Faster 16/32bit 2D Morton Code encode/decode functions

morton.h

uint16_t encode16_morton2(uint8_t x_, uint8_t y_)
{
  uint32_t res=x_|(uint32_t(y_)<<16);
  res=(res|(res<<4))&0x0f0f0f0f;
  res=(res|(res<<2))&0x33333333;
  res=(res|(res<<1))&0x55555555;
  return uint16_t(res|(res>>15));
}
//----

uint32_t encode32_morton2(uint16_t x_, uint16_t y_)
{
  uint64_t res=x_|(uint64_t(y_)<<32);
  res=(res|(res<<8))&0x00ff00ff00ff00ff;
  res=(res|(res<<4))&0x0f0f0f0f0f0f0f0f;
  res=(res|(res<<2))&0x3333333333333333;
  res=(res|(res<<1))&0x5555555555555555;
  return uint32_t(res|(res>>31));
}
//----

void decode16_morton2(uint8_t &x_, uint8_t &y_, uint16_t mc_)
{
  uint32_t res=(mc_|(uint32_t(mc_)<<15))&0x55555555;
  res=(res|(res>>1))&0x33333333;
  res=(res|(res>>2))&0x0f0f0f0f;
  res=res|(res>>4);
  x_=uint8_t(res);
  y_=uint8_t(res>>16);
}
//----

void decode32_morton2(uint16_t &x_, uint16_t &y_, uint32_t mc_)
{
  uint64_t res=(mc_|(uint64_t(mc_)<<31))&0x5555555555555555;
  res=(res|(res>>1))&0x3333333333333333;
  res=(res|(res>>2))&0x0f0f0f0f0f0f0f0f;
  res=(res|(res>>4))&0x00ff00ff00ff00ff;
  res=res|(res>>8);
  x_=uint16_t(res);
  y_=uint16_t(res>>32);
}

Based on some micro benchmarking, these functions are ~50% faster than classic bit twiddling implementations using separate2/compact2 for each coordinate. It's also better to use 16-bit version for performance if xy-coordinates are in range 0-255, which gains further ~30% boost.

[pvrtc] These formulas work for sequential scans:

twiddle1 = ((twiddle1 | 0xAAAAAAAAu) + 1u) & 0x55555555u;
twiddle2 = ((twiddle2 | 0x55555555u) + 2u) & 0xAAAAAAAAu;
twiddle = twiddle1 + twiddle2;