pinMode is much slower in 2.0.3 (and 2.0.4)

[jenscski]

Board

All boards

Device Description

Tested with featheresp32

Hardware Configuration

DS18B20 connected to A0

Version

v2.0.3

IDE Name

not applicable

Operating System

not applicable

Flash frequency

not applicable

PSRAM enabled

no

Upload speed

not applicable

Description

In version 2.0.3 the GPIO handing was rewritten using IDF instead of the old implementation.

This caused the time pinMode takes to finish from ~4us to ~16us.

In ESPHome the one wire implementation uses the normal Arduino calls to manipulate GPIO, and this change is probably the cause of One Wire not working in ESPHome when using Arduino 2.0.3.

esphome/issues#3401

Is it possible to speed up pinMode?

After studying the ESPHome and Arduino code, I see that when ESPHome is compiled using ESP-IDF, it uses a different approach than Arduino for setting pin mode.

ESPHome https://github.com/esphome/esphome/blob/ed26c57d990c77cf9eed9bd54da2681f8f7881c9/esphome/components/esp32/gpio_idf.cpp#L83-L106

Arduino

arduino-esp32/cores/esp32/esp32-hal-gpio.c

Lines 92 to 133 in 178aee8

	extern void ARDUINO_ISR_ATTR __pinMode(uint8_t pin, uint8_t mode)
	{
	#ifdef RGB_BUILTIN
	if (pin == RGB_BUILTIN){
	__pinMode(RGB_BUILTIN-SOC_GPIO_PIN_COUNT, mode);
	return;
	}
	#endif
	if (!GPIO_IS_VALID_GPIO(pin)) {
	log_e("Invalid pin selected");
	return;
	}
	gpio_hal_context_t gpiohal;
	gpiohal.dev = GPIO_LL_GET_HW(GPIO_PORT_0);
	gpio_config_t conf = {
	.pin_bit_mask = (1ULL<<pin), /*!< GPIO pin: set with bit mask, each bit maps to a GPIO */
	.mode = GPIO_MODE_DISABLE, /*!< GPIO mode: set input/output mode */
	.pull_up_en = GPIO_PULLUP_DISABLE, /*!< GPIO pull-up */
	.pull_down_en = GPIO_PULLDOWN_DISABLE, /*!< GPIO pull-down */
	.intr_type = gpiohal.dev->pin[pin].int_type /*!< GPIO interrupt type - previously set */
	};
	if (mode < 0x20) {//io
	conf.mode = mode & (INPUT | OUTPUT);
	if (mode & OPEN_DRAIN) {
	conf.mode |= GPIO_MODE_DEF_OD;
	}
	if (mode & PULLUP) {
	conf.pull_up_en = GPIO_PULLUP_ENABLE;
	}
	if (mode & PULLDOWN) {
	conf.pull_down_en = GPIO_PULLDOWN_ENABLE;
	}
	}
	if(gpio_config(&conf) != ESP_OK)
	{
	log_e("GPIO config failed");
	return;
	}
	}

Sketch

This is the sketch I ran using 2.0.2 and 2.0.3. pinMode was the only one with different timing


#include <Arduino.h>

void test()
{
  auto d1 = micros();
  pinMode(A0, OUTPUT);
  auto d2 = micros();
  digitalWrite(A0, LOW);
  auto d3 = micros();
  pinMode(A0, INPUT_PULLUP);
  auto d4 = micros();
  digitalRead(A0);
  auto d5 = micros();

  Serial.println(d2 - d1);
  Serial.println(d3 - d2);
  Serial.println(d4 - d3);
  Serial.println(d5 - d4);
  Serial.println();
}

void setup()
{
  Serial.begin(9600);
  Serial.println();
  Serial.println("Hello");

  test();
  test();
  test();
  test();
}

void loop()
{
  // put your main code here, to run repeatedly:
}

Debug Message

not applicable

Other Steps to Reproduce

No response

I have checked existing issues, online documentation and the Troubleshooting Guide

• I confirm I have checked existing issues, online documentation and Troubleshooting guide.

[Jason2866]

Speed improvement is always nice to have. Dropping the manipulating of registers and using IDF is imho a good decision. Otoh there is a bit more overhead for some functions, which can decrease speed. The better solution imho is to adopt the code (to the new general conditions) to be (more) timing tolerant, as to hope that a speed increase will happen
btw this OneWire lib does work: https://github.com/PaulStoffregen/OneWire

[TD-er]

Ah this is a plausible explanation why the Dallas plugin and several others like DHTxx no longer seem to work on ESP32 in ESPEasy.

[TD-er]

btw this OneWire lib does work: https://github.com/PaulStoffregen/OneWire

That looks like an easy approach to speed up switching the mode by simply keeping track of the register and bitmask for a given pin.

I will at least implement this for my One Wire implementation and probably also for other timing critical plugins like those DHTxx units.

[TD-er]

OK, setting pinMode to INPUT_PULLUP is even worse.
I was working on the implementation of the DHT22 support for ESPEasy to improve timing stability and apparently setting the pin mode to input + pull-up takes a whoppin' 228 usec.

The white line is the pin I set to INPUT_PULLUP, which is then pulled down by the DHT22 sensor.
The brown line is an extra pin (GPIO-27) I toggle using direct GPIO access:

    DIRECT_WRITE_HIGH(reg, mask_27);  // GPIO-27
    pinMode(DHT_pin, INPUT_PULLUP);
    DIRECT_WRITE_LOW(reg, mask_27);  // GPIO-27

This is it with direct access setting the pin mode to INPUT (thus not using the pull-up as I don't know how to do it via direct access)

    DIRECT_WRITE_HIGH(reg, mask_27);
    DIRECT_MODE_INPUT(reg, mask);
//    pinMode(DHT_pin, INPUT_PULLUP);
    DIRECT_WRITE_LOW(reg, mask_27);

This takes 4 usec to set the pin mode.

[P-R-O-C-H-Y]

@TD-er, GPIO has been changed to use the IDF API, rather than directly writing to registers in order to have support for new chips. I'm closing this and adding label "won't fix", as there is nothing to do about this issue.