Bo Haglund, Soren Hein, Bob Richardson

Rev X, 2014-11-16

Latest DLL issue with this description is available at http://www.bahnhof.se/wb758135/

Description of the DLL functions supported in Double Dummy Problem Solver 2.8

Callable functions

The callable functions are all preceded with extern "C" __declspec(dllimport) int __stdcall. The prototypes are available in dll.h in the “include” directory.

Return codes are given at the end.

Not all functions are present in all versions of the DLL. For historical reasons, the function names are not entirely consistent with respect to the input format. Functions accepting binary deals will end on Bin, and those accepting PBN deals will end on PBN in the future. At some point existing function names may be changed as well, so use the new names!

The Basic Functions

The basic functions SolveBoard and SolveBoardPBN each solve a single hand and are thread-safe, making it possible to use them for solving several hands in parallel. The other callable functions use the SolveBoard functions either directly or indirectly.

The Multi-Thread Double Dummy Solver Functions

The double dummy trick values for all 5 × 4 = 20 possible combinations of a hand’s trump strain and declarer hand alternatives are solved by a single call to one of the functions CalcDDtable and CalcDDtablePBN. Threads are allocated per strain in order to save computations.

To obtain better utilization of available threads, the double dummy (DD) tables can be grouped using one of the functions CalcAllTables and CalcAllTablesPBN.

Solving hands can be done much more quickly using one of the multi-thread alternatives for calling SolveBoard. Then a number of hands are grouped for a single call to one of the functions SolveAllBoards, SolveAllChunksBin and SolveAllChunksPBN. The hands are then solved in parallel using the available threads.

The number of threads is automatically configured by DDS on Windows, taking into account the number of processor cores and available memory. The number of threads can be influenced by calling SetMaxThreads. This function should probably always be called on Linux/Mac, with a zero argument for auto-configuration.

Calling FreeMemory causes DDS to give up its dynamically allocated memory.

The PAR Calculation Functions

The PAR calculation functions find the optimal contract(s) assuming open cards and optimal bidding from both sides. In very rare cases it matters which side or hand that starts the bidding, i.e. which side or hand that is first to bid its optimal contract.

Two alternatives are given:

The PAR scores / contracts are calculated separately for each side. In almost all cases the results will be identical for both sides, but in rare cases the result is dependent on which side that “starts the bidding”, i.e. that first finds the bid that is most beneficial for the own side. One example is when both sides can make 1 NT.
The dealer hand is assumed to “start the bidding”.

The presentation of the par score and contracts are given in alternative formats.

The functions Par, SidesPar and DealerPar do the par calculation; their call must be preceded by a function call calculating the double dummy table values.

The functions SidesParBin and DealerParBin provide binary output of the par results, making it easy to tailor-make the output text format. Two such functions, ConvertToSidesTextFormat and ConvertToDealerTextFormat, are included as examples.

It is possible as an option to perform par calculation in CalcAllTables and CalcAllTablesPBN.

The par calculation is executed using a single thread. But the calculation is very fast and its duration is negligible compared to the double dummy calculation duration.

Double Dummy Value Analyser Functions

The functions AnalysePlayBin, AnalysePlayPBN, AnalyseAllPlaysBin and AnalyseAllPlaysPBN take the played cards in a game or games and calculate and present their double dummy values.

Function	Arguments	Format	Comment
SolveBoard	struct deal dl, int target, int solutions, int mode, struct futureTricks *futp, int threadIndex	Binary	The most basic function, solves a single hand from the beginning or from later play
SolveBoardPBN	struct dealPBN dlPBN, int target, int solutions, int mode, struct futureTricks *futp, int threadIndex	PBN	As SolveBoard, but with PBN deal format.
CalcDDtable	struct ddTableDeal tableDeal, struct ddTableResults * tablep	Binary	Solves an initial hand for all possible declarers and denominations (up to 20 combinations)
CalcDDtablePBN	struct ddTableDealPBN tableDealPBN, struct ddTableResults * tablep	PBN	As CalcDDtable, but with PBN deal format.
CalcAllTables	struct ddTableDeals * dealsp, int mode, int trumpFilter[5], struct ddTablesRes resp, struct allParResults presp	Binary	Solves a number of hands in parallel. Multi-threaded.
CalcAllTablesPBN	struct ddTableDealsPBN dealsp, int mode, int trumpFilter[5], struct ddTablesRes resp, struct allParResults * presp	PBN	As CalcAllTables, but with PBN deal format.
SolveAllBoards	struct boardsPBN bop, struct solvedBoards solvedp	PBN	Consider using this instead of the next 3 “Chunk” functions”!
SolveAllChunksBin	struct boards bop, struct solvedBoards solvedp, int chunkSize	Binary	Solves a number of hands in parallel. Multi-threaded.
SolveAllChunks	struct boardsPBN bop, struct solvedBoards solvedp, int chunkSize	PBN	Alias for SolveAllChunksPBN; don’t use!
SolveAllChunksPBN	struct boardsPBN bop, struct solvedBoards solvedp, int chunkSize	PBN	Solves a number of hands in parallel. Multi-threaded.

Par	struct ddTableResults * tablep, struct parResults *presp, int vulnerable	No format	Solves for the par contracts given a DD result table.
DealerPar	struct ddTableResults * tablep, struct parResultsDealer * presp, int dealer, int vulnerable	No format	Similar to Par(), but requires and uses knowledge of the dealer.
DealerParBin	struct ddTableResults * tablep, struct parResultsMaster * presp, int dealer, int vulnerable	Binary	Similar to DealerPar, but with binary output.
ConvertToDealerTextFormat	struct parResultsMaster * pres, char *resp	Text	Example of text output from DealerParBin.
SidesPar	struct ddTableResults tablep, struct parResultsDealer presp, int vulnerable	No format	Par results are given for sides with the DealerPar output format.
SidesParBin	struct ddTableResults * tablep, struct parResultsMaster sidesRes[2], int vulnerable	Binary	Similar to SidesPar, but with binary output.
ConvertToSidesTextFormat	struct parResultsMaster * pres, struct parTextResults * resp	Text	Example of text output from SidesParBin.
CalcPar	struct ddTableDeal tableDeal, int vulnerable, struct ddTableResults * tablep, struct parResults * presp	Binary	Solves for both the DD result table and the par contracts. Is deprecated, use a CalcDDtable function plus Par() instead!
CalcParPBN	struct ddTableDealPBN tableDealPBN, struct ddTableResults * tablep, int vulnerable, struct parResults * presp	PBN	As CalcPar, but with PBN input format. Is deprecated, use a CalcDDtable function plus Par() instead!

AnalysePlayBin	struct deal dl, struct playTraceBin play, struct solvedPlay * solvedp, int thrId	Binary	Returns the par result after each card in a particular play sequence
AnalysePlayPBN	struct dealPBN dlPBN, struct playTracePBN playPBN, struct solvedPlay * solvedp, int thrId	PBN	As AnalysePlayBin, but with PBN deal format.
AnalyseAllPlaysBin	struct boards bop, struct playTracesBin plp, struct solvedPlays * solvedp, int chunkSize	Binary	Solves a number of hands with play sequences in parallel. Multi-threaded.
AnalyseAllPlaysPBN	struct boardsPBN bopPBN, struct playTracesPBN plpPBN, struct solvedPlays * solvedp, int chunkSize	PBN	As AnalyseAllPlaysBin, but with PBN deal format.
SetMaxThreads	int userThreads		Used at initial start and can also be called with a request for allocating memory for a specified number of threads. Is apparently mandatory on Linux and Mac (optional on Windows)
FreeMemory	void		Frees all allocated dynamical memory.
GetDDSInfo	DDSInfo * info
ErrorMessage	int code, char line[80]		Turns a return code into an error message string

Data structures

Common encodings are as follows.

Encoding

Element

Value

Suit

Spades

Hearts

Diamonds

Clubs

Hand

North

East

South

West

Vulnerable

None

Both

NS only

EW only

Side

N-S

E-W

Card

Bit 2

…

Bit 13

Bit 14

Rank of deuce

Rank of king

Rank of ace

Holding

A value of 16388 = 16384 + 4 is the encoding for the holding “A2” (ace and deuce).

The two lowest bits are always zero.

PBN

Whole hand

Example:

W:T5.K4.652.A98542 K6.QJT976.QT7.Q6 432.A.AKJ93.JT73 AQJ987.8532.84.K

struct	Field	Comment
deal	int trump;	Suit encoding
	int first;	The hand leading to the trick. Hand encoding
	int currentTrickSuit[3];	Up to 3 cards may already have been played to the trick. Suit encoding. Set to 0 if no card has been played.
	int currentTrickRank[3];	Up to 3 cards may already have been played to the trick. Value range 2-14. Set to 0 if no card has been played.
	unsigned int remainCards[4][4];	1st index is Hand, 2nd index is Suit. remainCards uses Holding encoding.

struct	Field	Comment
dealPBN	int trump;	Suit encoding
	int first;	The hand leading to the trick. Hand encoding
	int currentTrickSuit[3];	Up to 3 cards may already have been played to the trick. Suit encoding.
	int currentTrickRank[3];	Up to 3 cards may already have been played to the trick. Value range 2-14. Set to 0 if no card has been played.
	char remainCards[80];	Remaining cards. PBN encoding.

struct	Field	Comment
ddTableDeal	unsigned int cards[4][4];	Encodes a deal. First index is hand. Hand encoding. Second index is suit. Suit encoding.

struct	Field	Comment
ddTableDealPBN	char cards[80];	Encodes a deal. PBN encoding.

struct

Field

Comment

ddTableDeals

int noOfTables;

Number of DD table deals in structure, at most

MAXNOOFTABLES

struct ddTableDeal

deals[X];

X = MAXNOOFTABLES * DDS_STRAINS

struct

Field

Comment

ddTableDealsPBN

int noOfTables;

Number of DD table deals in structure

struct ddTableDealPBN

deals[X];

X = MAXNOOFTABLES * DDS_STRAINS

struct	Field	Comment
boards	int noOfBoards;	Number of boards
	struct deal [MAXNOOFBOARDS];
	int target [MAXNOOFBOARDS];	See SolveBoard
	int solutions [MAXNOOFBOARDS];	See SolveBoard
	int mode [MAXNOOFBOARDS];	See SolveBoard

struct	Field	Comment
boardsPBN	int noOfBoards;	Number of boards
	struct dealPBN [MAXNOOFBOARDS];
	int target [MAXNOOFBOARDS];	See SolveBoard
	int solutions [MAXNOOFBOARDS];	See SolveBoard
	int mode [MAXNOOFBOARDS];	See SolveBoard
struct	Field	Comment
futureTricks	int nodes;	Number of nodes searched by the DD solver
	int cards;	Number of cards for which a result is returned. May be all the cards, but equivalent ranks are omitted, so for a holding of KQ76 only the cards K and 7 would be returned, and the “equals” field below would be 2048 (Q) for the king and 54 (6) for the 7.
	int suit[13];	Suit of the each returned card. Suit encoding
	int rank[13];	Rank of the returned card. Value range 2-14.
	int equals[13];	Lower-ranked equals. Holding encoding.
	int score[13];	-1: target not reached. Otherwise: Target of maximum number of tricks.

struct

Field

Comment

solvedBoards

int noOfBoards;

struct futureTricks

solvedBoard

[MAXNOOFBOARDS];

Struct	Field	Comment
ddTableResults	int resTable[5][4];	Encodes the solution of a deal for combinations of denomination and declarer. First index is denomination. Suit encoding. Second index is declarer. Hand encoding. Each entry is a number of tricks.

Struct

Field

Comment

ddTablesRes

int noOfBoards;

Number of DD table deals in structure, at most

MAXNOOFTABLES

struct ddTableResults

results[X];

X = MAXNOOFTABLES * DDS_STRAINS

struct

Field

Comment

parResults

char parScore[2][16];

First index is NS/EW. Side encoding.

char parContractsString

[2][128];

First index is NS/EW. Side encoding.

struct

Field

Comment

allParResults

struct parResults

[MAXNOOFTABLES];

There are up to 20 declarer/strain combinations per DD table

struct	Field	Comment
parResultsDealer	int number;
	int score;
	char contracts[10][10];

struct	Field	Comment
parResultsMaster	int score;
	int number;
	struct contractType contracts[10];

struct	Field	Comment
contractType	int underTricks;
	int overTricks;
	int level;
	int denom;
	int seats;

struct	Field	Comment
parTextResults	char parText[2][128];
	int equal;

struct	Field	Comment
DDSInfo	int major, minor patch;
	char versionString[10];	Printable version string
	int system;	0 unknown, 1 Windows, 2 Cygwin, 3 Linux, 4 Apple
	int compiler;	0 unknown, 1 Microsoft Visual C++, 2 mingw, 3 GNU g++, 4 clang
	int constructor;	0 none, 1 DLLMain, 2 Unix-style
	int threading;	0 none, 1 Windows, 2 OpenMP
	int noOfThreads;
	char systemString[512];	Printable summary string

struct	Field	Comment
playTraceBin	int number;	Number of cards in the play trace, starting from the beginning of the hand
	int suit[52];	Suit encoding.
	int rank[52];	Encoding 2 .. 14 (not Card encoding).

struct	Field	Comment
playTracePBN	int number;	Number of cards in the play trace, starting from the beginning of the hand
	int cards[106];	String of cards with no space in between, also not between tricks. Each card consists of a suit (C/D/H/S) and then a rank (2 .. A). The string must be null-terminated.

struct

Field

Comment

playTracesBin

int noOfBoards;

struct playTraceBin

plays[MAXNOOFBOARDS];

struct

Field

Comment

playTracesPBN

int noOfBoards;

Struct playTracePBN

plays[MAXNOOFBOARDS];

struct	Field	Comment
solvedPlay	int number;
	int tricks[53];	Starting position and up to 52 cards

struct

Field

Comment

solvedPlays

int noOfBoards;

struct solvedPlay

solved[MAXNOOFBOARDS];

Functions

SolveBoard

struct deal dl,

int target,

int solutions,

int mode,

struct futureTricks *futp,

int threadIndex

SolveBoardPBN

struct dealPBN dl,

int target,

int solutions,

int mode,

struct futureTricks *futp,

int threadIndex

SolveBoardPBN is just like SolveBoard, except for the input format. Historically it was one of the first functions, and it exposes the thread index directly to the user. Later functions generally don’t do that, and they also hide the implementation details such as transposition tables, see below.

SolveBoard solves a single deal “dl” and returns the result in “*futp” which must be declared before calling SolveBoard.

If you have multiple hands to solve, it is always better to group them together into a single function call than to use SolveBoard.

SolveBoard is thread-safe, so several threads can call SolveBoard in parallel. Thus the user of DDS can create threads and call SolveBoard in parallel over them. The maximum number of threads is fixed in the DLL at compile time and is currently 16. So “threadIndex” must be between 0 and 15 inclusive; see also the function SetMaxThreads. Together with the PlayAnalyse functions, this is the only function that exposes the thread number to the user.

There is a “transposition table” memory associated with each thread. Each node in the table is effectively a position after certain cards have been played and other certain cards remain. The table is not deleted automatically after each call to SolveBoard, so it can be reused from call to call. However, it only really makes sense to reuse the table when the hand is very similar in the two calls. The function will still run if this is not the case, but it won’t be as efficient. The reuse of the transposition table can be controlled by the “mode” parameter, but normally this is not needed and should not be done.

The three parameters “target”, “solutions” and “mode” together control the function. Generally speaking, the target is the number of tricks to be won (at least) by the side to play; solutions controls how many solutions should be returned; and mode controls the search behavior. See next page for definitions.

For equivalent cards, only the highest is returned, and lower equivalent cards are encoded in the futureTricks structure (see “equals”).

target	solutions	Comment
-1	1	Find the maximum number of tricks for the side to play. Return only one of the optimum cards and its score.
-1	2	Find the maximum number of tricks for the side to play. Return all optimum cards and their scores.
0	1	Return only one of the cards legal to play, with score set to 0.
0	2	Return all cards that legal to play, with score set to 0.
1 .. 13	1	If score is -1: Target cannot be reached. If score is 0: In fact no tricks at all can be won. If score is > 0: score will always equal target, even if more tricks can be won. One of the cards achieving the target is returned.
1 .. 13	2	Return all cards meeting (at least) the target. If the target cannot be achieved, only one card is returned with the score set as above.
any	3	Return all cards that can be legally played, with their scores in descending order.

mode	Reuse TT?	Comment
0	Automatic if same trump suit and the same or nearly the same cards distribution, deal.first can be different.	Do not search to find the score if the hand to play has only one card, including its equivalents, to play. Score is set to –2 for this card, indicating that there are no alternative cards. If there are multiple choices for cards to play, search is done to find the score. This mode is very fast but you don’t always search to find the score.
1		Always search to find the score. Even when the hand to play has only one card, with possible equivalents, to play.
2	Always

Note: mode no longer always has this effect internally in DDS. We think mode is no longer useful, and we may use it for something else in the future. If you think you need it, let us know!

“Reuse” means “reuse the transposition table from the previous run with the same thread number”. For mode = 2 it is the responsibility of the programmer using the DLL to ensure that reusing the table is safe in the actual situation. Example: Deal is the same, except for deal.first. The trump suit is the same.

1^st call, East leads: SolveBoard(deal, -1, 1, 1, &fut, 0), deal.first=1

2^nd call, South leads: SolveBoard(deal, -1, 1, 2, &fut, 0), deal.first=2

3rd call, West leads: SolveBoard(deal, -1, 1, 2, &fut, 0), deal.first=3

4th call, North leads: SolveBoard(deal, -1, 1, 2, &fut, 0), deal.first=0

CalcDDtable

struct ddTableDeal tableDeal,
struct ddTableResults * tablep

CalcDDtablePBN

struct ddTableDealPBN tableDealPBN,

struct ddTableResults * tablep

CalcDDtablePBN is just like CalcDDtable, except for the input format.

CalcDDtable solves a single deal “tableDeal” and returns the double-dummy values for the initial 52 cards for all the 20 combinations of denomination and declarer in “*tablep”, which must be declared before calling CalcDDtable.

CalcAllTables

struct ddTableDeals *dealsp,

int mode,

int trumpFilter[5],

struct ddTablesRes *resp,

struct allParResults *presp

CalcAllTablesPBN

struct ddTableDealsPBN *dealsp,

int mode,

int trumpFilter[5],

struct ddTablesRes *resp,

struct allParResults *presp

CalcAllTablesPBN is just like CalcAllTables, except for the input format.

CalcAllTables calculates the double dummy values of the denomination/declarer hand combinations in “*dealsp” for a number of DD tables in parallel. This increases the speed compared to calculating these values using a CalcDDtable call for each DD table. The results are returned in “*resp” which must be defined before CalcAllTables is called.

The “mode” parameter contains the vulnerability (Vulnerable encoding; not to be confused with the SolveBoard mode) for use in the par calculation. It is set to -1 if no par calculation is to be performed.

There are 5 possible denominations or strains (the four trump suits and no trump). The parameter “trumpFilter” describes which, if any, of the 5 possibilities that will be excluded from the calculations. They are defined in Suit encoding order, so setting trumpFilter to {FALSE, FALSE, TRUE, TRUE, TRUE} means that values will only be calculated for the trump suits spades and hearts.

The maximum number of DD tables in a CalcAllTables call depends on the number of strains required, see the following table:

Number of strains	Maximum number of DD tables
5	32
4	40
3	53
2	80
1	160

SolveAllBoards

struct boards *bop,

struct solvedBoards

* solvedp

SolveAllChunksBin

struct boards *bop,

struct solvedBoards *solvedp,

int chunkSize

SolveAllChunksPBN

struct boardsPBN *bop,

struct solvedBoards *solvedp,

int chunkSize

SolveAllChunks is an alias for SolveAllChunksPBN; don’t use it.

SolveAllBoards used to be an alias for SolveAllChunksPBN with a chunkSize of 1; however this has been changed in v2.8, and we now recommend only to use SolveAllBoards and not the chunk functions any more; explanation follows.

The SolveAll* functions invoke SolveBoard several times in parallel in multiple threads, rather than sequentially in a single thread. This increases execution speed. Up to 200 boards are permitted per call.

It is important to understand the parallelism and the concept of a chunk.

If the chunk size is 1, then each of the threads starts out with a single board. If there are four threads, then boards 0, 1, 2 and 3 are initially solved. If thread 2 is finished first, it gets the next available board, in this case board 4. Perhaps this is a particularly easy board, so thread 2 also finishes this board before any other thread completes. Thread 2 then also gets board 5, and so on. This continues until all boards have been solved. In the end, three of the threads will be waiting for the last thread to finish, which causes a bit of inefficiency.

The transposition table in a given thread (see SolveBoard) is generally not reused between board 2, 4 and 5 in thread 2. This only happens if SolveBoard itself determines that the boards are suspiciously similar.

If the chunk size is 2, then initially thread 0 gets boards 0 and 1, thread 1 gets boards 2 and 3, thread 2 gets boards 4 and 5, and thread 3 gets boards 6 and 7. When a thread is finished, it gets two new boards in one go, for instance boards 8 and 9. The transposition table in a given thread is reused within a chunk.

No matter what the chunk size is, the boards are solved in parallel. If the user knows that boards are grouped in chunks of 2 or 10, it is possible to force the DD solver to use this knowledge. However, this is rather limiting on the user, as the alignment must remain perfect throughout the batch.

SolveAllBoards now detects repetitions automatically within a batch, whether or not the hands are evenly arranged and whether or not the duplicates are next to each other. This is more flexible and transparent to the user, and the overhead is negligible. Therefore, use SolveAllBoards!

Par