/*
V0.2
  overriding shiftOut
  direct manipulation of latchPin and clockPin
    to speed up
  now down from ~110 us to ~25 us
  needs some improvements for "addressing" the pins
V0.3
  using digitalWriteFast  
*/

#include <digitalWriteFast.h>

byte dataPin = 2;
byte clockPin = 3;
byte latchPin = 4;

uint8_t data = 0x01;


void setup()
{
  pinMode(dataPin, OUTPUT);
  pinMode(clockPin, OUTPUT);
  pinMode(latchPin, OUTPUT);

  Serial.begin(115200);

  //digitalWrite(dataPin, HIGH);

  //while (Serial.available() == 0) {}
}

void loop()
{
  //digitalWrite(latchPin, LOW);
  digitalWriteFast(latchPin, LOW);
  shiftOut(dataPin, clockPin, MSBFIRST, data);
  //digitalWrite(latchPin, HIGH);
  digitalWriteFast(latchPin, HIGH);

  delay(200);
  data++;
}

void shiftOut(uint8_t dataPin, uint8_t clockPin, uint8_t bitOrder, uint8_t val)
{
  //Serial.println("shift");
  uint8_t i;

  for (i = 0; i < 8; i++)
  {
    if (bitOrder == LSBFIRST)
    {
      //digitalWrite(dataPin, val & 1);
      digitalWriteFast(dataPin, val & 1);
      val >>= 1;
    }
    else
    {
      //digitalWrite(dataPin, (val & 128) != 0);
      digitalWriteFast(dataPin, (val & 128) != 0);
      val <<= 1;
    }

    //digitalWrite(clockPin, HIGH);
    digitalWriteFast(clockPin, HIGH);
    //digitalWrite(clockPin, LOW);
    digitalWriteFast(clockPin, LOW);
  }
}