Add support for ESP32

[alexthomazo]

Hello,

After some debugging this afternoon, I managed to have a working version of the library working on an ESP32.

It was not working because the ESP32 is way too faster that the Arduino UNO for setting a pin value so the clock was not within the specs of the TM1637. The datasheet specify a minimum clock pulse of 400ns but the ESP32 done that in ~50ns.

I add a delay after each write in order to comply with the TM1637 specs. The delay is only added for the ESP32 in order to not slow down the write on the Arduino.

related to #3

[RobTillaart]

Thanks for this addition I will do a code check

[RobTillaart]

It looks good, overal just one remark on code level to make other timing adjustments possible

However thinking about it, you need a delay of 400 ns and you use 1000 ns.

Is there a way to write the delayNanoseconds() - (drawback) it is CPU clock dependent...

something like

void delayNanoseconds(uint32_t  ns)
{
  ns >>= 5;   // every loop takes about 32 nanoseconds @ 240 MHz
  while (ns--)
  {
    __asm__ __volatile__ ("nop");  // 1 nop = 1000 /240 ~~ 4ns;
  }
}

One might need initially get the ESP32 board frequency - Freq = getCpuFrequencyMhz();

[RobTillaart]

small test

uint32_t start, duration;

void setup()
{
  Serial.begin(115200);

  for (uint16_t n = 100; n < 3100; n += 100)
  {
    start = micros();
    nanoDelay240(n);
    duration = micros() - start;
    Serial.print(n);
    Serial.print("\t");
    Serial.println(duration);
    delay(10);
  }
}

void loop()
{
}

void nanoDelay240(uint16_t n)    // 240 ==> at 240 MHz
{
  volatile uint16_t i;
  for (i = 0; i < n; ++i);
}

one iteration seems to be about 91 nanoseconds so to get a 400ns delay ==> n = 4 or 5 ?

Can you test?

[alexthomazo]

Hello Rob,

Thanks for all your suggestions, yes I will test it to see if it's possible to reduce the delay furthermore.

[alexthomazo]

After some tests, it's working with nanoDelay with only 2 loops. It's almost working with 1 loop but I see some weird issues on the display with it.

Here the timing diagram for ESP32 with 2 loops.

Same timings with the Arduino UNO:

Also, it's working as well when removing the delayMicroseconds when getting acknowledge (the diagrams are without), I don't really know if it's really needed as the ACK is not used afterward. So I don't remove it on my pull request but it's definitively working on Arduino and ESP32 without it.

[RobTillaart]

Wow, looks pretty clean signal to me.
How much time did you gain?
How much faster is the ESP compared to the UNO now?

[RobTillaart]

did a test here and found a version that is 40% faster so in theory more accurate nanoDelay
the -- instruction in combination with the automatic compare to zero is less machine instructions
as the ++ and the comparison with n.

Think you should be able to squeeze some nanoseconds of, - which add up to microseconds

void nanoDelay(uint16_t n)
{
  volatile uint16_t i = n;
  while (i--);
}

[RobTillaart]

Think the PR looks good, will merge after you confirmed (or not) if the new nanoDelay() works better.

[alexthomazo]

I tried with the new nanoDelay, succeed to gain a little more (113us instead of 133us), it's not working if I go lower.

[RobTillaart]

I tried with the new nanoDelay, succeed to gain a little more (113us instead of 133us), it's not working if I go lower.

had a thought,
when the CLOCK is set LOW, it is followed by a nanoDelay() to be sure the period is long enough.
However that LOW is followed by writing the DATA, with its own nanoDelay().
As the data is actually read after the CLOCK goes HIGH, the nanoDelay() should not be needed with the CLOCK LOW command.

(think that would keep al signals just limits but could save another 25-30% ?

That's my final squeeze :)

[alexthomazo]

Yes, I've got the same idea, already tried it but unfortunately it's not working and the datasheet is not very explicit for theses timings.

[Niv1900]

Thanks @alexthomazo and @RobTillaart!

When I tried to use level shifter to solve #3 it didn't work and I blame the fault on the level shifter not working properly.

[RobTillaart]

Merged into master branch,
I will update version numbers etc tonight and make a new release

Thanks @alexthomazo for code, investigating and testing!

[RobTillaart]

0.2.0 version released

[alexthomazo]

Thanks 🙂