1 ----------------------------------------------------------------------
\r
2 Bonanza 6.0 - Client Source Code
\r
3 Kunihito Hoki, May 2011
\r
4 ----------------------------------------------------------------------
\r
10 Bonanza is a state-of-the art computer shogi engine which runs on
\r
11 Windows and Linux machines, and this directory contains a
\r
12 platform-independent source code.
\r
14 This source code is distributed with a hope that it will be useful in
\r
15 addition to the main part of the shogi engine. This program includes
\r
16 many useful functions such as manipulating a shogi board, reading and
\r
17 writing a CSA record file, speaking of a CSA protocol with a socket
\r
18 communication, and controlling time, and etc.. I believe this program
\r
19 can be a good starting point if you are interested in making shogi
\r
22 One main feature of this program is that it employs a brute-force
\r
23 search method together with bitboard techniques as many chess programs
\r
24 do. Another notable feature is a machine learning of shogi evaluation
\r
25 functions. The details of the learning algorithm (aka Bonanza method)
\r
26 were already presented [1], and this source code provides an example
\r
27 of implementation of the learning method.
\r
29 I admit that some parts of the source code are cryptic, e.g. codes in
\r
30 "mate1ply.c". I hope that I will have some time to make a quality
\r
31 documentation and comments on the program code, or someone else could
\r
32 decrypt my program and provide a documentation.
\r
34 Any comments or suggestions are welcome [2].
\r
36 [1] K. Hoki, "Optimal control of minimax search results to learn
\r
37 positional evaluation", Game Programming Workshop, Hakone, Japan
\r
40 [2] Contact to "bonanza_query [at] hotmail.com".
\r
46 This program is protected by copyright. Without a specific
\r
47 permission, any means of commercial applications are prohibited.
\r
48 Furthermore, this program is distributed without any warranty.
\r
49 Within these limits, you can use, redistribute, and/or modify it.
\r
57 Search() function is modified to ignore moves which receive negative
\r
58 values from Static Exchange Evaluation (SEE) at frontier and
\r
59 pre-frontier nodes. Self-play experiments showed that this modification
\r
60 makes Bonanza slightly stronger. Similar results were also observed in
\r
61 experiments using a shogi program YSS,
\r
62 http://www32.ocn.ne.jp/~yss/horizon.txt. Moreover, the branch cut by
\r
63 means of SEE at frontier nodes can also be found in a chess program,
\r
66 Now Bonanza does Late Move Reduction (LMR) at root nodes, and recursive
\r
67 iterative deepening more often. Self-play experiments showed that these
\r
68 modifications make Bonanza slightly stronger.
\r
70 DFPN searcher is added to the set of Bonanza source codes (dfpn.c,
\r
71 dfpn.h, and dfpnhash.c). You can use the searcher by using 'dfpn'
\r
72 command with a compile option -DDFPN turned on. To make use of the DFPN
\r
73 searcher in ordinary search, use 'dfpn_client' command with a compile
\r
74 option -DDFPN_CLIENT turned on. To provide a proper service to the DFPN
\r
75 client, you also need to use a server script 'src/server/dfpn_server.pl'
\r
76 with one or more DFPN worker(s) connected with 'dfpn connect' command.
\r
77 I am thankful to Dr. A. Nagai for his kind and fruitful advice.
\r
79 Now Bonanza can be a client of parallel searches using cluster computing
\r
80 environment with mnj command. 'src/server/parallel_server.pl' is a
\r
81 server script of root-node parallelizations, and the server can be a
\r
82 client of majority voting cluster computation.
\r
84 Some techniques which increase computational efficiency of Bonanza are
\r
85 reported by E. Ito, http://aleag.cocolog-nifty.com/. In his website,
\r
86 two ideas are shown, i.e., (i) speed up of the computation of the static
\r
87 evaluation function, and (2) use of SSE instructions of Intel processors
\r
88 for bitboard operations. By adopting his ideas to Bonanza, NPS value
\r
89 nearly doubled. This is amazing. To use SSE instructions, specify -DSSE2
\r
90 or -DSSE4 compiler option. I am thankful to Mr. E. Ito for his kind and
\r
93 -DINANIWA_SHIFT compilation flag is supplied to enable an Inaniwa-strategy detection
\r
95 There are many other modifications and bug fixes.
\r
100 - Some results from Y. Sato's experiments with Bonanza inspired me
\r
101 to use the history table in LMR in Shogi, as Fruit did in chess. By
\r
102 doing so, some improvement in performance of the tree search is
\r
105 - From private communications with Yaneurao, it turned out that the
\r
106 3-ply-mate detections in 'mate3.c' can search game trees more
\r
107 efficiently by using several heuristic pruning techniques. Now
\r
108 'mate3.c' adopts some of these techniques.
\r
110 - E. Ito kindly reported two bugs in the source codes. One is in
\r
111 hash_store(), and the other is in ehash_probe() and
\r
112 ehash_store(). Now these are fixed.
\r
114 - The structure of history table is modified from 'square-to' and
\r
115 'square-from' method to an exact match method by using the perfect
\r
116 hash technique. The hash function of moves in 'phash.c' is generated
\r
117 by codes at http://burtleburtle.net/bob/hash/perfect.html.
\r
119 - A command 'mnj' is added to connect to the cluster-computing server
\r
122 - In client_next_game(), the way of dealing with REJECT message from a
\r
123 CSA Shogi server is revised. Now Bonanza connects to the server again
\r
124 when the game is rejected by an opponent.
\r
128 - The owner of 'LS3600 Blog Webpage' pointed out that there were two
\r
129 serious bugs. Thank you very much! According to his indication,
\r
130 is_hand_eq_supe() in 'utility.c' and read_CSA_line() in 'csa.c' are
\r
132 - In MPV code in 'searchr.c', a test of 'root_abort' flag had been
\r
133 forgotten. Now the flag is tested after every call of search().
\r
134 - 'lan.txt' is added to 'src/executable/' directory. This is an
\r
135 example of an input sequence to connect to CSA Shogi server.
\r
136 - 'book.bin' now has smaller moves and positions than previous one does.
\r
137 Also, 'book_anti.csa' is added to 'src/executable/' directory. This
\r
138 is an example of bad moves which apear in records of human experts.
\r
139 - 'Legal Notices' in this document is corrected.
\r
142 - In 'Makefile' and 'Makefile.vs', targets which require profile-guided
\r
143 optimization are removed. Furthermore, an option, which controls
\r
144 optimization, has been reverted from the aggressive flag, -O3, to a
\r
145 moderate one, '-O2'. These modifications were necessary for avoiding
\r
146 abnormal terminations of the program.
\r
147 - In 'ini.c', the attribute of POSIX thread in a global-variable
\r
148 'pthread_attr' is set to 'detach-state'. Because threads will never
\r
149 join with any other threads in this program, the thread should be
\r
150 created in the detached state to free system resources.
\r
151 - In 'ini.c', the default size of the transposition table is lifted
\r
152 from 12MByte to 48MByte.
\r
153 - In 'iterate.c", probing the opening book is avoided when the move
\r
154 history of a current game has repetitions.
\r
155 - In 'shogi.h', the margins of futility pruning are increased in
\r
156 accord with new positional evaluation of the feature vector in
\r
158 - The quality of an opening book, 'winbin/book.bin', has been improved
\r
159 at the expense of quantity.
\r
161 Version 4.1.0 and 4.1.1 (26 Apr 2009)
\r
162 - In 'Makefile' and 'Makefile.vs', some options of Intel C compiler
\r
163 are modified. Here, agressive optimization '-O3' is substituted for
\r
164 the default '-O2', pthreads support '-pthread' is substituted for
\r
165 '-lpthread', and an obsolete '-static-libcxx' is removed.
\r
166 - In 'Makefile', the conformance of GNU C and Intel C compilers are
\r
167 set to GNU extensions of ISO C99 by setting '-std=gnu99' because a
\r
168 POSIX function 'strtok_r()' is not in the C99 standard library.
\r
169 - In 'Makefile', targets for GNU C with gprof and profile-guided
\r
170 optimization of GNU C are removed.
\r
171 - In 'shogi.h', inline assemblies and intrinsics are used on x86-64 as
\r
172 well as x86. To detect the targets, pre-defined macros '__i386__'
\r
173 and '__x86_64__' are examined.
\r
174 - In 'evaluate.c', the evaluation function looks up the table
\r
175 'stand_pat[ply]' to see if the position have been evaluated since
\r
176 the previous move is made by 'make_move_[bw]()' functions. Also, the
\r
177 evaluation function probes the hash table 'ehash_tbl[]' to avoid
\r
178 re-evaluation of the same position.
\r
179 - In 'io.c', an immediate value 0 is substituted for 'fileno(stdin)'
\r
180 because the POSIX function 'fileno()' is not part of ANSI C. POSIX
\r
181 requires that the file descriptor associated with 'stdin' be 0.
\r
182 - In 'iterate.c', a criterion for aging of transposition-table, i.e.,
\r
183 increment of a global variable 'trans_table_age', is lifted from 7%
\r
184 to 9% of saturation ratio of the table.
\r
185 - In 'learn1.c', the macro 'SEARCH_DEPTH' is set to 2. The macro
\r
186 specifies the depth threashold of searches for finding all of
\r
187 principle variations of positions in a set of games.
\r
188 - In 'learn1.c', a step size of increment/decrement of feature vectors
\r
189 is fixed to 1 in 'learn_parse2()' function.
\r
190 - In 'hash.c', when a node can be pruned by using a hash value based on
\r
191 futility pruning, a return value of the node is set to beta.
\r
192 - In 'rand.c', some codes of initialization of PRNG variables are
\r
193 moved to 'ini_genrand()' from 'ini()' function.
\r
194 - In 'sckt.c', a function 'send_crlf()' is removed.
\r
195 - In 'search.c', now 'search()' calls 'search_quies()' and returns if
\r
196 a search depth reaches a threashold. So that we can call 'search()'
\r
197 function anytime, and this makes source codes simple.
\r
198 - In 'search.c', the static evaluation value is evaluated and used to
\r
199 see if the late-move reduction is applicable or not. The evaluation
\r
200 value is also used by the futility pruning.
\r
201 - In 'search.c', the futility pruning is not applied if the node is in
\r
202 check or the move is a check.
\r
203 - In 'search.c', now Bonanza sends the keep-alive command '0x0a' to
\r
204 the server in 'detect_signals()'.
\r
205 - In 'time.c', 'set_seach_limit_time()' is simplified.
\r
206 - In 'time.c', the second argument of 'gettimeofday()' is set to 'NULL'.
\r
208 Version 4.0.4 (2 Feb 2009)
\r
209 - An error of GCC inline assembly for spinlock in "thread.c" is fixed.
\r
210 - In Windows OS, Bonanza now opens all streams with file sharing by
\r
211 using "_SH_DENYNO" constant in "io.c".
\r
212 - GCC built-in functions are substituted for GCC inline assemblies for
\r
213 bit-scan operations in "bitop.h". Furthermore, "bitop.h" is removed,
\r
214 and some of macros in the header are integrated into "shogi.h".
\r
216 Version 4.0.3 (Jan 2008)
\r
222 Here is a list of files you can find in this directory.
\r
225 - param.h piece values
\r
226 - shogi.h main header
\r
227 - bitop.h bit operations
\r
228 - dfpn.h header of DFPN codes
\r
231 - main.c main function of C program
\r
232 - data.c definition of global variables
\r
233 - ini.c initializations
\r
234 - rand.c pseudo random number generator
\r
235 - time.c time functions
\r
236 - bitop.c bit operations
\r
237 - utility.c misc. functions
\r
240 - proce.c input procedures
\r
241 - csa.c csa file format I/O
\r
243 - sckt.c Socket communications
\r
245 bitboard manipulations
\r
246 - attack.c piece attacks
\r
247 - genchk.c move generation (checks)
\r
248 - genevasn.c move generation (evasions)
\r
249 - gendrop.c move generation (drops)
\r
250 - gennocap.c move generation (non-captures)
\r
251 - gencap.c move generation (captures)
\r
252 - movgenex.c move generation (inferior moves)
\r
253 - makemove.c make moves
\r
254 - unmake.c unmake move
\r
255 - mate1ply.c 1-ply mate detection
\r
256 - debug.c examine bitboard validity
\r
259 - iterate.c iterative deepning search at root node
\r
260 - searchr.c alpha-beta search at root node
\r
261 - search.c alpha-beta search
\r
262 - next.c obtains next move
\r
263 - quiesrch.c quiescence search
\r
264 - evaluate.c static eveluation function
\r
265 - evaldiff.c easy and fast evaluation function
\r
266 - swap.c static exchange evaluation
\r
267 - hash.c transposition table
\r
268 - phash.c perfect hash function of moves
\r
269 - thread.c thread-level parallelization
\r
270 - root.c root move genelation and shallow min-max search
\r
271 - mate3.c 3-ply mate detection
\r
272 - ponder.c pondering
\r
273 - book.c creates and probes opening book
\r
274 - problem.c auto problem solver
\r
275 - valid.c examine move validity
\r
278 - dfpn.c DFPN search
\r
279 - dfpnhash.c Hashing for DFPN search
\r
281 optimal control of min-max search
\r
282 - learn1.c main functions
\r
283 - learn2.c feture vector manipuration
\r
286 - bonanza.txt which now you are looking at
\r
287 - Makefile makefile for gnu make.exe
\r
288 - Makefile.vs makefile for Microsoft nmake.exe
\r
289 - bonanza.ico icon file for windows
\r
290 - bonanza.rc resource-definition file for windows
\r
291 - lan.txt example of input sequence to connect CSA Shogi server
\r
292 - book_anti.csa example of a set of bad moves which apear in records
\r
293 of human exparts. This is used by 'book create' command.
\r
295 4. How to build Bonanza
\r
296 -----------------------
\r
298 You can build Bonanza by means of GNU Make on Linux or Microsoft NMAKE
\r
299 on Windows. Here is some examples:
\r
302 > make -f Makefile gcc
\r
304 - Intel C++ Compiler on Linux
\r
305 > make -f Makefile icc
\r
307 - Microsoft C/C++ Compiler on Windows
\r
308 > nmake -f Makefile.vs cl
\r
310 - Intel C++ Compiler on Windows
\r
311 > nmake -f Makefile.vs icl
\r
313 The C source codes are written by using ANSI C plus a small number of
\r
314 new features in ISO C99. Therefore, I think this can be easily built
\r
315 in many platforms without much effort.
\r
317 It may be necessary to define some macros in Makefile or
\r
318 Makefile.vs. The macros are:
\r
320 - NDEBUG (DEBUG) builds release (debug) version of Bonanza
\r
322 - MINIMUM disables some auxiliary functions that are not
\r
323 necessary to play a game, e.g., book composition
\r
324 and optimization of evaluation functions
\r
326 - TLP enables thread-level parallel search
\r
328 - MPV enables multi-PV search
\r
330 - CSA_LAN enables Bonanza to communicate by CSA Shogi TCP/IP
\r
333 - DEKUNOBOU enables dekunobou interface (available only for
\r
336 - CSASHOGI builds an engine for CSA Shogi (available only for
\r
339 - NO_LOGGING suppresses dumping log files
\r
341 - HAVE_SSE2 use SSE2 instructions for speed
\r
343 - HAVE_SSE4 use SSE2 and SSE4.1 instructions for speed
\r
345 - MNJ_LAN enables a client-mode of cluster searches
\r
347 - INANIWA_SHIFT enables an Inaniwa strategy detection
\r
349 - DFPN build the DFPN worker of mate-problems server
\r
351 - DFPN_CLIENT enables the client-mode of mate-problem server
\r
354 Bonanza is an application that does not provide graphical user
\r
355 interface. If you could build "bonanza.exe" properly without CSASHOGI
\r
356 macro, it shows a prompt "Black 1>" when you execute it at a computer
\r
359 Bonanza uses three binary files: a feature vector of static evaluation
\r
360 function "fv.bin", an opening book "book.bin", and a
\r
361 position-learning database "hash.bin". You can find these in "winbin/"
\r
362 directory. Without the NO_LOGGING option, Bonanza must find "log/"
\r
363 directory to dump log files.
\r
371 These commands enable (on) or disable (off) a beep when Bonanza
\r
372 makes a move. The default is on.
\r
376 These commands enable (on) or disable (off) to probe the opening
\r
377 book, "./book.bin". The default is on.
\r
381 When the command with "narrow" is used, Bonanza selects a book
\r
382 move from a small set of opening moves. The default is "wide". The
\r
383 narrowing of the opening moves is useful if you want Bonanza
\r
384 choose a common opening line.
\r
387 This command creates the opening book file, "./book.bin", by using
\r
388 numerous experts' games in a single CSA record file, "./book.csa".
\r
389 It also uses another CSA record file, "book_anti.csa", where you
\r
390 can register bad moves that may appear in the experts' games at
\r
391 the last moves in the record file. Here is the example:
\r
393 ----------------------------------------
\r
394 PI, +, +6978KI, %TORYO
\r
396 PI, +, +6978KI, -8384FU, %TORYO
\r
398 PI, +, +7776FU, -4132KI, %TORYO
\r
400 PI, +, +7776FU, -4132KI, +2726FU, %TORYO
\r
401 ----------------------------------------
\r
403 This command becomes effective when MINIMUM macro is not defined
\r
406 - connect 'addr' 'port' 'id' 'passwd' ['ngame']
\r
407 This command connects Bonanza to a shogi server by using the CSA
\r
408 protocol. The first four arguments specify the network address,
\r
409 port number, user ID, and password, respectively. The last
\r
410 argument limits a number of games that will be played by Bonanza.
\r
411 This command becomes effective when CSA_LAN macro is defined in
\r
415 Bonanza does DFPN search at a current position.
\r
418 This command is used to initialize the transposition table of
\r
419 DFPN search and set the size of the table to 2^'num'. This
\r
420 command becomes effective when DFPN macro is defined in the
\r
423 - dfpn connect 'hostname' 'port#' 'ID'
\r
424 This command is used to connect to the server script
\r
425 dfpn_server.pl as a worker. This command becomes effective when
\r
426 DFPN macro is defined in the Makefile.
\r
428 - dfpn_client 'hostname' 'port#'
\r
429 This command is used to connect to the server script
\r
430 dfpn_server.pl as a client. With this, a root and its child
\r
431 posittions are examined by the DFPN worker(s). This command
\r
432 becomes effective when DFPN_CLIENT macro is defined in the
\r
436 This command prints the shogi board. If you want to flip the
\r
437 board, set 'num' to 2. If not, set it to 1.
\r
440 Bonanza makes a prompt reply while thinking as soon as this
\r
444 This command is used to initialize the transposition table and
\r
445 set the size of the table to 2^'num'.
\r
447 - hash learn create
\r
448 This command is used to make a zero-filled position-lerning
\r
449 database, "hash.bin". This command becomes effective when MINIMUM
\r
450 macro is not defined in the Makefile.
\r
454 These commands enable (on) or disable (off) the position learning.
\r
457 - learn 'str' 'steps' ['games' ['iterations' ['num1' ['num2']]]]
\r
458 This command optimizes a feature vector of the static evaluation
\r
459 function by using numorous experts' games in a single CSA record
\r
460 file, "./records.csa". If you want to use a zero-filled vector as
\r
461 an initial guess of the optimization procedure, set 'str' to
\r
462 "ini". If not, set it to "no-ini". The third argument 'games' is a
\r
463 number of games to be read from the record file. If the third
\r
464 argument is negative or omitted, all games are read from the file.
\r
466 The learning method iterates a set of procedures, and the number
\r
467 of iteration can be limited by the fourth argument. It continues
\r
468 as long as the argument is negative. The procedures consist of two
\r
469 parts. The first part reads the record file and creates principal
\r
470 variations by using 'num1' threads. The default value of 'num1' is
\r
471 1. The second part renews the feature vector 'steps' times by using
\r
472 'num2' threads in accord with the principal variations. The default
\r
473 value of 'steps' and 'num2' is 1. Note that each thread in the
\r
474 second procedure uses about 500MByte of the main memory. The two
\r
475 arguments 'num1' and 'num2' become effective when TLP macro is
\r
476 defined in the Makefile. After the procedures, the optimized
\r
477 vector is saved in "./fv.bin". This command become effective when
\r
478 MINIMUM macro is not defined in the Makefile.
\r
480 - limit depth 'num'
\r
481 This command is used to specify a depth, 'num', at which Bonanza
\r
482 ends the iterative deepening search.
\r
484 - limit nodes 'num'
\r
485 When this command is used, Bonanza stops thinking after searched
\r
486 nodes reach to 'num'.
\r
488 - limit time 'minute' 'second' ['depth']
\r
489 This command limits thinking time of Bonanza. It tries to make
\r
490 each move by consuming the time 'minute'. When the time is spent
\r
491 all, it makes each move in 'second'. The last argument 'depth' can
\r
492 be used if you want Bonanza to stop thinking after the iterative
\r
493 deepening searches reach sufficient depth.
\r
495 - limit time extendable
\r
496 - limit time strict
\r
497 The command, "limit time extendable", allows Bonanza to think
\r
498 longer than the time limited by the previous command if it wishes
\r
499 to. The default is "strict".
\r
501 - mnj 'sd' 'seed' 'addr' 'port' 'id' 'factor' 'stable_depth'
\r
502 This command connects Bonanza to the majority or parallel server in
\r
503 src/server/. The first two integers specify the standard
\r
504 deviation and initial seed of pseudo-random numbers which are
\r
505 added to the static evaluation function. Experiments suggested
\r
506 that an appropriate value for the standard deviation is 15. Note
\r
507 that all clients should use different seeds. The following three
\r
508 arguments are network address, port number, user ID,
\r
509 respectively. You can specify factor as a voting weight. Also
\r
510 stable_depth is useful to detect opening positions. Here, when
\r
511 Bonanza reaches the specified depth, then this is reported to the
\r
512 server to reduce thinking time. This command becomes effective
\r
513 when MNJ_LAN macro is defined in the Makefile.
\r
516 Bonanza makes a move of 'str'. If the argument is omitted, Bonanza
\r
517 thinks of its next move by itself.
\r
520 - mpv width 'threshold'
\r
521 These commands control the number of root moves, 'nroot', to
\r
522 constitute principal variations. The default number is 1. A root
\r
523 move that yields a smaller value than the best value by 'threshold'
\r
524 is neglected. The default threshold is about 200. These commands
\r
525 become effective when MPV macro is defined in the Makefile.
\r
528 This command initializes the shogi board. The argument 'str'
\r
529 controls an initial configuration of the board. If you want to
\r
530 play a no-handicapped game, set 'str' to "PI" and this is the
\r
531 default value. In a handicapped game, specify squares and pieces
\r
532 to drop, e.g. "new PI82HI22KA" or "new PI19KY".
\r
536 The command "peek on (off)" enables (disables) peeks at a buffer
\r
537 of the standard input file while Bonanza is thinking. The default
\r
538 is on. This command is useful when you want to process a set of
\r
539 commands as "> ./bonanza.exe < infile".
\r
542 Prompt Bonanza to print "pong".
\r
546 The command "ponder on (off)" enables (disables) thinks on the
\r
547 opponent's time. The default is on.
\r
550 This command is used to solve problems in "./problem.csa". Here
\r
551 is an example of the problem file.
\r
553 -----------------------------
\r
555 P1-KY-KE-OU-KI * * * * -KY
\r
556 P2 * * * * * -KI * * *
\r
557 P3 * * -FU-GI-FU * -KE * -KA
\r
558 P4-FU * * -FU-GI-FU-HI * -FU
\r
559 P5 * * * * * * * -FU+KY
\r
560 P6+FU+KA+FU+FU+GI+FU+KI * *
\r
561 P7 * +FU * * +FU * * * *
\r
562 P8 * +OU+KI+GI * * +HI * *
\r
563 P9+KY+KE * * * * * +KE *
\r
568 $ANSWER:+0087KY:+0088KY
\r
569 P1-OU-KE * * * * * +GI *
\r
570 P2-KY-KI * * * * * * *
\r
571 P3-FU-HI * -KI * * -GI * *
\r
572 P4 * * -KE * * * * * -FU
\r
573 P5 * +GI * -FU-FU-FU-FU-FU *
\r
574 P6+FU+HI-FU * * * * * *
\r
575 P7 * * * +FU * * * * +FU
\r
576 P8 * * +OU+KI+KI * * * *
\r
577 P9+KY+KE * * * * * +KE+KY
\r
578 P+00KA00GI00KY00FU00FU
\r
579 P-00KA00FU00FU00FU00FU00FU
\r
581 -----------------------------
\r
583 The argument 'num' specifies the number of problems to solve.
\r
586 The quit command and EOF character will exit Bonanza.
\r
588 - read 'filename' [(t|nil) ['num']]
\r
589 This command is used to read a CSA record 'filename' up to 'num'
\r
590 moves. Set the second argument to "nil" when you want to ignore
\r
591 time information in the record. The default value is "t". Bonanza
\r
592 reads all move sequence if the last argument is neglected. If
\r
593 'filename' is ".", the command reads an ongoing game from the
\r
597 Use this command when you resign a game.
\r
600 This command specifies the threshold to resign. 'num' is a value
\r
601 of the threshold. The default is around 1000.
\r
605 When the command "stress on" is used, last-move shown in shogi
\r
606 board is stressed. The default is on.
\r
608 - time remain 'num1' 'num2'
\r
609 This command tells Bonanza the remaining time. 'num1' ('num2') is
\r
610 the remaining time of black (white) in seconds.
\r
612 - time response 'num'
\r
613 This command specifies a margin to control time. The time margin
\r
614 saves Bonanza from time up due to TCP/IP communication to a server
\r
615 program, sudden disc access, or imperfection of time control of
\r
616 Bonanza. 'num' is the time margin in milli-second. The default
\r
620 This command controls the number of threads to be created when
\r
621 Bonana considers a move to make. The command becomes effective
\r
622 when TLP macro is defined in the Makefile. 'num' is the number of
\r
623 threads. The default value is 1.
\r
626 A line beginning with # causes all characters on that line
\r
630 A move command consists of four digits followed by two
\r
631 capital alphabets, e.g. 7776FU. The first two digits
\r
632 are a starting square and the last two are a target square. The
\r
633 starting square is "OO" if the move is a dorp, e.g. 0087FU. The
\r
634 following two alphabets specify a piece type as the following,
\r
636 FU - pawn (Fuhyo) TO - promoted pawn (Tokin)
\r
637 KY - lance (Kyousha) NY - promoted lance (Narikyo)
\r
638 KE - knight (Keima) NK - promoted knight (Narikei)
\r
639 GI - silver general (Ginsho) NG - promoted silver (Narigin)
\r
640 KI - gold general (Kinsyo)
\r
641 KA - Bishop (Kakugyo) UM - Dragon horse (Ryuma)
\r
642 HI - Rook (Hisha) RY - Dragon king (Ryuo)
\r
645 Here, words in parentheses are romanization of Japanese words.
\r