转置位矩阵

/*
 * http://www-graphics.stanford.edu/~seander/bithacks.html#ReverseByteWith64Bits
 * 
 * 0010 0100 -- 24
 * 0010 0001 -- 21 
 * 1111 0000 -- F0
 * 0111 1111 -- 7F
 * 1000 0000 -- 80
 * 0011 0111 -- 37
 * 1111 1111 -- FF
 * 0001 1111 -- 1F
 * |||| ||||
 * 4CFC 898A
 * 546E CEEE
 * 
 */

uint8_t t1[ 8 ] =
{ 0x24, 0x21, 0xf0, 0x7f, 0x80, 0x37, 0xff, 0x1f };

uint8_t t2[ 8 ];

union
{
  uint32_t l[ 2 ];
  uint8_t c[ 8 ];
} Tmp;

void foo( void )
{
  long Ll, Lh;
  uint8_t i;
  for ( i = 0; i < 8; i++ )
    Tmp.c = t1;
  
  i = 8;
  while ( i-- )
  {
    Ll = ( Tmp.l[ 0 ] & 0x01010101 ) * 0x01020408;  // 0x08040201
    Lh = ( Tmp.l[ 1 ] & 0x01010101 ) * 0x01020408;  // Reserved
      
    t2 = (uint8_t) ( ( ( Lh & 0x0F000000 ) >> 20 )
      | ( ( Ll & 0x0F000000 ) >> 24 ) );
    
    Tmp.l[ 0 ] >>= 1;
    Tmp.l[ 1 ] >>= 1;
  }
}

// 《高效程序的奥秘》翻译 Henry S. Warren. Jr. 的“Hacher's Delight"
// 0x57 0x17 0x97 0x13 0x37 0xf6 0x32 0x2a
void transpose8( uint8_t i[ 8 ], uint8_t o[ 8 ] )
{
  uint32_t x, y, t;
  
  x = ( i[ 0 ] << 24 ) | ( i[ 1 ] << 16 ) | ( i[ 2 ] << 8 ) | i[ 3 ];
  y = ( i[ 4 ] << 24 ) | ( i[ 5 ] << 16 ) | ( i[ 6 ] << 8 ) | i[ 7 ];
  
  t = ( x & 0xf0f0f0f0 ) | ( ( y >> 4 ) & 0x0f0f0f0f );
  y = ( ( x << 4 ) & 0xf0f0f0f0 ) | ( y & 0x0f0f0f0f );
  x = t;
  
  t = ( x ^ ( x >> 14 ) ) & 0x0000cccc;
  x = x ^ t ^ ( t << 14 );
  t = ( y ^ ( y >> 14 ) ) & 0x0000cccc;
  y = y ^ t ^ ( t << 14 );
  
  t = ( x ^ ( x >> 7 ) ) & 0x00aa00aa;
  x = x ^ t ^ ( t << 7 );
  t = ( y ^ ( y >> 7 ) ) & 0x00aa00aa;
  y = y ^ t ^ ( t << 7 );
  
  o[ 7 ] = x >> 24;
  o[ 6 ] = x >> 16;
  o[ 5 ] = x >> 8;
  o[ 4 ] = x;
  o[ 3 ] = y >> 24;
  o[ 2 ] = y >> 16;
  o[ 1 ] = y >> 8;
  o[ 0 ] = y;
}

int main( void )
{
  int i;
  uint8_t t1[ 8 ] =
  { 0x24, 0x21, 0xf0, 0x7f, 0x80, 0x37, 0xff, 0x1f };
  uint8_t t2[ 8 ];
  
  transpose8( t1, t2 );
  for ( i = 0; i < 8; i++ )
    printf( "0x%x ", t2 );
}