color encoding for ws2812, MCU and why not, x86

yeppee, i had the time to code.

the lib can use a palette or a direct value.
if it uses a palette, the pixels are stored in one byte.
with direct value, it is like original with 3bytes.
the palette can be any size (no check at runtime) between 0 and 256 colors, stored as 4 bytes.

https://gitlab.com/foolavr/ws2812

here is the atomic part responsible of the NRZ transmission with palette.
the timings seems to be ok even if i use dirty C adressing to lookup the palette.

Code:

if (isPalette) {
    i = numBytes; 
    volatile uint8_t c;
    volatile uint16_t p;    
    while (i != 0) {
      p = *ptr;
      for (c = 0; c < 3; c++) {
        bit = 8;
        b = ((uint8_t*)(palette))[(p<<2)+c];
        asm volatile(
         "head201:"                   "\n\t" // Clk  Pseudocode    (T =  0)
          "st   %a[port],  %[hi]"    "\n\t" // 2    PORT = hi     (T =  2)
          "sbrc %[byte],  7"         "\n\t" // 1-2  if(b & 128)
           "mov  %[next], %[hi]"     "\n\t" // 0-1   next = hi    (T =  4)
          "dec  %[bit]"              "\n\t" // 1    bit--         (T =  5)
          "st   %a[port],  %[next]"  "\n\t" // 2    PORT = next   (T =  7)
          "mov  %[next] ,  %[lo]"    "\n\t" // 1    next = lo     (T =  8)
          "breq nextbyte201"          "\n\t" // 1-2  if(bit == 0) (from dec above)
          "rol  %[byte]"             "\n\t" // 1    b <<= 1       (T = 10)
          "rjmp .+0"                 "\n\t" // 2    nop nop       (T = 12)
          "nop"                      "\n\t" // 1    nop           (T = 13)
          "st   %a[port],  %[lo]"    "\n\t" // 2    PORT = lo     (T = 15)
          "nop"                      "\n\t" // 1    nop           (T = 16)
          "rjmp .+0"                 "\n\t" // 2    nop nop       (T = 18)
          "rjmp head201"              "\n\t" // 2    -> head20 (next bit out)
         "nextbyte201:"               "\n\t" //                    (T = 10)
          "st   %a[port], %[lo]"     "\n\t" // 2    PORT = lo     (T = 15)
    
          : [port]  "+e" (port),
            [byte]  "+r" (b),
            [bit]   "+r" (bit),
            [next]  "+r" (next)
          : [hi]     "r" (hi),
            [lo]     "r" (lo)
        );
      }
      ptr++;
      i--;
    }
  }

this is a trim of adafruit lib to fit my specific requirements, WS2812 with a ATMEGA328 at 16MHz, using arduino IDE.

it looks to work very well with my 16*16 matrix and i can now return to the project that needed this ram use reduction.