RGB565 SIMD (blend,...)

[BitsForPeople]

Hi :)
As hinted at on reddit, I have made some helper routines for RGB565 processing via the S3's PIE/SIMD instructions which you may want to give a try for their performance gains:
https://github.com/BitsForPeople/esp-bits-and-pieces/blob/b04a1481d0ea70b7a6edcca04d89ad5878887d4a/esp32-s3/rgb565/rgb565_s3_simd.hpp#L109

The idea is to use rgb565::RGBVec like

if(alphaQ8 == 0) {
  // just copy src1 to dst
} else 
if(alphaQ8 > 255) {
  // just copy src2 to dst
} else {
  rgb565::RGBVec rgb1;
  rgb565::RGBVec rgb2;

  while( haveMorePixels() ) {
    rgb1.from565(p_src1);
    rgb2.from565(p_src2);

    rgb1.blend(rgb2, alphaQ8);

    rgb1.to565(p_dst);

    p_src1 += 16;
    p_src2 += 16;
    p_dst += 16;
  }
}

This way, blending 16+16 RGB565 source pixels into 16 RGB565 destination pixels takes about 90-100 CPU clock cycles - a ~5x speed-up vs. a scalar implementation of the same.

If you give it a try, I'd be interested to hear about any performance differences you get.