00000656 FF030000 00000080 FFFF1F00 FF03C0FF FFFF3F80 FFFF1F00 ..................?.....
00000680 0000C07F 00002000 FF030000 0000C07F 00002000 FF030000 ...... ........... .....
00000704 0000C07F 00002000 FF030000 0000C07F 00002000 FF030000 ...... ........... .....
00000728 00FC3F00 FFFF3F80 00002000 00FC3F00 FFFF3F80 00002000 ..?...?... ...?...?... .
00000752 00FC3F00 FFFF3F80 00002000 00FC3F00 FFFF3F80 00002000 ..?...?... ...?...?... .
00000776 00F83F80 00002000 00FCFFFF 00F83F80 00002000 00FCFFFF ..?... .......?... .....
00000800 00000080 00002000 00FC3F00 00000080 00002000 00FC3F00 ...... ...?....... ...?.
00000824 00F81F00 00F83F80 0004C0FF 00F81F00 00F83F80 00040000 ......?...........?.....
00000848 00F81F00 00F83F80 00040000 00F81F00 FF070080 00040000 ......?.................
00000872 FF072000 FF070080 FFFBFFFF FF072000 FF070080 FFFBFFFF .. ........... .........
00000896 FF072000 FF070080 FFFBFFFF FF072000 FF070080 FFFBFFFF .. ........... .........
these right here.
...
maybe I'll try xoring them

To clarify: these are the compressed normals and tangents and binormals of the model raw of a cube. so values are 0, -1 and 1.
But obviously there's more then one 0 or something?
Code: Select all
ushort radix = (ushort)(bits >> 00 & 0x7FF); // retrieve the bits of this componant (first 11 bits)
if (radix == 0x7FF || radix == 0) return 0; // two special cases for zero: return zero on either
if ((radix & 0x400) == 0x400) // if sign bit is set, output should be negetive
return -(float)(~(radix) & 0x3FF) * xy_max_inverse; /* return the radix 'ones compliment' trimming the sign bit
* return the negetive value*/
else /* else just return the radix value multiplied by the pre-
* calculated ratio */
return (float)(radix) * xy_max_inverse;