nibble swapping

[lucasart]

I'm trying to write a function that swaps the nibbles of a bitboard two by two.

Just to clarify what that means:
- a nibble is 4 bits (half a byte)
- a bitboard is therefore 16 nibbles: b = N0,N1, ..., N14,N15
- function should return: f(b) = N1,N0, N3,N2, ..., N15,N14

Is there a smart way to perform that operation really fast ?

Thank you

[mar]

I'm trying to write a function that swaps the nibbles of a bitboard two by two.

Just to clarify what that means:
- a nibble is 4 bits (half a byte)
- a bitboard is therefore 16 nibbles: b = N0,N1, ..., N14,N15
- function should return: f(b) = N1,N0, N3,N2, ..., N15,N14

Is there a smart way to perform that operation really fast ?

Thank you

Maybe shift left and mask or shift right and mask?
Like ((b << 4) & maskhi) | ((b>>4) & masklo)?

[Evert]

Maybe shift left and mask or shift right and mask?
Like ((b << 4) & maskhi) | ((b>>4) & masklo)?

That's almost certainly the only portable solution. Note that it can be done on an entire bitboard in one go, assuming the encoding is A1=0, B1=1,...,H8=63, using exactly that code (and using suitable bitmasks).

Should be fast enough.

[lucasart]

I'm trying to write a function that swaps the nibbles of a bitboard two by two.

Just to clarify what that means:
- a nibble is 4 bits (half a byte)
- a bitboard is therefore 16 nibbles: b = N0,N1, ..., N14,N15
- function should return: f(b) = N1,N0, N3,N2, ..., N15,N14

Is there a smart way to perform that operation really fast ?

Thank you

Maybe shift left and mask or shift right and mask?
Like ((b << 4) & maskhi) | ((b>>4) & masklo)?

Thanks. Simple and portable

[Evert]

There may also be something relevant at http://chessprogramming.wikispaces.com/ ... Techniques

[hgm]

Maybe shift left and mask or shift right and mask?
Like ((b << 4) & maskhi) | ((b>>4) & masklo)?

Indeed, that is also the method I use for diagonal mirroring of position indexes in the fairygen EGT generator. (Except that the groups of bits I want to swap are only 3-wide rank and file number there. So I use b<<3 & 07070707070 | b>>3 & 0707070707).

For non-8x8 square boards I had a hard time finding something similarly simple. In the end I chose for keeping two separate keys, one collecting all rank coordinates, the other all file coordinates, so that I can do mirroring in one dimension or the other by subtracting the index from a constant. The total index is then rankIndex*WIDTH + fileIndex or fileIndex*WIDTH + rankIndex, depending on the diagonal mirror class I need.

[Don]

I'm trying to write a function that swaps the nibbles of a bitboard two by two.

Just to clarify what that means:
- a nibble is 4 bits (half a byte)
- a bitboard is therefore 16 nibbles: b = N0,N1, ..., N14,N15
- function should return: f(b) = N1,N0, N3,N2, ..., N15,N14

Is there a smart way to perform that operation really fast ?

Thank you

You could try using a 64k table that does the swapping 16 bits at a time. I don't know how fast or slow it would be but it would be worth a quick test.

My best guess is that the shift intensive expression will be faster because even with lookup tables you will have to do a fair amount of shifting as well as 4 memory accesses and additional cache pressure.