Hi all, in this post I will be explaining the code of the Diablo Chess Engine which I made while learning about artificial intelligence and game bots. This post also serves as the documentation for the code of Diablo Chess Engine
I would recommend opening the code side by side to understand this documentation. In this explanation I will explain every file of the code of Diablo Chess Engine .
Files
I will go on describing each file and the functions in those files.
def.h
define various enums and macros for easy indexing of all types of pieces.
//enumurating costants for later use in arrays
//enumurating costants for later use in arrays
enum {FALSE,TRUE};
enum {WHITE,BLACK,BOTH};
enum {EMPTY,wP,wN,wB,wR,wQ,wK,bP,bN,bB,bR,bQ,bK};
enum {FILE_A,FILE_B,FILE_C,FILE_D,FILE_E,FILE_F,FILE_G,FILE_H,FILE_NONE};
enum {RANK_1,RANK_2,RANK_3,RANK_4,RANK_5,RANK_6,RANK_7,RANK_8,RANK_NONE};
enum{
A1=21,B1,C1,D1,E1,F1,G1,H1,
A2=31,B2,C2,D2,E2,F2,G2,H2,
A3=41,B3,C3,D3,E3,F3,G3,H3,
A4=51,B4,C4,D4,E4,F4,G4,H4,
A5=61,B5,C5,D5,E5,F5,G5,H5,
A6=71,B6,C6,D6,E6,F6,G6,H6,
A7=81,B7,C7,D7,E7,F7,G7,H7,
A8=91,B8,C8,D8,E8,F8,G8,H8,NONE,OFF
};
enum {WKCA=1,WQCA=2,BKCA=4,BQCA=8};
//enum weather castleing is possible for the the four types or not...represeted using first 4 bits
the board which we use is from 0 to 120 which is playable only from 21 to 98… which will be equivalent to 0 to 63 for a chess broad so we use the function and two arrays which inter-converts these boards…
we then define the board structure with the contents as –
Bord
int Pieces[SQNO]; //to know which type of piece is present at the respective index int Pieces[SQNO]; //to know which type of piece is present at the respective index ULL Pawns[3]; //bitboards for white,black,both pawns int PceNum[13]; //number of each type of piece present on the board int BigPce[2]; //number of all pieces except pawns int MajPce[2]; //number of rooks and queens of white,black; int MinPce[2];//number of knights and bishops of white,black int KingPos[2];//position of white and black kings int ValPce[2]; //stores the total materialistic value for white and black sides int Ply,HisPly;//number of half moves made and history to store previous plays int FiftyMov; int EnPas; //check for EnPass move int Side; //to know which side has to move int CastlePerm; //if the current king can castle or not rpresented by four bits for the above four enumrations ULL PosKey; //unique PosKey generated for every move UNDO History[MAXMOVES]; //this array stores all the previous game states with essential things which are stated in the undo //structure and also it helps us to check if a move is repeated or not,we can do so by using the //HisPly as the index and going back to all the states and see if the PosKey is repeated or not  int PceList[13][10]; PVTABLE PvTable[1];//we keep one PVTABLE in the main structure for the current board position...in the form of   // a pointer which is already given memeory... int PvArray[MAXDEPTH];//this array store the best line of moves upto certain depth...for the given position   int SearchHistory[13][120];//this will be used for move ordering for beta cutoff  int SearchKillers[2][MAXDEPTH];//this will also be only used for move ordering...basically these are the  //non capture moves which are causing the beta cutoff
then we define out UNDO structure to keep track of all the previous game moves so that we can crosscheck for 50moves and move repetition then we define out UNDO structure to keep track of all the previous game moves so that we can crosscheck for 50moves and move repetition also we can undo back to any move we want to…
Undo
//this structures is to undo any move if we want to to go back to certain previous stage of the game
//this structures is to undo any move if we want to to go back to certain previous stage of the game
typedef struct{
ULL PosKey;
int move;
int EnPas;
int FiftyMov;
int CastlePerm;
}UNDO;
ClearBit(b,sq) this Macro unsets the bit at number sq from the bitboard
SetBit(b,sq) this Macro sets the bit at number sq from the bitboard
we also define all the global arrays and function with extern keyword so that the can be accessed globally
Move
typedef struct{
int Move;
int Score;
}MOVE; //this structure stores the information about the move and in a very intresting wa of bitmasking...
we can see that we any square on the playable board i.e 21 to 98 can be represented using a 7bit number because the max value is 64+32+4 which can be made by setting some bits in a 7 digit number so we can see that all the aspects of a move can be stored in a number which has 32 bit…
0000 0000 0000 0000 0000 0111 1111 ->from (the masked value of first 7 bits will we be the index from where we start the move) 0000 0000 0000 0011 1111 1000 0000 ->to(the masked value of next 7 bits sotre the index where we want to go) 0000 0000 0011 1100 0000 0000 0000 ->captured(the masked value of next four bit ranges from 0 to 15 which will give us the piece (wP to bK) or value (0 to 12) which is being captured) 0000 0000 0100 0000 0000 0000 0000 ->EnPas (seting of this bit will tell that this move is an EnPas move) 0000 0000 1000 0000 0000 0000 0000 ->PawnSt(if this bit is set that means that this is a starting move of a pawn) 0000 1111 0000 0000 0000 0000 0000 ->Promoted(these 4 bits indicate that this move can promote to a Knight,Bishop,Queen or Rook each for respective bit) 0001 0000 0000 0000 0000 0000 0000 ->Castle(setting of this bit tells us that the current move will be a castling move)
so as we can see we can easily store all the information about a move in just a single 32 bit number so as we can see we can easily store all the information about a move in just a single 32 bit number so we don’t need to maintain separate variables to store all these
//macros for simple shifting and anding to find the masked value which will give us the index of movement
//macros for simple shifting and anding to find the masked value which will give us the index of movement
//and captured type index and promoted type index
#define FromSq(m) (m & (0x7F))
#define ToSq(m) ((m>>7) &(0x7F))
#define Captured(m) ((m>>14) & (0xF))
#define Promoted(m) ((m>>20) & (0xF))
//these macros are simply the values which we will & with our move to see if the corrsponding bits are set or not
#define FlagEnPas 0x40000
#define FlagPawnSt 0x80000
#define FlagCastle 0x1000000
//these flag are flag to see if the capturing is happening or not so for that we need to any one bit from 
//EnPas bit to 4 capyured bit must be on so the on coresponding hexadecimal number will be 0x7C000
#define FlagCap 0x7C000
//to see if the promotion in this move promotion is hapenning or not so for that any of the 4 bits representing
//the promotion must be on so the corresponding on bit structure in hexadecimal is 0xF00000 #define FlagPromoted 0xF00000
//this structure is the entries which will be present in the PVTABLE...it is a simple pair of the PosKey and the
//move which will be stored in our principal variation table
typedef struct{
ULL PosKey;
int Move; }PVENTRY;
//this is the actual PVTABLE structure which simply stores the entries array...which is the hash array for this
//PV hash table...and it contains the count of entries
typedef struct{
PVENTRY PvEntries[131070];
int Count; }PVTABLE;
//this structure is simply for the use f GUI....when connecting it to arena
typedef struct{
int starttime;
int stoptime;
int depth;
int depthset;
int timeset;
int movestogo;
int infinite;
long nodes;
int quit;
int stopped;
float fh,fhf; }SEARCHINFO;
INIT.cpp
this file does all our initialisation task…..which are : –
Convert()
this function inter converts the two types of boards using arrays sq120to64[120] and sq64to120[64]
InitHashKeys()
this function initializes our hashkey values from our arrays
ULL HashKey[13][120]; //to store a random 64bit no which states that that type of element is present on that sq no ULL SideKey; //to store a random number if it is the turn of white in the game ULL CastleKey[16]; //to store 64 random numbers which represent each possible type of castlepositions which are 0 to 15 //which are later xored according to the position of piece to generate a unique Hash Key value for current board condtion...repetition of a HashKey //value 3 times in a games indicates 3xrepetition and the game is draw
InitializeFileAndRanks()
this function initialises our RankNo and FileNo array which for any given 12 based square as an index will give its rank and file uses the arrays
FileNo[] and RankNo[];
AllInit()
this function just act to call the renaming functions…..
#define rand_64 ((ULL)rand() | \ (ULL)rand()<<15 | \ (ULL)rand()<<30 | \ ((ULL)rand()<<45 | \ ((ULL)rand() &0xf)<<60)) //this is a modified rand function which generates 64 bit random numbers for our hashkey values&amp;amp;nbsp;
bitboard.cpp
this file is just about our pawn bitboards and has the functions this file is just about our pawn bitboards and has the functions :-
void PrintBitBoard(ULL b)
this function is simply printing out a bit board by representing set bits as ‘X’ and unset as ‘-‘ we print the ranks from 8 to 1….remember in our program we represent the ranks opposite to what the appear in a game so while printing we simply reverse them….see the bitboard video for any confusion…..keep in mind that all the calculation and working is done on the structure where ranks are 1 to 8 from tops down….but only to show it to the user while printing we reverse the ranks….so that ranks become 8 to 1 as they appear in a normal chess board
int POP(ULL *bb)
return the first index of the set bit int the bit boards and sets it to 0 code taken from chessprogramming.wki and also the bitboard is used only for this function….we can also use the simple brute force method but it may be slower it use this table–
const int BitTable[64] = {
63, 30, 3, 32, 25, 41, 22, 33,
15, 50, 42, 13, 11, 53, 19, 34,
61, 29, 2, 51, 21, 43, 45, 10,
18, 47, 1, 54, 9, 57, 0, 35,
62, 31, 40, 4, 49, 5, 52,26,
60, 6, 23, 44, 46, 27, 56, 16,
7, 39, 48, 24, 59, 14, 12, 55,
38, 28, 58, 20, 37, 17, 36, 8 };
int Count(ULL b)
counts the number of set bits in a a bitboard
hashkey.cpp
this file is just to generate the hashkeys this file is just to generate the hash keys
ULL GenerateHashKeys(const BOARD *board)
this function takes in the current board status and according to the positions of pieces,Side to play,CastlePerm and the EnPas position generates the unique PosKey value by xoring all the random 64 bit numbers for those states of piece for that board and returns it….
board.cpp
this file has all the necessary functions that concern our chess board…this file has all the necessary functions that concern our chess board…
void ResetBoard(BOARD *b)
this function resets the entire board by setting all positions to empty or off board and the count of all other this to 0 or their respective starting value in the board structure
int CalcFen(char *arr,BOARD *board)
this function is the most important one as it takes in a FEN notation character array….and a point er to BOARD structure and then process that FEN notation to initialise the members of board structure with respect to the configuration in the FEN array….except for PceList,and maj,min and Big pieces….because as we will so they are initialised in other function as they help us to initialise the PceList effectively
void PrintBoard(BOARD *board)
simply function that prints the board when it has been set up by an FEN notation….it uses
char PceChar[]=".PNBRQKpnbrqk"; char SideChar[]="wb-"; char RankChar[]="12345678"; char FileChar[]="abcdefgh";
these arrays for easy indexing and printing using the type of pieces in the ENUM as the indexes we can effectively print them
void UpdateMaterial(BOARD *board)
function to update the remaining items i.e PceList,MajPce,MinPce…pawn bitboards etc…; once we have the board structure ready….we can easily count the number of big,maj,min,material value of pieces by simply running a loop from 0 to 120…and store the position or counts….we use various Boolean array which are defined in data.cpp which fasten our approach to see what type of a piece the piece is
extern int CheckBoard(BOARD *board)
this function crosschecks everything we have done to our board so far with various assert statements…it ensure that what we have stored on the board structure is actually present in the board…it is not that necessary if everything is correct but we need to be sure in such a big program…so we make a temporary arrays system which are same as those present in the board structure then using various value of board structure it cross checks other values in the board structure
data.cpp
this file contains all the data declarations needed for the program this file contains all the data declarations needed for the program
char PceChar[]=".PNBRQKpnbrqk";
char PceChar[]=".PNBRQKpnbrqk";
char SideChar[]="wb-";
char RankChar[]="12345678";
char FileChar[]="abcdefgh";
char Fen[]="rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1";
int PceBig[13]={FALSE,FALSE,TRUE,TRUE,TRUE,TRUE,TRUE,FALSE,TRUE,TRUE,TRUE,TRUE,TRUE};
int PceMaj[13]={FALSE,FALSE,FALSE,FALSE,TRUE,TRUE,TRUE,FALSE,FALSE,FALSE,TRUE,TRUE,TRUE};
int PceMin[13]={FALSE,FALSE,TRUE,TRUE,FALSE,FALSE,FALSE,FALSE,TRUE,TRUE,FALSE,FALSE,FALSE};
int PceVal[13]={0,100,325,325,550,1000,50000,100,325,325,550,1000,50000};
int PceCol[13]={BOTH,WHITE,WHITE,WHITE,WHITE,WHITE,WHITE,BLACK,BLACK,BLACK,BLACK,BLACK,BLACK};
int IsKnight[13];
int IsKing[13];
int IsRQ[13];
int IsBQ[13];
int IsSlide[13];
we have defined so many array as now when we use a type of piece say wN as in index to these arrays we will get the corresponding properties for that type of piece…i.e whether it is a Big,Maj,Min piece…its printing value…it colour…ad its Material Value….whether it is a knight,bishop,rook or a queen…so its very helpful to do so….
ghost.cpp
this is our main source file which will be executed and contains the main function…..it will call all the other remaining this is our main source file which will be executed and contains the main function…..it will call all the other remaining functions and will be used for testing ….
attack.cpp
int IsSqAttack(int sq,int side,BOARD *board)
this function here checks if the current given square is attacked by any side which is specified…on the current board…this is fairly easy we just need to set few arrays to mark the directions from this square from where certain types of pieces can attack…such as diagonals,vertical,horizontal etc..which are done by :-
const int KnDir[8] = {-8,-19,-21,-12,8,19,21,12 };
const int RkDir[4] = {-1,-10,1,10};
const int BiDir[4] = {-9,-11,11,9};
const int KiDir[8] = {-1,-10,1,10,-9,-11,11,9};
queen will take both the direction of the rook and bishop further these arrays make sure the whether the current piece is of the given type or not
extern int IsPawn[13]; extern int IsKnight[13]; extern int IsKing[13]; extern int IsRQ[13]; extern int IsBQ[13]; extern int IsSlide[13];
io.cpp
void PrintSq(int sq)
a simple function to print the file and rank of a current square using the file and rank arrays
void PrintMove(int move)
simple function to print the from,to and promoted bit values of the move using file and ranks
void PrintMoveList(MOVELIST *MoveList)
again a simple function to print a move list for a given side
movegenerator.cpp
Note:- remember there are only 2 move lists for a each side and not individually for each piece
#define MkMove(f,t,cap,prom,flag) (f|(t<<7)|(cap<<14)|(prom<<20)|flag) //a macro define to reduce our task to type every time to make a move from the values
void AddMove(BOARD *board, int Move,MOVELIST *MoveList)
function that takes in a move list say for WHITE and a possible move and add its to the move list of WHITE it takes in all the three types of moves CaptureMove,SimpleMove…and special EnPas move for pawns
void GenMoves(BOARD *board,MOVELIST MoveList)
generates all the the possible moves to be added to a particular side whose turn it is…first it checks if its white side…if it is…it starts adding by pawn moves now to add moves from pawns there can be 4 types…
- simple one square ahead move for wP it is +10 s check if its empty and in board…if it is then add wp move is called because it can lead to a promotion if a wP is present on rank 6..
- starting move if the sq+20 is empty and the pawn is at rank2…then we also add a PawnSt flag.
- diagonal moves if it is a capture move then we check sq+9 and sq+11…it can also lead to promotion so we call the addwpMove instead of normal Addmove…because it takes care of promotions…
- and the last one is that it can be an EnPAs move..if that square on the board is marked as EnPas then we add flag with EnPas…in it
void AddwPMove(BOARD *board ,int from,int to,int cap,int flag,MOVELIST *MoveList)
adds moves of white pawns using the add move function…it takes into account that if the from of any pawn is from rank 2 then in its next simple or capturing move it will promote..else it will not
void AddbPMove(BOARD *board ,int from,int to,int cap,int flag,MOVELIST *MoveList)
similar function for black pawns….just the promotion rank check change to rank7
//these arrays easily help us to generate moves for sliding pieces of both sides as they define the starting
//index of white and black sliding pieces and when we reach a 0 we have completed one sides sliding pieces
int SlidePce[]={ wB,wR,wQ ,0,bB,bR,bQ,0}; int SlideIndex[]={0,4};
//similar arrays for non sliding pieces...
int NonSlidePce[]={wN,wK, 0 , bN,bK,0};
int NonSlideIndex[]={0,3};
//these store the direction for each piece which need to be added to the current square...to get all the squares
//where a type of piece indexed by rows can move;
int PceDir[13][8] = {
{ 0, 0, 0, 0, 0, 0, 0 },
{ 0, 0, 0, 0, 0, 0, 0 },
{ -8, -19, -21, -12, 8, 19, 21, 12 },
{ -9, -11, 11, 9, 0, 0, 0, 0 },
{ -1, -10, 1, 10, 0, 0, 0, 0 },
{ -1, -10, 1, 10, -9, -11, 11, 9 },
{ -1, -10, 1, 10, -9, -11, 11, 9 },
{ 0, 0, 0, 0, 0, 0, 0 },
{ -8, -19, -21, -12, 8, 19, 21, 12 },
{ -9, -11, 11, 9, 0, 0, 0, 0 },
{ -1, -10, 1, 10, 0, 0, 0, 0 },
{ -1, -10, 1, 10, -9, -11, 11, 9 },
{ -1, -10, 1, 10, -9, -11, 11, 9 } };
//these are the number of directions for each piece type for queen=8,bishp and rook=4
int NumDir[13] = { 0, 0, 8, 4, 4, 8, 8, 0, 8, 4, 4, 8, 8 };
for sliding and non sliding pieces we add move in 2 parts….
- for non sliding pieces….we first get the boar->Side in the variable Side…then with that side we get the starting index for non sliding pieces…which will be 0 when side i white and 4 when side is black….and then we run a wile loop till we get and EMPTY(0) piece which tells us that non sliding pieces for the current side are over…we loop each of them and inside this loop we loop till the number of direction of that piece…and then we test each direction by adding that may squares…if the new square is not off board we check if it is empty…if it is then it is a normal move else if its not empty and has a piece of opposite side (which we simply check by xoring with that pieces colour)..then it is a capture move….we do this for all direction for all non sliding pieces…
- all is same for sliding pieces just wile checking directions we run an inside loop to move to all possible squares in that direction (sliding)…until we reach an opposite colour square or which will be capture move or and EMPTY move which will be normal move or we get off board.. after that we test the next direction…and do this for all the sliding pieces of that colour
int MoveExists(BOARD *board,int Move)
it simply checks if a given move is possible on the board or not
int VictimScore[13]={0,100,200,300,400,500,600,100,200,300,400,500,600};
static int MvvLvaScores[13][13]={0};
these arrays are for the move ordering method of most valuable victim least valuable attacker like a pawn captures queen…we initially assign everything to zero…so that the score of non capture moves remain zero
void InitMvvLva()
this function is the actual function to initialise the MvvLvaScores array for every combination of 2 pieces
void GenCapMoves(BOARD *board,MOVELIST *MoveList)
this is 95% same function as GenMove…but it only generates the capture move which are required in the Quiescence search function
makemove.cpp
extern void RemovePiece(BOARD *board,int sq)
change the values of the board structure when a piece is removed
extern void AddPiece(BOARD *board,int sq,int Pce)
opposite to removePiece
extern void MovePiece(BOARD *board,int from,int to)
call first remove the addpiece for the given squares
extern int MakeMove(BOARD *board,int Move)
now this function is the most important one….it makes actual changes to the board structure and actually makes a move….to do so…we us our history array which stored the unique information about all our previous moves….when we make a move we firstly store all the unique values to the history structure indexed by HisPly..
board->History[board->HisPly].Move=Move; board->History[board->HisPly].PosKey=board->PosKey; board->History[board->HisPly].EnPas=board->EnPas; board->History[board->HisPly].FiftyMove=board->FiftyMove; board->History[board->HisPly].CastlePerm=board->CastlePerm;
we then increment the hisply and ply as the move has been made so we need to increase the index
now we first check if the moves are special moves like EnPas or castling..if so we remove the pawns of other color from the respective position..or if it is castling we move the rooks to the respective position decided by the move to where they are going if there is any EnPas position…it should now be removed so we unhash it from the hashkey and also the castlePerm we unhash it…it may remain same or get changed…we then update the castlepem and hash it back and we marks EnPas as NONE..we use the array CastlePerm for the updation
const int CastlePerm[120] = {
15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 13, 15, 15, 15, 12, 15, 15, 14, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 7, 15, 15, 15, 3, 15, 15, 11, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15 };
now if it is a capturing move we need to remove the captured piece..and increase the fifty moves counter
now we check if it was a starting move for a pawn…if it was then we need to add the enpas position for that pawns…and also hash it back to the position key
after this we see if there was a promotion…if so then we remove the piece moved from the previous step and add the promoted piece…finally we change the side and hash it to the unique key now only we check if this move has made the king into check position..if so we take back the move as it will be illegal and return false stating that this move cannot be made
so if the king is in a mate the entire move list will be traversed but every move will be taken back and returned false if even after that move king still remains in mate…so only those moves will be evaluated in which either the king moves or any other piece blocks the mate
extern void TakeMove()
this function is the exact mirror of the the make move function and we firstly restore all the unique value back from the history array after decreasing its size
board->EnPas=board->History[board->HisPly].EnPas; board->CastlePerm=board->History[board->HisPly].CastlePerm; board->FiftyMove=board->History[board->HisPly].FiftyMove;
we the nreverse the sides if it was an EnPas capture we add back the respective pawn pieces to their places or the rooks back to their original position…we finally move back the piece from (to 2 from)..we update the kingpos and see if it was a capture move…if it was we add the captured piece back to the board…if the pieces was promoted we remove theat promoted piece which is now standing at the from square with and add a pawn of the same colour over there
finally we restore back the unique PosKey value….we do it at the last because if we did it before then the AddPieceand RemovePiece functions would have altered its value which we didn’t wanted… either we can do that or we can let it be and and get back to its history value by manually xoring and with the help of add and remove functions
perftest.cpp
void PerftTest(int depth, BOARD *pos)
this function is basically backtracking to a given depth to calculate all the valid move that can be made from a given board position…it first makes one move …the generate a new move list for that position and then again makes a move and makes a new list …and so on till depth is 0…then it comes back one step and makes the second moves and goes on…just like dfs…and count the number of leaf nodes in the moves tree
void Perft(int depth,BOARD *board)
this is just a helper function for the above function
misc.cpp
this file contains all the misc functions this file contains all the misc functions if running the program in windows them WIN32 is defined and we get the time buy the library windows.h else if running on linux then we include sys/time.h to get the time…
#ifdef WIN32 #include "windows.h" #else #include "sys/time.h" #endif
int GetTime()
this function return the value of current time….it does not matter which system it is linux or windows we will use this when we enable the engine with the gui…then it will sent us a stop interrupt then we stop the searching and make the move
pvtable.cpp
this file has all the functions for our principal variation tablethis file has all the functions for our principal variation table
the pvtable is represented using two structure…
- a structures used as hashing element..it has a PosKey…and a best move to be hashed for that PosKey
- the table structure which which has the array of above structure
typedef struct{
ULL PosKey;
int Move; }PVENTRY;
typedef struct{
PVENTRY PvEntries[131070];
int Count; }PVTABLE;
void ClearPvTable(PVTABLE *PvTable)
this function is used to reset the Pvtable by setting everything to 0 and the clear the pvtable for use.. before the next search is applied
void InitPvTable(PVTABLE *PvTable)
this function is to initialise the count
void StorePvMove(BOARD *board,int Move)
to store a move in the pvtable…using hashing function as poskey%count…
int ProbePvMove(BOARD *board)
retrieve the move if it is present for that given current poskey in the board WORKING– it serves two purpose
- everytime we search a given depth it may happen that a certain position has already been reached by some other way in the same ongoing search before….so when we again encounter the same position we have a best move for this position stored which will most likey give us the best pv line from there on….
- now at the end of the search we have a caching of moves saved…called the Pv line which is the best possible moves upto a given depth…how do we get this…after the search we get the best move for the current position…then we make that move and get the next position…and we repeat the same step until we reach the max depth of we reach a none move..the the pvtable may get change multiple times for the same position..whenever we reach an alpha cuttoff..indicating we have a better move from this position..so we hash this new move tho that position…but at the end of the search we get the best move for all the index that are filled
why do we clear it every time when we have the best moves for a given position which may occur…after some time in the game the we do not need to search again..?
no, we will search for every move…because although we may get a best move say for a position at a depth 6…in the last search…and after several moves say we reach that position…but now if we use this previous value then we only get the result which is one length deep…because remember we got back after 6 depth so no further level analysis was done for the 6th depth…so we need to clear the pvtable every time….before a search…but if in an ongoing search a position is repeated then we can use it because now we only need the first best move…as w e may get maximum beta cutoff and the best pvline staring form this move
int GetPvLine(BOARD *board,int depth)
as the name suggest we get the best pvline stored in the PvArray…for a given depth..but we may also get an answer greater than the depth
evaluate.cpp
this file gives us the the score which a given move makes….by calculating the sum of values values which every pieces get at a given position for WHITE and the subtracting it with BLACK it also takes in the material value of the sides
if the side is white it return a +ve score else for black side it return a -ve score it is because it is alpha beta with negamax…so we need to return the score with the sides point of view…
this if when white makes a move and then we get the evaluation value of the current board state
/* * we are getting the -score from the evaluation function because we want to see the score we gets from the * points sides of view in the alpha beta pruning ...so lets say from the starting position we mak a move * from a2 to a4 of a pawn....then as soon as we make the move the side is changed..in the makemove function * so now when we evaluate the current position we will get the value from blacks point of view and that is * what we want...because the second node was the node for black or the min node so we want the value * from the mins point of view or blacks point of view...and as now black makes a move further in the alpha * beta search the sides again gets changed...and now we reach a max node...so we need to evaluate the * board from whites point of view and so on.... */
see it this way say white has three options which gives the score of
10 20 50
but these values will be returned from blacks point of view giving
-10 -20 -50
but the -sign during the calling of the alphabeta function again reverse it and takes the max score which will be 50
now say in the next depth black makes a move and now the the side is change to white..and in the evaluation function we check the change side…so from whites point of view the pints are 10 20 30
again the -ve sign at the call of the function will change them to
-10 -20 -30
and now black returns the maximum value which will be -10
and further back when the final return occurs for alpha this value again change to 10 by the function call by that node
so we get a final value of 10 by this path which is correct as black minimises and white tries to maximise
search.cpp
the actual search function for iterative deepening and alpha beta and quiescence search the actual search function for iterative deepening and alpha beta and quiescence search
#define INF 30000
#define MATE 29000
int IsRepeat(BOARD *board)
this functions return true if a certain board position is repeated…and of course if a pawn is moved or a piece is captured…then the fifty moves resets…because we cannot move a pawn back or bring back the captured piece….so from there on we all the positions before that cannot be repeated so we start the loop from the last time when the fifty moves was reset
static void ClearForSearch(BOARD *board, SEARCHINFO *info)
this functions clears all the previous data beforn starting a search..it clears our searchkillers for beta cutoff and searchjistory for alphacutoff..also it clears the pvtalbe
void SearchPos(BOARD *board,SEARCHINFO *info)
this function is out iterative deepening function…iterative deepening has an advantage that we already have a best move even if the time runs out…because searching for the current depth we have already searched and updated the table for the previous depth search…and we have stored the best move for that depth which will now help us in our pvmove cutoff discussed before….which will cutoff the maximum node and lead us to the best pvline
static int AlphaBeta(int Alpha,int Beta,int Depth,BOARD *board,SEARCHINFO *info,int Flag);
this is the final alphabeta function in negamax fashion…to get the best pvline inside this we sort our moves in decreasing order of their score…this is called move ordering and is a crucial part of alphabeta search as it cutoffs the maximum nodes…inside this we update our pvtable every time whenever alpha is improved….say in one path alpha was 10 but from the second path the score returned was 20…so the new best moves has change to that path an we need to update it in the pvtable
if we get a mate we send a very large value so that this path is taken above all..that is there is no other legal move for the other side we are getting a mate so we have to take this path…so we send a large negative value added with ply which gives us after how many moves we can get a mate
for eg say white can give a mate to black after 1 move then it must take this path say we move this path…now the side is changed…now it s blacks turn..since it is a mate black has now legal moves…so it return say -2900 value..which gets change to 2900 which is the maximum score white will ever get…so it will follow this path…
now say white is getting a mate after 3 steps…so it must avoid this path at all cost…so at the last depth when white also return -2900 value to the black nod it gets changed to 2900 since it is negamax fashion we always take the maximum fro all the nodes…both alpha and beta and it automatically gets minimised for beta and maximised for alpha but in this case when maximised black get the value of 2900 so it return the value of -2900 value to the first alpha node…so it knows that it must not take this path…whereas if it takes this path…now black has the value 2900 stored for this position which tell it to take this path…so its correct
static int Quiescence(int Alpha,int Beta,BOARD *board,SEARCHINFO *info)
this search is added to improve the way the engine plays…to improve its game
Prakhar's Blog 