press |   2/  2/  4 | 2002    |
------------------------+-------------+---------+