faster!

[sweh]

Based on a hint elsewhere, I tried to see if I could get faster output...

So without getting into assembler, I looked at direct port manipulation:

  DDRD=B11111111;
  boolean x=false;
  while(1)
  {
     x=!x;
     if (x) { PORTD=255; } else { PORTD=0;}
  }

(Yeah, I'm just setting all the pins for now).

That gives me a 748Khz signal. Amazingly quicker!

Removing the boolean test

  while (1)
  {
    PORTD=255;
    PORTD=0;
  }

This is the resulting assembler:

        ldi r24,lo8(-1)
  .L4:
        out 43-32,r24
        out 43-32,__zero_reg__
        rjmp .L4

Gives me a 3.9Mhz signal!

What's interesting, here, isn't the asymmetric on/off cycle (expected; need some NOPs in there to balance it out) but that the output is really noisy with massive overshoot on both low->high and high->low
transitions!


Of course we see these spikes at lower frequencies as well, but at this high speed the stabilisation time is a large portion of the signal.


FWIW, with two asm("nop") instructions in between the two PORTD settings I get 2.6Mhz

 0000 8FEF                  ldi r24,lo8(-1)
                    .L4:
 0002 8BB9                  out 43-32,r24
 0004 0000                  nop
 0006 0000                  nop
 0008 1BB8                  out 43-32,__zero_reg__
 000a 00C0                  rjmp .L4

Interesting stuff!