Support any number of strands from 1 to 8; make demo more like arduino demo

[jepler]

Also closes #5

[jepler]

@gamblor21 do you want to test this PR with fewer than 8 pins?

[gamblor21]

@gamblor21 do you want to test this PR with fewer than 8 pins?

Hope to test it tomorrow

[gamblor21]

Nice way of accounting for less then 8 strands. One issue though - bitops.bit_tranpose does not work for 1 strand (works 2-8).

I did test my setup (which is 1 strand) but setting num_strands to 2 and doubled the pixels and it did work. But either the library has to exclude 1 strand or bit_transpose needs a tweak or a 3rd PIO program that handles just 1 strand with a different data format.

Good news, filling 300 pixels (well 600 but half the fake ones) went from 73fps to about 182fps so I think getting it to work even for 1 strand may have worth.

[gamblor21]

I did manage to make a quick core change to bitops.bit_transpose to handle one strand. I can make a PR for it tomorrow (too late now) but basically in shared-module/bitops/__init__.c in transpose_var change the end of the switch to:

        case 2:
            y |= *src << 8;
            src -= src_stride;
            MP_FALLTHROUGH;
        case 1:
            y |= *src;

and in the shared-bindings change the check to 1 instead of 2 as the minimum num_strands.

Also noticed there are some print statements still in the code.

[jepler]

You're not wrong to want support for 1 strip using background writes! Let's make that a separate issue, especially in case it requires a core change.

However, I think I'd rather have a separate program for 1 strip, because the extra memory usage and processing time for the bit_transpose operation can be eliminated entirely. A 2nd copy of the data is still needed in order to be free from tearing (writing a partially updated strip) but not a copy that's enlarged by 8x because each bit becomes a full byte! And it means no core change is required.

(a smaller savings could be had for 2 to 4 strands, with a core change, but I'm less convinced this is worth it)

[jepler]

I've filed an issue for the 1 strand case, so hopefully you can approve this PR as it stands.

[gamblor21]

Debugging print left behind. Also on line 171. Otherwise looks great.

[jepler]

I went ahead and added 1-strand support, but changed the structure a bit so it probably deserves a full re-test

I tested the 1-strand version on the neopixel bff. The text scroller demo running unthrottled increases from 167 to 210fps! (and just commenting out pixels.show also gives 210fps 🤣 )

[jepler]

There was a mistake in my benchmarking of the neobff text scroller. The actual frame rates were 928fps (neopxl8 background) and 446fps (standard neopixel) 🚀

[gamblor21]

Tested with 1 and 2 strands, both worked great.

1 strand measure a speed-up from 73.2fps to 99.6fps over 300 neopixels. A 36% increase in speed.