Attempt to vectorize memory access while writing pixels to 8 ports.

gistfile1.txt

template<FAB_TDEF>
inline void
avrBitbangLedStrip<FAB_TVAR>::eightPortSoftwareSendBytes(const uint16_t count, const uint8_t * array)
{
	const uint16_t blockSize = count / 8;
	uint8_t r0 __asm__("r2") = 0;
	uint8_t r1 __asm__("r3") = 0;
	uint8_t r2 __asm__("r4") = 0;
	uint8_t r3 __asm__("r5") = 0;
	uint8_t r4 __asm__("r6") = 0;
	uint8_t r5 __asm__("r7") = 0;
	uint8_t r6 __asm__("r8") = 0;
	uint8_t r7 __asm__("r9") = 0;

	for(register uint16_t c = 0; c < blockSize + 8; c++) {
		for(register int8_t b = 7; b >= 0; b--) {
			uint8_t bitmask __asm__("r10");

//			bitmask |=  r0 & 1      ; r0 >>= 1;
//			bitmask |= (r1 & 1) << 1; r1 >>= 1;
//			bitmask |= (r2 & 1) << 2; r2 >>= 1;
//			bitmask |= (r3 & 1) << 3; r3 >>= 1;
//			bitmask |= (r4 & 1) << 4; r4 >>= 1;
//			bitmask |= (r5 & 1) << 5; r5 >>= 1;
			bitmask |= (r6 & 1) << 6; r6 >>= 1;
			bitmask |= (r7 & 1) << 7; r7 >>= 1;

			// Load ONE byte per iter.
			// Skip end condition check to reduce CPU cycles.
			// It is expected that the LED strip won't have extra pixels.
			switch (b) {
/*
				case 0:
					// if (c < blockSize)
					r0 = array[c];
					break;
				case 1:
					// if (c < blockSize+1)
					r1 = array[c + 1 * blockSize];
					break;
				case 2:
					r2 = array[c + 2 * blockSize];
					break;
				case 3:
					r3 = array[c + 3 * blockSize];
					break;
				case 4:
					r4 = array[c + 4 * blockSize];
					break;
				case 5:
					r5 = array[c + 5 * blockSize];
					break;
*/
				case 6:
					r6 = array[c + 6 * blockSize];
					break;
				case 7:
					r7 = array[c + 7 * blockSize];
					break;
			}

			// Set all HIGH, set LOW all zeros, set LOW zeros and ones.
			FAB_PORT(dataPortId, 0xFF);
			DELAY_CYCLES(4); //high0 - sbiCycles);

//			FAB_PORT(dataPortId, bitmask);
			AVR_PORT(dataPortId) &= bitmask;
			DELAY_CYCLES(4); //high1 - sbiCycles - high0);

			FAB_PORT(dataPortId, 0x00);
			// Let's assume we'll spend enough time doing math to not need to wait
			//DELAY_CYCLES(20); //low0 - cbiCycles);
		}
	}
}