[solved : 200ns] pin interrupt latency = 2.6us : too long !

[trimarco232]

I made a simple test :
// first, in the setup, pwm and interrupt for pin 10
analogWriteFreq(10000); analogWriteRange(256); // can it be different for each (pair of) pin ?
attachInterrupt(digitalPinToInterrupt(10), isr_test_interrupt_rising, RISING) ;
analogWrite(10, 64);
// than the isr, outside the loop
void isr_test_interrupt_rising() {
static int fall_rise;
if (fall_rise) {
digitalWrite(11, HIGH);
digitalWrite(12, HIGH);
fall_rise = 0 ;
}
else {
digitalWrite(11, LOW);
digitalWrite(12, LOW);
fall_rise = 1 ;
}
}

it works, but my 10€ logic analyser show me :
delay between pin 10 and pin 11 : 3.2us
delay between pin 11 and pin 12 : 0.6us ; so the time for a digitalWrite is 0.6us ...
interrupt latency : 3.2 - 0.6 = 2.6us = 346 cycles @ 133MHz
it's not so ugly than an ESP32 (4us) , but far too much for an ARM based chip
any solution , and how to :

• use a more efficient ISR ?
• change the ISR's priority ?
• put it in RAM ?
  thanks a lot !

[Andy2No]

@trimarco232 You could try the gpio_... functions, instead of relying on digitalWrite(). That might be a touch faster:

https://raspberrypi.github.io/pico-sdk-doxygen/group__hardware__gpio.html

static void | gpio_set_mask (uint32_t mask)
-- | --
  | Drive high every GPIO appearing in mask.  
 
static void | gpio_clr_mask (uint32_t mask)
  | Drive low every GPIO appearing in mask.  
 
static void | gpio_xor_mask (uint32_t mask)
  | Toggle every GPIO appearing in mask.  
 
static void | gpio_put_masked (uint32_t mask, uint32_t value)
  | Drive GPIO high/low depending on parameters.  

I've found gpio_put_masked() to be the most generally useful one, but one of the simpler ones might be faster still - I've never measured it.

You can try putting it in RAM by using __not_in_flash_func in a function definition, apparently. See: #421

[earlephilhower]

The XIP cache is non-deterministic, so definitely put the ISR code in RAM
void __not_in_flash_func)(my_isr)()...

The Arduino ISR API needs to do some processing before it calls your function, so that adds overhead. If you use the raw SDK functions and none of the Arduino GPIO ISR API you should be safe to skip all that (but note you only get 1 GPIO IRQ callback, not one per pin this way).

[trimarco232]

well, thanks @earlephilhower
I tried this example from the SDK : https://forums.raspberrypi.com/viewtopic.php?t=306132
just copied the code, and it works, I got the 200ns latency . It's a normal value ; imho an important avantage of pi pico vs esp32 ...
but I got unsolved compiling errors while trying to custom the isr with __not_in_flash_func ...
.
@Andy2No : thanks, yes , I got very fast io using , for example, gpio_put(11, 0); or gpio_put(11, 1);
of course, I will use the masks, when needed, in order to drive more gpios at the same time

[Andy2No]

I got unsolved compiling errors while trying to custom the isr with __not_in_flash_func ... .

Maybe just a typo? The use of brackets is a bit unfamiliar. A misplaced bracket could produce compiler errors that aren't all that clear.

You could try copying the line from #421 and pasting in your function name and parameters list. Failing that, paste in your code and the error message in a reply.

I'm not clear what adding the "static" keyword does in that context - there's an example with and one without.

[trimarco232]

// not that simple , I have this setting :
irq_set_exclusive_handler(IO_IRQ_BANK0, gpio_toggle);
// and the isr itself :
static void gpio_toggle ()
{
gpio_xor_mask(1 << 11);
gpio_acknowledge_irq(10, IO_IRQ_BANK0);
}
// this works

// when i try to put the isr in ram ...
static void __not_in_flash_func(gpio_toggle) ()
{
gpio_xor_mask(1 << 11);
gpio_acknowledge_irq(10, IO_IRQ_BANK0);
}
// ... I get the error
tesst_pi_pico_isr:34:43: error: 'gpio_toggle' was not declared in this scope
34 | irq_set_exclusive_handler(IO_IRQ_BANK0, gpio_toggle);
| ^~~~~~~~~~~
// ( same thing when I used attachInterrupt( )
.
// when I try, naively, to change it in the setting :
irq_set_exclusive_handler(IO_IRQ_BANK0, __not_in_flash_func(gpio_toggle));
// ... of course, I get an error : expected primary-expression before '_attribute
// who knows ?

[Andy2No]

What happens if you remove the static keyword?

If I've understood what I've read correctly, in this context it's just an instruction to the linker to not make it available outside the source file where it's declared, so you could probably just omit "static" and not cause any problems for yourself.

[trimarco232]

same thing without "static" or when one use arduino's "attachInterrupt( "

[earlephilhower]

__not_in_flash_func is a function decorator attribute so you need to put it in the function definition, not in the call. static can cause the GCC attributes it's defining to be in the wrong order, so don't make the ISR static.

Just testing this in an empty sketch builds in the IDE

void __not_in_flash_func(gpio_toggle)() {
  gpio_xor_mask(1 << 11);
}

void setup() {
  // put your setup code here, to run once:

}

void loop() {
  // put your main code here, to run repeatedly:
irq_set_exclusive_handler(IO_IRQ_BANK0, gpio_toggle);

}

[trimarco232]

of course , this works , but what I had like , if it is possible and if it has sense, is to put the whole ISR in RAM , not only a function located in the ISR
I have experience with ESP32 : the attribute IRAM_ATTR puts the whole ISR in RAM
void IRAM_ATTR anyISR() {
Statements;
}

[earlephilhower]

I don't understand what you mean by "whole ISR in IRAM". The function is the whole ISR. I think the set-exclusive thing actually writes the IRQ handler address to the vector table so it's really all there is executing there...

This is equivalent to the IRAM_ATTR on the ESPs, in fact. We just set an attribute saying "store in DATA, not in TEXT" (on the ESP8266 it's in the IRAM segment, but same idea).

[trimarco232]

sorry , ok , it works so
the compiling error was just because I have placed the ISR at the end of the sketch, after the loop
(one have to place it before , or to declare it at the beginning)
great thanks Earle