Conway's Game Of Life

[HenrikOlsson]

In the Grid Evalutaion section, precalculating the Array index values and replacing the AND-logic with nested IF/THEN saves Another bunch of cycles

'Grid Evaluation	

CntPlus33 VAR WORD
CntPlus32 VAR WORD
CntPlus31 VAR WORD
CntPlus1  VAR WORD

CntMinus33 VAR WORD
CntMinus32 VAR WORD
CntMinus31 VAR WORD
CntMinus1  VAR WORD

CntPlus33 = Counter + 33
CntPlus32 = Counter + 32
CntPlus31 = Counter + 31
CntPlus1 = Counter +1

CntMinus33 = Counter - 33
CntMinus32 = Counter - 32
CntMinus31 = Counter - 31
CntMinus1 = Counter - 1	        	
	
	IF led[CntMinus1] > 0 THEN
      IF counter > 0 THEN
	    Colour[population] = led[CntMinus1]
		population = population + 1
	  ENDIF
	ENDIF
	
	IF led[CntPlus1] > 0 THEN
      IF counter < 1023 THEN
	    Colour[population] = led[CntPlus1]
		population = population + 1  		
       ENDIF
    ENDIF
	
	IF botedge = 0 THEN
		IF led [CntMinus33] > 0 THEN
		    Colour[population] = led[CntMinus33]
			population = population + 1			
		ENDIF
		IF led [CntMinus32] > 0 THEN
		    Colour[population] = led[CntMinus32]  
			population = population + 1
		ENDIF
		IF  rigedge = 0 and led [CntMinus33] > 0 THEN
		    Colour[population] = led[CntMinus33]
			population = population + 1
		ENDIF
	ENDIF
	
	IF topedge = 0 THEN
		IF led [CntPlus31] > 0 THEN
		    Colour[population] = led[CntPlus31]
			population = population + 1
		ENDIF
		IF led [CntPlus32] > 0 THEN
		    Colour[population] = led[CntPlus32]
			population = population + 1
		ENDIF
		IF  rigedge = 0 and led [CntPlus33] > 0 THEN 
		    Colour[population] = led[CntPlus33]
			population = population + 1
		ENDIF
	ENDIF

It's not nearly as elegant but it does increase the performance a bit. All in all that 1024 iteration FOR/NEXT loop took 265ms to execute on my test board, now it's 67. Obviously I don't know if it still WORKS so it might all be for nothing :-)

/Henrik.

[retepsnikrep]

I'm very grateful for all these interesting replies. Thank you.

There are three basic 1024 element byte arrays.


Led var byte[1024] used to hold the current generation of cells


LedNew var byte [1024] used to to hold the next generation of cells (This is copied into the Led array after evaluation is completed)


NeoLed var byte [1024] used to hold the NeoPixel led data.

This third array is required because the layout of the NeoPixels grid is one continuous string of 1024 leds arranged like a snakes and ladders board.

This is different to the other two arrays which are a conventional X/Y grid layout, so the Led data has to be parsed and re-organised to display correctly on the neopixel grid.


How are people timing the loops etc.
I would be interested in that code so I can see what difference changes make.

[HenrikOlsson]

There's also an array called colour, how large is that?

For this I just toggle an output and measure it with the scope but for more precise measurements I use one of the onboard timers. Reset it, start it, execute code, stop it, output its value.
Doing that it's important to have SOME idea about what you're measuring in order to set up the timer with proper prescaling value to make sure you don't overflow, giving you false readings.

[retepsnikrep]

I incorporated all those ideas and it is now running at over 13hz v 3hz before. Well done Team!

The Colour Array is only 8 elements and holds the colour of any adjacent cells to the one being tested.



Which of these two is technically quickest? Both seem to work well but I haven't got timing set up yet.

  Temp = Counter // 32

  If Temp = 0 THEN
    lefedge = 1
  ENDIF

  IF Temp = 31 THEN
    rigedge = 1
  ENDIF

or

Temp = Counter & 65504
If Temp = 0 THEN
lefedge = 1
ENDIF


IF Temp = 31 THEN
rigedge = 1
ENDIF

Here we have it at 13hz..

[richard]

hmm i think i got that the wrong way round
try Temp = Counter & 31

the bitwise and is faster than a // which uses a divide

i wonder about led/LedNew arrays as byte . a cell is either dead or not dead a bit can store that info

[HenrikOlsson]

Great that it all still works!
I timed it and Richards is a quite a bit quicker. On my test setup it went from 67 to 48ms which is not surprising since, generally speaking, anything involving division is a killer.
I also found that I had missed one AND statement, replaced that with nested IFs and I'm now measuring 44.7ms

[retepsnikrep]

That all works and I changed the AND statements.
Maybe 15hz now but I need to do some proper timing.

I wonder about led/LedNew arrays as byte . a cell is either dead or not dead a bit can store that info

The arrays also store LED RGB colour and brightness information so one bit is not sufficient.
The colour of a new cell is dependent on the adjacent three populated cells that spawn it.

[HenrikOlsson]

From 3 to 15 frames per second in a couple of hours, I call that improvement.
At one place (at least) you first do IF LED[Counter] = 0 and then a couple of lines further down you do IF LED[Counter] > 0 which seems a bit of a wate since you already know if it IS 0 and if it's not it's got to be > 0 (since it can't be negative in case you're not using LONGs). There's also Population = Population + 1 in 8 places, can you move that outside each of the IF/THEN statements and only have ONE of them? That likely won't increase performance but will save space (if that matter at all).

We're probably talking microseconds here but since it all executes 1024 times per frame it does add up.

Anyhow, nice job and keep those updates and videos coming!