While trying to bitbang all 8 pins of a port to speed up ws2812b display, I run out of time extracting each bit to push on each port. I have to create a byte where each bit comes from a different part of the array. Below is code that works, using 2 pins. Adding a 3rd one breaks.

gistfile1.txt

template<FAB_TDEF>
inline void
avrBitbangLedStrip<FAB_TVAR>::eightWireSoftwareSendBytes(const uint16_t count, const uint8_t * array)
{
	const uint16_t blockSize = count/2;

	for(uint16_t pos = 0; pos < blockSize; pos++) {
		for(int8_t bit = 7; bit >= 0; bit--) {
			const uint8_t mask = 1 << bit;
			// overlap bits set to 1 to save time. generating the boolean seems to take the longest.
			const bool bitHigh5 = array[0*blockSize + pos] & mask;
			if (bitHigh5) SET_PORT_HIGH(dataPortId, dataPortPin);
			const bool bitHigh6 = array[1*blockSize + pos] & mask;
			if (bitHigh6) SET_PORT_HIGH(clockPortId, clockPortPin);

			// bits set to zero must cycle hi-lo fast so no overlap
			if (!bitHigh5) {
				SET_PORT_HIGH(dataPortId, dataPortPin);
				//DELAY_CYCLES(high0 - sbiCycles);
				DELAY_CYCLES(0);
				SET_PORT_LOW(dataPortId, dataPortPin);
			}
			if (!bitHigh6) {
				SET_PORT_HIGH(clockPortId, clockPortPin);
				//DELAY_CYCLES(high0 - sbiCycles);
				DELAY_CYCLES(0);
				SET_PORT_LOW(clockPortId, clockPortPin);
			}
			// normally I put delay loops but it's not needed at 16mhz
			DELAY_CYCLES(0);
			// I could set low only bits that are one, but i just set low all the bits to reduce instruction count.
			SET_PORT_LOW(dataPortId, dataPortPin);
			SET_PORT_LOW(clockPortId, clockPortPin);
			DELAY_CYCLES(0);
		}
	}
}