The heart of the LED cube code is the interrupt routine.
Every time this interrupt runs, the cube is cleared, data for the new layer is loaded onto the latch array, and the new layer is switched on. This remains on until the next time the interrupt runs, where the cube is cleared again, data for the next layer is loaded onto the latch array, and the next layer is switched on.
The ATmega32 has 3 timer/counters. These can be set to count continuously and trigger an interrupt routine every time they reach a certain number. The counter is reset when the interrupt routine is called.
We use Timer2 with a prescaler of 128 and an Output Compare value of 10. This means that the counter is incremented by 1 for every 128th cpu cycle. When Timer2 reaches 10, it is reset to 0 and the interrupt routine is called. With a cpu frequency of 14745600 Hz, 128 prescaler and output compare of 10, the interrupt routine is called every 1408th CPU cycle (128*11) or 10472.7 times per second. It displays one layer at a time, so it takes 8 runs of the interrupt to draw the entire cube once. This gives us a refresh rate of 1309 FPS (10472.7/8). At this refresh rate, the LED cube is 100% flicker free. Some might say that 1300 FPS is overkill, but the interrupt routine is quite efficient. At this high refresh rate, it only uses about 21% of the CPU time. We can measure this by attaching an oscilloscope to the output enable line (OE). This is pulled high at the start of each interrupt and low at the end, so it gives a pretty good indication of the time spent inside the interrupt routine.
Before any timed interrupts can start, we have to set up the Timer 2. This is done in the ioinit() function.
TCCR2 (Timer Counter Control Register 2) is an 8 bit register that contains settings for the timer clock source and mode of operation. We select a clock source with a 1/128 prescaler. This means that Timer/counter 2 is incrementet by 1 every 128th CPU cycle.
We set it to CTC mode. (Clear on Timer Compare). In this mode, the counter value TCNT2 is continuously compared to OCR2 (Output Compare Register 2). Every time TCNT2 reaches the value stored in OCR2, it is reset to 0 and starts counting from from 0. At the same time, an interrupt is triggered and the interrupt routine is called.
For every run of the interrupt, the following takes place:
1) All the layer transistors are switched off.
2) Output enable (OE) is pulled high to disable output from the latch array.
3) A loop runs through i = 0-7. For every pass a byte is outputed on the DATA bus and the i+1 is outputed on the address bus. We add the +1 because the 74HC138 has active low outputs and the 74HC574 clock line is triggered on the rising edge (transition from low to high).
4) Output enable is pulled low to enable output fro the latch array again.
5) The transistor for the current layer is switched on.
6) current_layer is incremented or reset to 0 if it moves beyond 7.
That's it. The interrupt routine is quite simple. I'm sure there are some optimizations we could have used, but not without compromising human readability of the code. For the purpose of this instructable, we think readability is a reasonable trade-off for a slight increase in performance.