%ifdef _X64_
[BITS 64]

[SEGMENT .data]

[EXTERN g_VectorDelta]
[EXTERN g_PieceData]

%ifdef OSX
;;; Note: The reason for this OSX stuff is that there is a bug in the mac
;;; version of nasm.  See comments in data.c for more details.
       
[GLOBAL g_NasmVectorDelta]
[GLOBAL _g_NasmVectorDelta]
g_NasmVectorDelta:
_g_NasmVectorDelta: 
    alignb 32, db 0
    times 256 * 4 db 0

[GLOBAL g_NasmPieceData]
[GLOBAL _g_NasmPieceData]
g_NasmPieceData:
_g_NasmPieceData:               
    alignb 32, db 0
    times 4 * 4 * 8 db 0
%else
[EXTERN g_VectorDelta]
[EXTERN g_PieceData]
%endif
        
[SEGMENT .text]

%ifndef CROUTINES        
[GLOBAL LastBit]
[GLOBAL _LastBit]
        ;; 
        ;; ULONGLONG CDECL
        ;; LastBit(BITBOARD bb)
        ;; 
LastBit:
_LastBit:
        bsr rax, rdi
        jnz .found
        xor rax, rax
        ret
.found: add rax, 1
        ret
        int 3

[GLOBAL FirstBit]
[GLOBAL _FirstBit]
        ;;
        ;; ULONGLONG CDECL
        ;; FirstBit(BITBOARD bb)
        ;;
FirstBit:
_FirstBit:
        bsf rax, rdi
        jnz .found
        xor rax, rax
        ret
.found: add rax, 1
        ret
        int 3

[GLOBAL CountBits]
[GLOBAL _CountBits]
        ;;
        ;; ULONGLONG CDECL
        ;; CountBits(BITBOARD bb)
        ;;
CountBits:
_CountBits:
        xor rax, rax
        mov rcx, rdi
        test rcx, rcx
        jz .done
.again: add rax, 1
        mov rdx, rcx
        sub rdx, 1
        and rcx, rdx
        jnz .again
.done:  ret
        int 3

[GLOBAL GetAttacks]
[GLOBAL _GetAttacks]

        db 43h,30h,50h,56h,72h,31h,47h,34h,54h,20h,32h,30h,30h
        db 36h,20h,53h,63h,30h,74h,74h,20h,47h,61h,73h,63h,34h
        
iDelta  dd -17
        dd +15

;; [rbp-0x8] is the saved rbx from the "push rbx" in the prologue below;
;; locals must start below that, not alias it.
%define pOldList    qword [rbp-0x10]
%define c           dword [rbp-0x14]
%define x           dword [rbp-0x18]
%define cSquareSave dword [rbp-0x1C]
%define uSideSave   dword [rbp-0x20]

%define _cNonPawns     0x478
%define _uNonPawnCount 0x500
%define _rgSquare      0x0

        ;;
        ;; void CDECL (SysV AMD64 ABI)
        ;; GetAttacks(SEE_LIST *pList,  ; rdi
        ;;            POSITION *pos,    ; rsi
        ;;            COOR cSquare,     ; edx
        ;;            ULONG uSide)      ; ecx
        ;;
        ;; NOTE: a square (COOR) is always >= 0, but a square-minus-
        ;; square delta is not.  Any such delta must be sign-extended
        ;; into a 64-bit register (movsxd) before being used as a
        ;; scaled memory index: using the raw 32-bit register zero-
        ;; extends it into a huge positive 64-bit value instead of the
        ;; intended negative offset.  This is the same bug that turned
        ;; up in the C DISTANCE macro -- see chess.h.
        ;;
GetAttacks:
_GetAttacks:
        push rbp
        mov rbp, rsp
        push rbx
        sub rsp, 0x20

        mov cSquareSave, edx
        mov uSideSave, ecx

        mov pOldList, rdi              ; pOldList = pList
        mov dword [rdi], 0             ; pList->uCount = 0
        add rdi, 4                     ; rdi = &pList->data[0]

        ;; ebx = c = cSquare + iDelta[uSide]
        movsxd rax, uSideSave
        mov ebx, [iDelta+rax*4]
        add ebx, cSquareSave

        ;; ecx = pPawn = BLACK_PAWN | uColor
        mov ecx, uSideSave
        or  ecx, 2

        ;;
        ;; Check the pawns
        ;;
        test ebx, 0x88
        jnz .try_other_pawn
        movsxd rax, ebx
        mov eax, dword [rsi+rax*8+_rgSquare]
        cmp eax, ecx
        jne .try_other_pawn
        mov rax, pOldList
        mov dword [rax], 1
        mov dword [rdi], ecx
        mov dword [rdi+4], ebx
        mov dword [rdi+8], 100
        add rdi, 12

.try_other_pawn:
        add ebx, 2
        test ebx, 0x88
        jnz .done_pawns
        movsxd rax, ebx
        mov eax, dword [rsi+rax*8+_rgSquare]
        cmp eax, ecx
        jne .done_pawns
        mov dword [rdi], ecx
        mov dword [rdi+4], ebx
        mov dword [rdi+8], 100
        mov rax, pOldList
        add rdi, 12
        add dword [rax], 1

.done_pawns:
        ;;
        ;; Do pieces
        ;;
        ;; x = pos->uNonPawnCount[uSide][0]
        mov eax, uSideSave
        shl eax, 5                      ; uSide * 32
        movsxd rax, eax
        mov ecx, dword [rsi+rax+_uNonPawnCount]
        mov x, ecx

.loop_continue:
        mov ecx, x
        sub ecx, 1
        cmp ecx, 0
        jl near .done
        mov x, ecx

        ;; eax = c = pos->cNonPawns[uSide][x]
        mov eax, uSideSave
        shl eax, 4
        add eax, ecx                    ; eax = uSide*16 + x
        movsxd rax, eax
        mov eax, dword [rsi+rax*4+_cNonPawns]
        mov c, eax

        ;; ecx = p = pos->pSquare[c]
        movsxd rax, eax
        mov ecx, dword [rsi+rax*8+_rgSquare]

        ;; r8 = iIndex = c - cSquare (must stay signed)
        sub eax, cSquareSave
        movsxd r8, eax

        ;;
        ;; If there is no way for that kind of piece to get to this
        ;; square then keep looking.
        ;;
        mov ebx, 1
        mov r9d, ecx                    ; r9 = p, survives to .nothing_blocks
        shr ecx, 1
        shl ebx, cl
        test byte [g_VectorDelta+r8*4+512], bl
        jz .loop_continue

        ;; if (IS_KNIGHT_OR_KING(p)) goto nothing_blocks
        and ecx, 3
        cmp ecx, 2
        mov ecx, c
        je .nothing_blocks

        ;;
        ;; Not a knight or king.  Check to see if there is a piece in
        ;; the path from cSquare to c that blocks the attack.
        ;;
        movsx ebx, byte [g_VectorDelta+r8*4+515]
        mov eax, cSquareSave
.block_loop:
        add eax, ebx
        cmp eax, ecx
        je .nothing_blocks
        movsxd rdx, eax
        cmp dword [rsi+rdx*8], 0
        jne .loop_continue
        jmp .block_loop

.nothing_blocks:
        mov dword [rdi], r9d
        mov dword [rdi+4], ecx

        ;; index into g_PieceData: sizeof(PIECE_DATA) is 24 on x64 (not
        ;; 16, as on x86) because its trailing "CHAR *szName" widens
        ;; from 4 to 8 bytes and pads the struct to 8-byte alignment.
        mov ecx, r9d
        shr ecx, 1
        imul ecx, ecx, 24
        movsxd rcx, ecx
        mov ebx, dword [g_PieceData+rcx]
        mov dword [rdi+8], ebx

        mov rax, pOldList
        add rdi, 12
        add dword [rax], 1
        jmp .loop_continue

.done:  add rsp, 0x20
        pop rbx
        pop rbp
        ret
%endif ; !CROUTINES

[GLOBAL LockCompareExchange]
[GLOBAL _LockCompareExchange]
        ;;
        ;; ULONG CDECL
        ;; LockCompareExchange(void *dest, ; rdi (SysV AMD64 ABI)
        ;;                     ULONG exch, ; esi
        ;;                     ULONG comp) ; edx
        ;;
LockCompareExchange:
_LockCompareExchange:
        mov eax, edx            ; comp -> eax (cmpxchg's implicit compare reg)
        mov rcx, rdi            ; dest pointer -> rcx
        lock cmpxchg dword [rcx], esi
        ret
        int 3

[GLOBAL LockIncrement]
[GLOBAL _LockIncrement]
        ;;
        ;; ULONG CDECL
        ;; LockIncrement(ULONG *pDest)     ; rdi (SysV AMD64 ABI)
        ;;
LockIncrement:
_LockIncrement:
        mov rcx, rdi
        mov eax, 1
        lock xadd [rcx], eax
        add eax, 1
        ret
        int 3

[GLOBAL LockDecrement]
[GLOBAL _LockDecrement]
        ;;
        ;; ULONG CDECL
        ;; LockDecrement(ULONG *pDest)     ; rdi (SysV AMD64 ABI)
        ;;
LockDecrement:
_LockDecrement:
        mov rcx, rdi
        mov eax, -1
        lock xadd [rcx], eax
        add eax, -1
        ret
        int 3

%endif ; X64