[Double-O-Seven]

Very nice!

//Edit: To the speed comparator above:
This is no WTF. 8-bit and 32-bit is already built in in PAWN, 16-bit not. So 16-bit needs some calculations and this takes time.

//Edit 2: Why is there a DBIT_VAR?

pawn Код:

#define Bit16_Get(%0,%1)                                                        \
        (((DBITS_VAR=(%1)),DBITS_VAR&1)?(%0[(%1)/2]&0x0000FFFF):((%0[(%1)/2]>>>16)&0x0000FFFF))

You could just do this:

pawn Код:

#define Bit16_Get(%0,%1)                                                        \
        ((((%1)&1)?(%0[(%1)/2]&0x0000FFFF):((%0[(%1)/2]>>>16)&0x0000FFFF))

//Edit 3:

pawn Код:

#define Bit16:%0<%1>                                                            \
        %0[(%1)/2]

I'm not sure, but maybe you should round the real array size up:

pawn Код:

#define Bit16:%0<%1>                                                            \
        %0[((%1)+2)/2] // Round up here

Assume you want 31 16-bit cells, now you divide 31 by 2 and recieve 15. So there would be no 31st cell...
If you round up, there's a 32nd 16-bit cells, which you don't need, but it's still better.
Too much is better than not enough.