improve port expander handling #282
Conversation
The output register status buffer was initialized fix to {0xff,0xff}
although the output registers were set to {0x00,0x00}.
That buffer is used by Port_Write() to determine the values of the
bits for the other pins. So those get set to 1 (high).
Since the peripheral power on the Espuino 4-layer mini PCB is switched
on by setting a pin on the PCA9555 low, that power is first turned on
(by Port_Init()), then turned off (by the Port_Write() call in
AudioPlayer_Init()) and then turned on again (by Power_PeripheralOn()).
That caused a 12.5 millisecond disruption of the external 3.3V which
some SD-cards don't like.
The PORT_ExpanderISR() is called very often although no interrupt was triggered by the PCA9555. The interrupt line is held low by the PCA9555 until the input registers are read. So if the interrupt line is not low, there is no need to read the registers.
The interrupt line is active low so switch from FALLING to ONLOW. To then prevent continuous calling of the ISR, it needs to be disabled. Port_ExpanderHandler() re-enables it after handling things. It also takes care to enable it in the first place. While at it, use digitalPinToInterrupt() to convert the pin to the interrupt number as recommended by the Arduino reference manual (https://www.arduino.cc/reference/en/language/functions/external-interrupts/attachinterrupt/).
Port_ExpanderHandler() was using 4 transfers to read 2 registers from the PCA9555: START-DEV_ADDR-REG_ADDR(0)-STOP START-DEV_ADDR-REG_VAL(@0)-STOP START-DEV_ADDR-REG_ADDR(1)-STOP START-DEV_ADDR-REG_VAL(@1)-STOP This takes at least 766.1 microseconds on a 100kHz bus. Simplifying this by using repeated start and reading both registers at once to: START-DEV_ADDR-REG_ADDR(0)-START-DEV_ADDR-REG_VAL(@0)-REG_VAL(@1)-STOP Now it takes at least 466.7 microseconds i.e. almost 300 microseconds less.
Instead of checking if the whole register is unchanged, check the bits independently so a toggling input doesn't prevent stable inputs from being accepted.
|
@r-schmidt Thank you very much for your contribution! Please do the PR in the DEV branch first if possible, we can still test the extensions intensively there. As discussed we use uint_8 not unsigned int in loop: Where exactly did you encounter problems with the current implementation? Perhaps you can briefly describe it in the forum so that everyone can get an idea of how you have done speedup & debounce |
|
Seems like I need to make a new PR to change the base branch on my side. |
|
@r-schmidt I have tested your code & there does seem to be an optimised & faster access to the PE (less CPU time is used). Only PE port 2 I could not test now. It's worth transferring the PR against the DEV branch, thanks for your effort! |
|
Follow up #283 |
This is a collection of changes I made for the port expander: