[solved] Speed up call of Interrupt Service Routine

Official Hardware Mega Due
1

I am interfacing an ADC (AD7892-2) with my DUE. Eventually I want to run it at 400 ksps, but while I debug I've slowed it down to just 30ksps. The DUE is interrupted by the "end of conversion" pulse from the ADC, when it will eventually read the result in either parallel or serial mode. For now, I am running a dummy interrupt service routine which is as short as I can make it (direct port manipulation to make a very short pulse). It seems like it takes about 4us just to call the routine, which is too long for what I need.
Here is my code:

void setup(){
   pinMode(11, OUTPUT);
   attachInterrupt(6,getData,FALLING);
  delay(100);
}
void loop(){
}
void getData(){
  REG_PIOD_ODSR |= (1<<7);
  REG_PIOD_ODSR = 0;
}

I'm watching pins 11 and 6 on an oscilloscope and attached two oscillograms. The first shows that the pulse on pin 11 is just a couple of clock cycles long, as expected. This is the entire ISR. The second shows the trigger which should interrupt the DUE, and the ISR running after a wait of about 4us.
Is this delay unavoidable, or can I change something to call the ISR quicker? Is there a better way overall of doing this?
Thanks for reading!
-Matt
--I seem to be unable to upload the oscillograms right now. They are really pretty straightforward and show just what I've described.

2

Have you tried running the Interrupt on a different pin?
I heard that the interrupt handler runs an "if" on every pin on the port to see which pin has been interrupted.

You could try a pin lower on the port like pin 14 ( Pin 4 on Port D ) and see if it is quicker.

Or you could work directly with the registers instead of attachinterrupt.

3

Thanks EliteEng!
I moved the interrupt to pin 25 (Which is pin 0 on port d, and I had to move my output pulse to another port), and the time to call the ISR has dropped to about 800 ns. This is fast enough for me, so I won't attempt to learn how to handle interrupts using port registers.
Just for fun I experimented with using a few different pins on the same port to attach the interrupt. The time seemed to decrease pretty linearly with decreasing pin number on the port. It came to about 130 ns/port number. This is a good trick to know for when speed is critical!