move-do: use castling flags, other improvements (~12% on perft)

- always set K square and castling flags (remove expensive tests on K and castling flags)
- simplify castling rook move
- more restrictive sq_castle[]

Makefile

@@ -126,7 +126,7 @@ else # ifeq ($(build),dev)
         #        fen.c
         #CPPFLAGS  += -DDEBUG_FEN                           # FEN decoding
         #        hash / TT
-        #CPPFLAGS  += -DZOBRIST_VERIFY                      # double chk zobrist
+        CPPFLAGS  += -DZOBRIST_VERIFY                       # double chk zobrist
         #CPPFLAGS  += -DPERFT_MOVE_HISTORY                  # perft, keep prev moves
         #        attack.c
         #CPPFLAGS  += -DDEBUG_ATTACK_ATTACKERS              # sq_attackers
@@ -159,22 +159,22 @@ ifeq ($(build),release)
         #CFLAGS    += -g
         #CFLAGS    += -ginline-points                   # inlined funcs debug info
         LDFLAGS   += -flto
+else ifeq ($(build),perf)
+        CFLAGS    += -O3
+        CFLAGS    += -ggdb0                            # symbols (gdb, perf, etc.)
+        CFLAGS    += -ginline-points                   # inlined funcs debug info
+        CFLAGS    += -funroll-loops
 else ifeq ($(build),dev)
-        CFLAGS    += -Og
-        CFLAGS    += -g                                # symbols (gdb, perf, etc.)
+        CFLAGS    += -O1
+        CFLAGS    += -ggdb                             # symbols (gdb, perf, etc.)
         CFLAGS    += -ginline-points                   # inlined funcs debug info
         #CFLAGS += -pg                                 # gprof
         # Next one may be useful for valgrind (when invalid instructions)
         #CFLAGS += -mno-tbm
-else ifeq ($(build),perf)
-        CFLAGS    += -O3
-        CFLAGS    += -g                                # symbols (gdb, perf, etc.)
-        CFLAGS    += -ginline-points                   # inlined funcs debug info
-        CFLAGS    += -funroll-loops
 else ifeq ($(build),debug)
-        CFLAGS    += -Og
-        CFLAGS    += -g                                # symbols (gdb, perf, etc.)
-        CFLAGS    += -ginline-points                   # inlined funcs debug info
+        CFLAGS    += -O0                               # -Og hides some variables
+        CFLAGS    += -ggdb3                            # symbols (including macro)
+        #CFLAGS    += -ginline-points                   # inlined funcs debug info
         # for gprof
         #CFLAGS += -pg
         # Next one may be useful for valgrind (when invalid instructions)
@@ -229,16 +229,16 @@ else
 all: libs testing $(TARGET)
 
 release:
-	$(MAKE) BUILD=release clean all
+	$(MAKE) build=release clean all
 
 dev:
-	$(MAKE) BUILD=dev clean all
+	$(MAKE) build=dev clean all
 
 perf:
-	$(MAKE) BUILD=perf clean all
+	$(MAKE) build=perf clean all
 
 debug:
-	$(MAKE) BUILD=debug clean all
+	$(MAKE) build=debug clean all
 
 compile: brlib objs
 

src/bitboard.h

@@ -182,19 +182,6 @@ static __always_inline bitboard_t bb_file(int file)
 }
 */
 
-/**
- * bb_first_bb() - return bitboard of first square of a bitboard.
- * @bb: bitboard
- *
- * bb must be non-zero.
- *
- * @return: bitboard of first square (lsb) of @bb.
- */
-static __always_inline square_t bb_first_bb(bitboard_t bb)
-{
-    return bb & -bb;
-}
-
 /**
  * bb_next() - clear and return next (lsb) square of a bitboard.
  * @bb: &bitboard
@@ -205,7 +192,7 @@ static __always_inline square_t bb_first_bb(bitboard_t bb)
  */
 static __always_inline square_t bb_next(bitboard_t *bb)
 {
-    square_t sq = ctz64(*bb);
+    square_t sq = lsb64(*bb);
     *bb &= *bb - 1;
     return sq;
 }

src/chessdefs.h

@@ -16,9 +16,10 @@
 
 #include <brlib.h>                                /* brlib types */
 
-#define ONE                1ul
 #define U64(const_u64)     const_u64##UL
-#define BIT(i)             ( (u64) (ONE << (i)) )
+#define BIT(i)             ( U64(1) << (i) )
+#define BIT_NONE           ( U64(0) )
+#define BIT_ALL            ( ~BIT_NONE )
 
 #define BOARDSIZE          (8*8)
 #define GAMESIZE           1024                   /* max game size (512 moves) */
@@ -50,7 +51,7 @@
 
 /* castle_t bits structure
  */
-typedef enum {
+enum {
     CASTLE_NONE = 0,
     CASTLE_WK   = (1 << 0),                       /* 0001 */
     CASTLE_WQ   = (1 << 1),                       /* 0010 */
@@ -64,7 +65,9 @@ typedef enum {
     CASTLE_K  = (1 << 0),                         /* generic K/Q, bits 0 and 1 */
     CASTLE_Q  = (1 << 1),
     CASTLE_KQ = (CASTLE_K |CASTLE_Q),
-} castle_rights_t;
+};
+
+typedef u8 castle_rights_t;
 
 /* determine is oo or ooo is possible with castle flags f and color c
  */

src/eval-defs.c

@@ -256,8 +256,7 @@ static const struct pst {
              * A1 ..... H1
              */
             [PAWN] = {
-                { /* H1                                 H8 */
-                    /* midgame */
+                {                                 /* midgame */
                     +0,  0,   0,   0,   0,   0,  0,  0,
                     50, 50,  50,  50,  50,  50, 50, 50,
                     10, 10,  20,  30,  30,  20, 10, 10,
@@ -575,8 +574,8 @@ void pst_init(int set)
         eval_t mid_pc = piece_midval(pt);
         eval_t end_pc = piece_endval(pt);
         for (square_t sq = 0; sq < SQUARE_NB; ++sq) {
-            eval_t mid_pst = pst->val[MIDGAME][pt][sq];
-            eval_t end_pst = pst->val[ENDGAME][pt][sq];
+            eval_t mid_pst = pst->val[pt][MIDGAME][sq];
+            eval_t end_pst = pst->val[pt][ENDGAME][sq];
 
             pst_mg[BLACK][pt][sq] = (mid_pc * wmat + mid_pst * wpst) / 100;
             pst_eg[BLACK][pt][sq] = (end_pc * wmat + end_pst * wpst) / 100;

src/eval.c

@@ -41,7 +41,7 @@ s16 calc_phase(pos_t *pos)
 
     phase = max(phase, 0);
 #   ifdef DEBUG_EVAL
-    printf("calc phase:%d\n", phase);
+    printf("calculated phase:%d\n", phase);
 #   endif
     return phase;
 }

src/hash.c

@@ -202,7 +202,7 @@ int tt_create(s32 sizemb)
         hash_tt.bytes    = hash_tt.nbuckets * sizeof(bucket_t);
         hash_tt.mb       = hash_tt.bytes / 1024 / 1024;
 
-        hash_tt.mask     = -1ull >> (64 - nbits);
+        hash_tt.mask     = BIT_ALL >> (64 - nbits);
 
         hash_tt.keys     = safe_alloc_aligned_hugepage(hash_tt.bytes);
 
@@ -295,6 +295,7 @@ hentry_t *tt_probe_perft(const hkey_t key, const u16 depth)
     /* find key in buckets */
     for (i = 0; i < ENTRIES_PER_BUCKET; ++i) {
         entry = bucket->entry + i;
+        bug_on(entry->key && (key & hash_tt.mask) != (entry->key & hash_tt.mask));
         if (key == entry->key && HASH_PERFT_DEPTH(entry->data) == depth) {
             hash_tt.hits++;
             /*
@@ -390,7 +391,7 @@ hentry_t *tt_store_perft(const hkey_t key, const u16 depth, const u64 nodes)
 void tt_info()
 {
     if (hash_tt.keys) {
-        printf("TT: Mb:%d buckets:%'lu (bits:%u mask:%#x) entries:%'lu\n",
+        printf("TT: Mb:%d buckets:%'lu (bits:%u mask:%#lx) entries:%'lu\n",
                hash_tt.mb, hash_tt.nbuckets, hash_tt.nbits,
                hash_tt.mask, hash_tt.nkeys);
     } else {

src/hash.h

@@ -76,15 +76,15 @@ typedef struct {
 
     /* memory size in bytes/mb */
     size_t bytes;
-    u32  mb;
+    u32 mb;
 
     /* size in buckets/keys */
     size_t nbuckets;
     size_t nkeys;                                 /* nbuckets * NBUCKETS */
 
     /* internal representation */
-    u32  nbits;                                   /* #buckets in bits, power of 2 */
-    u32  mask;                                    /* nbuckets - 1, key mask */
+    u32 nbits;                                    /* #buckets in bits, power of 2 */
+    u64 mask;                                     /* nbuckets - 1, bucket mask */
 
     /* stats - unsure about usage */
     //size_t used_buckets;
@@ -106,8 +106,8 @@ typedef struct {
 
 extern hkey_t zobrist_pieces[16][64];
 extern hkey_t zobrist_castling[4 * 4 + 1];
-extern hkey_t zobrist_turn;                        /* for black, XOR each ply */
-extern hkey_t zobrist_ep[9];                       /* 0-7: ep file, 8: SQUARE_NONE */
+extern hkey_t zobrist_turn;                       /* for black, XOR each ply */
+extern hkey_t zobrist_ep[9];                      /* 0-7: ep file, 8: SQUARE_NONE */
 
 extern hasht_t hash_tt;                           /* main transposition table */
 

src/move-do.c

@@ -25,6 +25,20 @@
 #include "move-do.h"
 #include "hash.h"
 
+/**
+ * sq_castle - castling rights per square.
+ *
+ * position castling rights mask has to be AND'ed with this array 'from' and 'to'.
+ */
+castle_rights_t sq_castle[64] = {
+    [A1] = ~CASTLE_WQ, [B1 ... D1] = CASTLE_ALL,
+    [E1] = ~CASTLE_W,  [F1 ... G1] = CASTLE_ALL,
+    [H1] = ~CASTLE_WK, [A2 ... H7] = CASTLE_ALL,
+    [A8] = ~CASTLE_BQ, [B8 ... D8] = CASTLE_ALL,
+    [E8] = ~CASTLE_B,  [F8 ... G8] = CASTLE_ALL,
+    [H8] = ~CASTLE_BK,
+};
+
 /**
  * move_do() - do move.
  * @pos:    &pos_t position
@@ -60,9 +74,8 @@ pos_t *move_do(pos_t *pos, const move_t move, state_t *state)
 
     *state = pos->state;                          /* save irreversible changes */
 
-    /* update key: switch turn, reset castling and ep */
+    /* update key: switch turn, reset ep */
     key ^= zobrist_turn;
-    key ^= zobrist_castling[pos->castle];
     key ^= zobrist_ep[EP_ZOBRIST_IDX(pos->en_passant)];
 
     ++pos->clock_50;
@@ -74,80 +87,69 @@ pos_t *move_do(pos_t *pos, const move_t move, state_t *state)
 
     bug_on(COLOR(piece) != us);
 
-    if (is_promotion(move)) {
-        bug_on(sq_rank(to) != sq_rel_rank(RANK_8, us));
-        new_piece = MAKE_PIECE(move_promoted(move), us);
-    }
-
-    if (captured != EMPTY) {
+    /* special moves: captures, pawn double push, en-passant, castling.
+     * All these moves are exclusive (so we use if ... else)
+     */
+    if (captured != EMPTY) {                      /* capture: remove piece */
         pos->clock_50 = 0;
         bug_on(pos->board[to] == EMPTY || COLOR(pos->captured) != them);
         key ^= zobrist_pieces[captured][to];
-        pos_clr_sq(pos, to);                      /* clear square */
-    } else if (is_castle(move)) {                 /* handle rook move */
-        square_t rookfrom, rookto;
-        if (to > from) {
-            rookfrom = sq_rel(H1, us);
-            rookto = sq_rel(F1, us);
-        } else {
-            rookfrom = sq_rel(A1, us);
-            rookto = sq_rel(D1, us);
-        }
-        key ^= zobrist_pieces[pos->board[rookfrom]][rookto] ^
-            zobrist_pieces[pos->board[rookfrom]][rookfrom];
-        pos_set_sq(pos, rookto, pos->board[rookfrom]);
-        pos_clr_sq(pos, rookfrom);
-        pos->castle = clr_castle(pos->castle, us);
-    } else if (ptype == PAWN) {                   /* pawn non capture or e.p. */
+        pos_clr_sq(pos, to);
+    } else if (ptype == PAWN) {
         pos->clock_50 = 0;
-        if (from + up + up == to) {               /* if pawn double push, set e.p. */
+        if (from + up + up == to) {               /* double push: set e.p. */
             square_t ep = from + up;
+            /* check that opponent can e.p. */
             if (bb_pawn_attacks[us][ep] & pos->bb[them][PAWN]) {
                 pos->en_passant = ep;
                 key ^= zobrist_ep[EP_ZOBRIST_IDX(pos->en_passant)];
             }
-        } else if (is_enpassant(move)) {          /* clear grabbed pawn */
+        } else if (is_enpassant(move)) {          /* e.p.: clear grabbed pawn */
             square_t grabbed = to - up;
             piece_t pc = pos->board[grabbed];
             key ^= zobrist_pieces[pc][grabbed];
             pos_clr_sq(pos, grabbed);
         }
+    } else if (is_castle(move)) {                 /* castling: handle rook move */
+        square_t rookfrom, rookto;
+        if (to > from) {                          /* o-o */
+            rookfrom = to + 1;
+            rookto = to - 1;
+        } else {                                  /* o-o-o */
+            rookfrom = to - 2;
+            rookto = to + 1;
+        }
+        key ^= zobrist_pieces[pos->board[rookfrom]][rookto] ^
+            zobrist_pieces[pos->board[rookfrom]][rookfrom];
+        pos_set_sq(pos, rookto, pos->board[rookfrom]);
+        pos_clr_sq(pos, rookfrom);
+    }
+
+    /* common part: captures and non captures
+     * We could improve slightly, as en-passant and double push will make
+     * the following tests.
+     * Improvement could be to add promotion test and king test twice above.
+     * (in capture and after ).
+     */
+    if (is_promotion(move)) {
+        bug_on(sq_rank(to) != sq_rel_rank(RANK_8, us));
+        new_piece = MAKE_PIECE(move_promoted(move), us);
     }
 
     key ^= zobrist_pieces[piece][from] ^ zobrist_pieces[new_piece][to];
-    pos_clr_sq(pos, from);                        /* clear "from" and set "to" */
+    pos_clr_sq(pos, from);
     pos_set_sq(pos, to, new_piece);
 
-    if (ptype == KING)
-        pos->king[us] = to;
-
-    /* update castling flags
-     * As we always consider flags are valid, we :
-     * - adjust our flags if relative from is "E1", "A1", H1"
-     * - adjust opp flags if relative to   if "A8", H8"
-     */
-    if (can_castle(pos->castle, us)) {            /* do we save time with this test ? */
-        square_t rel_e1 = sq_rel(E1, us);
-        square_t rel_a1 = sq_rel(A1, us);
-        square_t rel_h1 = sq_rel(H1, us);
-        if (from == rel_e1)
-            pos->castle = clr_castle(pos->castle, us);
-        else if (from == rel_a1)
-            pos->castle = clr_ooo(pos->castle, us);
-        else if (from == rel_h1)
-            pos->castle = clr_oo(pos->castle, us);
-    }
-    if (can_castle(pos->castle, them)) {
-        square_t rel_a8 = sq_rel(A8, us);
-        square_t rel_h8 = sq_rel(H8, us);
-        if (to == rel_a8)
-            pos->castle = clr_ooo(pos->castle, them);
-        else if (to == rel_h8)
-            pos->castle = clr_oo(pos->castle, them);
-    }
-
-    /* update castling rights key */
+    /* update castling flags */
     key ^= zobrist_castling[pos->castle];
+    pos->castle &= sq_castle[from] & sq_castle[to];
+    key ^= zobrist_castling[pos->castle];
+
+    //if (ptype == KING) {
+    //    pos->king[us] = to;
+    //} else
+    /* do this always, cheaper than test on K move */
+    pos->king[us] = ctz64(pos->bb[us][KING]);
 
     pos->key = key;
 
@@ -180,33 +182,43 @@ pos_t *move_undo(pos_t *pos, const move_t move, const state_t *state)
     piece_t piece = pos->board[to];
     int up = sq_up(them);
 
+    /* common part: captures and non captures
+     * We could improve slightly, as en-passant and double push will make
+     * the following tests.
+     * Improvement could be to add promotion test and king test twice above.
+     * (in capture and after ).
+     */
     if (is_promotion(move))
         piece = MAKE_PIECE(PAWN, us);
 
-    pos_clr_sq(pos, to);                          /* always clear "to" ... */
-    pos_set_sq(pos, from, piece);                 /* ... and set "from" */
+    pos_clr_sq(pos, to);
+    pos_set_sq(pos, from, piece);
 
-    if (PIECE(piece) == KING)
-        pos->king[us] = from;
-
-    if (pos->captured != EMPTY) {
-        pos_set_sq(pos, to, pos->captured);       /* restore captured piece */
-    } else if (is_castle(move)) {                 /* make reverse rook move */
+    /* special moves: capture, king (+ castling), en-passant
+     */
+    if (pos->captured != EMPTY) {                 /* captured: restore piece */
+        pos_set_sq(pos, to, pos->captured);
+    } else if (is_castle(move)) {                 /* castle: rook move */
         square_t rookfrom, rookto;
-        if (to > from) {
-            rookfrom = sq_rel(F1, us);
-            rookto = sq_rel(H1, us);
-        } else {
-            rookfrom = sq_rel(D1, us);
-            rookto = sq_rel(A1, us);
+        if (to > from) {                          /* o-o */
+            rookfrom = to - 1;
+            rookto = to + 1;
+        } else {                                  /* o-o-o */
+            rookfrom = to + 1;
+            rookto = to - 2;
         }
         pos_set_sq(pos, rookto, pos->board[rookfrom]);
         pos_clr_sq(pos, rookfrom);
-    } else if (is_enpassant(move)) {              /* restore grabbed pawn */
+    } else if (is_enpassant(move)) {              /* e.p.: restore grabbed pawn */
         square_t grabbed = to + up;
         pos_set_sq(pos, grabbed, MAKE_PIECE(PAWN, them));
     }
 
+    /* do this always, cheaper than test on K move */
+    pos->king[us] = ctz64(pos->bb[us][KING]);
+    //if (PIECE(piece) == KING)
+    //    pos->king[us] = from;
+
     pos->state = *state;                          /* restore irreversible changes */
     pos->turn = us;
     return pos;
@@ -284,7 +296,7 @@ pos_t *move_do_alt(pos_t *pos, const move_t move, state_t *state)
     }
 
     key ^= zobrist_pieces[piece][from] ^ zobrist_pieces[new_piece][to];
-    pos_clr_sq(pos, from);                        /* clear "from" and set "to" */
+    pos_clr_sq(pos, from);
     pos_set_sq(pos, to, new_piece);
 
     if (ptype == KING)

src/move-do.h

@@ -14,8 +14,11 @@
 #ifndef MOVE_DO_H
 #define MOVE_DO_H
 
+#include "chessdefs.h"
 #include "position.h"
 
+extern castle_rights_t sq_castle[64];             /* to adjust castling rights */
+
 pos_t *move_do(pos_t *pos, const move_t move, state_t *state);
 pos_t *move_undo(pos_t *pos, const move_t move, const state_t *state);
 

src/position.h

@@ -88,8 +88,8 @@ static __always_inline void pos_set_sq(pos_t *pos, square_t square, piece_t piec
     bug_on(pos->board[square] != EMPTY);
 
     pos->board[square] = piece;
-    pos->bb[color][type] |= BIT(square);
-    pos->bb[color][ALL_PIECES] |= BIT(square);
+    pos->bb[color][type] ^= BIT(square);
+    pos->bb[color][ALL_PIECES] ^= BIT(square);
     //if (type == KING)
     //    pos->king[color] = square;
 
@@ -114,12 +114,38 @@ static __always_inline void pos_clr_sq(pos_t *pos, square_t square)
     //pos->key ^= zobrist_pieces[piece][square];
 
     pos->board[square] = EMPTY;
-    pos->bb[color][type] &= ~BIT(square);
-    pos->bb[color][ALL_PIECES] &= ~BIT(square);
+    pos->bb[color][type] ^= BIT(square);
+    pos->bb[color][ALL_PIECES] ^= BIT(square);
     //if (type == KING)
     //    pos->king[color] = SQUARE_NONE;
 }
 
+/**
+ * pos_mv_sq - move a piece from a square to another one.
+ * @pos: position
+ * @from: source square
+ * @to: dest square
+ *
+ * Both position board and bitboards are modified.
+ * This function is equivalent to:
+ *      pos_set_sq(pos, to, pos->board[from]);
+ *      pos_clr_sq(pos, from);
+ */
+static __always_inline void pos_mv_sq(pos_t *pos, square_t from, square_t to)
+{
+    piece_t pc = pos->board[from];
+    piece_type_t pt = PIECE(pc);
+    color_t color = COLOR(pc);
+    bitboard_t bb_squares = BIT(from) | BIT(to);
+
+    bug_on(pc == EMPTY || pos->board[to] != EMPTY);
+
+    pos->board[from] = EMPTY;
+    pos->board[to] = pc;
+    pos->bb[color][pt] ^= bb_squares;
+    pos->bb[color][ALL_PIECES] ^= bb_squares;
+}
+
 /**
  * pos_occ() - get position occupation (all pieces)
  * @pos: position

test/common-test.h

@@ -198,6 +198,10 @@ struct fentest {
     /*****************************************************
      * tests from talkchess                              *
      *****************************************************/
+    { __LINE__, FEN | MOVEGEN | MOVEDO | PERFT,
+      "https://talkchess.com/viewtopic.php?t=21343",
+      "8/k7/3p4/p2P1p2/P2P1P2/8/8/K7 w - - 0 1"
+    },
     { __LINE__, FEN | MOVEGEN | MOVEDO | PERFT,
       "https://www.talkchess.com/forum3/viewtopic.php?f=7&t=71379",
       "8/6kR/8/8/8/bq6/1rqqqqqq/K1nqnbrq b - - 0 1"
@@ -447,7 +451,12 @@ struct fentest {
 
 static int fentest_cur = -1;
 
-static char *next_fen(uint module)
+static __unused void restart_fen(void)
+{
+    fentest_cur = -1;
+}
+
+static __unused char *next_fen(uint module)
 {
     fentest_cur++;
 

test/movedo-test.c

@@ -52,8 +52,8 @@ int main(int __unused ac, __unused char**av)
         for (move = movelist.move; move < last; ++move) {
             //pos_print(pos);
             //printf("i=%d j=%d  turn=%d move=[%s]\n", i, j, pos->turn,
-            //       move_str(movebuf, move, 0));
-            //move_p
+            //       move_to_str(movebuf, *move, 0));
+
             move_do(pos, *move, &state);
             //pos_print(pos);
             //fflush(stdout);
@@ -67,6 +67,7 @@ int main(int __unused ac, __unused char**av)
             move_undo(pos, *move, &state);
             pos->state = state;
             if (!pos_ok(pos, false)) {
+                //pos_print(pos);
                 printf("*** fen %d [%s] move %d [%s] invalid position after move_undo\n",
                        test_line, fen, j, movebuf);
                 exit(0);

test/perft-test.c

@@ -242,7 +242,9 @@ static int usage(char *prg)
     fprintf(stderr, "Usage: %s [-cms][-d depth] [-p version] [-t size:\n", prg);
     fprintf(stderr, "\t-c:         do *not* print FEN comments\n");
     fprintf(stderr, "\t-d depth:   perft depth (default: 6)\n");
+    fprintf(stderr, "\t-l line:    start from 'line' test\n");
     fprintf(stderr, "\t-m:         print moves details\n");
+    fprintf(stderr, "\t-n number:  do 'number' tests (default: all)\n");
     fprintf(stderr, "\t-s:         use Stockfish to validate perft result\n");
     fprintf(stderr, "\t-t size:    Transposition Table size (Mb). Default: 32\n");
     fprintf(stderr,
@@ -252,7 +254,7 @@ static int usage(char *prg)
 
 int main(int ac, char**av)
 {
-    int curtest = 0;
+    int curtest = 0, totalfen = 0;
     u64 sf_count = 0, my_count;
     bool comment = true, sf_run = false, divide = false;
     char *fen;
@@ -262,6 +264,8 @@ int main(int ac, char**av)
     FILE *outfd = NULL;
     s64 ms, lps;
     int opt, depth = 6, run = 3, tt, newtt = HASH_SIZE_DEFAULT;
+    int startline = 1, ntests = INT_MAX;
+
     struct {
         s64 count, countskipped, ms;
         s64 minlps, maxlps;
@@ -273,7 +277,8 @@ int main(int ac, char**av)
         { .minlps = LONG_MAX },
     };
 
-    while ((opt = getopt(ac, av, "cd:mp:st:")) != -1) {
+    printf("Perft " VERSION "\n");
+    while ((opt = getopt(ac, av, "cd:l:mn:p:st:")) != -1) {
         switch (opt) {
             case 'c':
                 comment = false;
@@ -283,9 +288,15 @@ int main(int ac, char**av)
                 if (depth <= 0)
                     depth = 6;
                 break;
+            case 'l':
+                startline = atoi(optarg);
+                break;
             case 'm':
                 divide = true;
                 break;
+            case 'n':
+                ntests = atoi(optarg);
+                break;
             case 'p':
                 run = atoi(optarg);
                 break;
@@ -325,14 +336,23 @@ int main(int ac, char**av)
     if (sf_run)
         outfd = open_stockfish();
 
+    /* count total fen tests we will do */
+    while (next_fen(PERFT | MOVEDO))
+        totalfen++;
+    restart_fen();
+
     CLOCK_DEFINE(clock, CLOCK_MONOTONIC);
     while ((fen = next_fen(PERFT | MOVEDO))) {
+        if (cur_line() < startline)
+            continue;
+        if (curtest >= ntests)
+            break;
         if (!(fenpos = fen2pos(pos, fen))) {
             printf("wrong fen line:%d fen:%s\n\n", cur_line(), fen);
             continue;
         }
         curtest++;
-        printf("test:%d line:%d fen:%s\n", curtest, cur_line(), fen);
+        printf("test:%d/%d line:%d fen:%s\n", curtest, totalfen, cur_line(), fen);
         if (comment)
             printf("\t\"%s\"\n",
                    *cur_comment()? cur_comment(): "no test desc");