Release Notes DDS 1.0.1
-----------------------
Includes support for reuse of the transposition table
contents for a subsequent search. (This idea was brought forward 
independently by Alex Martelli and Flip Cronje.)
When setting the SolveBoard solutions to 2, the transposition 
table contents will be reused when searching for alternative cards,
giving a slightly decreased search time in most cases, in a few cases
the decrease is substantial.
Earlier the contents of the transposition table was always emptied 
between searches.

The way to control whether or not to empty the transposition table
is to properly set the parameter mtd in the InitSearch function:

mtd=FALSE means that the transposition table is emptied, mtd=TRUE
that it is not emptied.


Release Notes DDS 1.0.2
-----------------------
Includes a bug fix for the case when the SolveBoard parameter
"solutions" is set to 3. SolveBoard did not handle correctly the case
when the scores of the alternative cards were a mix of zero and 
non-zero. This is now fixed in 1.0.2.


Release Notes DDS 1.0.3
----------------------- 
Includes:
a) An improvement of the move ordering algorithm. When LHO ruffs,
this is now given a low weight if the highest rank of RHO beats both
the rank of the leading card and the highest rank of the leading suit
at the partner of the leading hand, assuming that the partner of the 
leading hand is not void in the suit of the leading card.
If these conditions do not hold, the weight for the LHO ruffing card is 
high.
b) For the SolveBoard parameter solutions=3, reuse of the transposition
table contents between card searches have been added.
a) and b) together give on the average an improvement performance. 

Release Notes DDS 1.0.4
-----------------------
When introducing the possibility of reusing the transposition table
between subsequent searches in 1.0.1, unfortunately a bug was introduced 
as well. It occassionally gives wrong scores when evaluating the last hand 
of the trick using solutions=2 or 3.  
This bug is fixed in 1.0.4.

Release Notes DDS 1.0.5
--------------------------
1) About 25% speed improvement due to a search algorithm improvement:
For moves generated in the search tree, only one "small" card move per hand 
and suit is generated. By "small" card is meant a card with a rank that is 
not winning due to its rank value.
Earlier all "small" cards generated their own moves.
This improvement comes from a tip by Hans Kuijf.

2) The size of the transposition table has been increased by 25%. This was 
done after realizing that for some extreme deals, rarely occuring, when
solutions=2 and 3 the transposition table could get completely filled up, 
making the search time extremely long for those cases.
The increased transposition table size should still fit well inside a 256 MB
machine.

3) Parameter mode=2 indicates that the transposition table is reused from an
earlier SolveBoard call. It is restricted to the case where the same conditions
apply for the 2 calls, except that the leading hand in the second call is the
partner to the leading hand in the first call.
I.e. same deal incl same trump suit, same target (not equal to -1), solutions=1. 


Release Notes DDS 1.0.6
--------------------------
1) A bug has been fixed in the DDS start up code.
The bug did not affect the accuracy of the solver.
Thanks to Fabien Bradmetz for spotting this!

2) About 10% increased speed due to improvements in Quick Tricks. 

3) An error code has been added for checking the input of of cards earlier played
in the trick.

4) A document (mode2.txt) describes source code examples how to call
SolveBoard using the parameter mode set to 2.


Release Notes DDS 1.1.0
---------------------------
1) Performance improvement giving a speed of about twice that of 1.0.6.

2) Redesign of the transposition table solution, see algorithm description.

3) Improvements in the Quicktricks algorithm, see algorithm description.

4) Dump in text file of SolveBoard parameters if SolveBoard returns with
error code.


Release Notes DDS 1.1.1
---------------------------
1) For target=-1 and solutions=1/2/3, the transposition table contents is now
reused also when SolveBoard is called with a new target value.

2) Slight improvement of the QuickTricks algorithm.

3) Moderate speed improvement compared to 1.1.1.


Release Notes DDS 1.1.2
---------------------------
1) Increased performance (about 15%) due to improved
algorithm for move ordering. For more information, look in the
algorithm description.


Release Notes DDS 1.1.3
---------------------------
1) Compiles with Visual C++ 2005 Express edition
2) Some improvements in move ordering, quick tricks
and the new later tricks algorithms.
3) Total improvement about 20% including the faster
code generated by Visual C++ and the algorithm 
improvements.


Release Notes DDS 1.1.4
---------------------------
Improvements at usage of the SolveBoard parameter
mode set to 2.
The mode parameter can now be set to 2 for SolveBoard
calls with parameter target set to -1 or a value 0-13,
and solutions set to 1, 2 or 3.
Following the first SolveBoard call with mode=1, mode
can be set to 2 for any further SolveBoard call for
the same deal, also when an opponent hand leads the trick.

If scores are to be calculated for all 4 possible leading hands
of the deal, 1 SolveBoard call can be made with mode=1, and the
other 3 SolveBoard calls for the other 3 leading hands can be
made with mode=2.
Compared to dds 1.1.3, the better support of mode=2 means improved
calculation speed.
Compared to using only mode=1 at calling Solveboard 4 times, each
time for each alternative leading hand, the calculation time is
roughly halfed using mode=2 as described above. 


Release Notes DDS 1.1.5
---------------------------  
A correction has been made removing a risk of DD value miscalculation.
The fault has not however yet showed up as far as I know.

A survival mechanism has been added to handle the situation if the
transposition table gets full:
The transposition table is emptied.
Before this, a full transposition table would in practise lock up the program.
Therefore the present size of the transposition table is quite large (393 MB)
to handle the most difficult deals.
This could lead to extensive memory swapping with the hard drive if the
PC RAM size is 256 MB or less, slowing down the program.
With the new survival mechanism, the program will not lockup at a full
transposition table. The calculation time will increase because of the extra time 
to build up stored positions in the table, but  the increase will be reasonable.
This means that DDS should be possible to run with such small PC RAM sizes 
as 64 MB (this has not been tested though).


Release Notes DDS 1.1.6
---------------------------  
A bug has been fixed. It could show up (but haven't as far as I know)
after the transposition table got full. 
Only a minor part of the maximum transposition table, about 30 MB, is 
now allocated at initialization, if more is needed it is dynamically 
added as need arises. Each incremental addition is about 2 MB. 
This continues until the maximum allowed space is reached, for 512 MB 
PC about 400 MB. Then the incrementally added transposition table pieces 
are wiped out and the initially allocated table is cleared. 


Release Notes DDS 1.1.7
---------------------------
A bug fix and recoding of the dynamic transposition table implementation. 


Release Notes DDS 1.1.8
--------------------------- 
The transposition table solution has been redesigned.
Each "winning ranks node" now contains all ranks for one suit.

When a new transposition table entry is created, it is now
positioned at the front of the alternative "winning ranks node"
list rather than at the end as before.

Each leading hand has now own root pointers pointing to
the "suit lengths combination" tree.

Some changes in QuickTricks.


Release Notes DDS 1.1.9
-----------------------
Speed increase 10-15% compared to 1.1.8.
Main improvements from 1) and 2) below.

1) Much faster initialization of variable rel
(struct type relRanksType).
Implementation done by Thomas Andrews.

2) Move generation:
If the hand-to-play is trick leading hand, the first
positions in the move list now come each from a different suit.
In 1.1.8, the first positions were groups of 2 cards from
each suit.

3) DismissX renamed to NextMove. A bug was corrected
in NextMove, increasing performance. (The bug did not give
incorrect result.)
The bug was found by Jol Bradmetz.

4) The function CountOnes replaced by a table look up.
Implementation by Thomas Andrews.

5) The handStore variable replaced by a macro. Based on
an implementation by Thomas Andrews.

6) The WinAdapt function replaced by a table look up.

7) A new table highestRank gives the highest rank from a
bitmap of ranks. Makes functions UpdateWinner, UpdateSecondBest
and QuickTricks slightly faster.



Release Notes DDS 2.0.0
-----------------------
A bug was fixed.
SolveBoard is now thread-safe.
The SolveBoard parameter "mode" no longer needs to be set to 2.
DDS automatically detect situations when the transposition
table contents can be reused.
Used with a single thread, DDS 2.0.0 has about the same speed as DDS 1.1.9.
Used with 2 parallel threads, DDS 2.0.0 is twice as fast as the single
thread case. 

A new function, CalcDDtable, has been added.
CalcDDtable calls SolveBoard using parallel threads. The number of
parallel threads is the same as the number of processor cores.
CalcDDtable calculates the double dummy values of the initial 52 cards
for all the 20 trump suit/leading hand combinations. 


Release Notes DDS 2.0.1
-----------------------
In 2.0.0, the contents of the transposition table could be erronously
be reused when the previous position contained a different number of
cards. This was caused by an erroneous implementation of the
deal similarity test. This bug was fixed by removing the similarity test.

The DDS version number is defined by a #define statement in dll.h. 


Release Notes DDS 2.1.0
-----------------------
Added OpenMP as multi-thread support for CalcDDtable when compiling with 
gcc 4.4.0 or later.

Added a similarity deals test function for reuse of the transposition
table contents when the current deal is similar to the previous deal. 


Release Notes DDS 2.1.1
-----------------------
The maximum number of threads is configurable depending on the size of the
physical memory. The configuration is either done automatically by reading
out the physical memory size by the operating system, or by supplying
parameter values in InitStart.


Release Notes DDS 2.1.2
-----------------------
2 new callable functions have been added:
SolveBoardPBN and CalcDDtablePBN.
They both accept PBN format for the remaining cards in the deal information
instead of the bit codes.

Code improvements have been done concerning quick tricks detection and move
ordering. 2.1.2 is about 10% faster than 2.1.1.


Release Notes DDS 1.1.12
------------------------
2 new callable functions have been added:
SolveBoardPBN and CalcDDtablePBN.
They both accept PBN format for the remaining cards in the deal information
instead of the bit codes.

Code improvements have been done concerning quick tricks detection and move
ordering. 1.1.12 is about 15-20% faster than 1.1.11.


Release Notes DDS 2.2.0
----------------------- 
1 new callable function has been added:
SolveAllBoards.

Problems that have been fixed in 2.2.0:
1)  DDS used a deficient function in Windows to detect RAM memory size, which
    could cause too few threads to be allocated. DDS now uses a correct Windows
    function to detect the RAM size.
2)  Failure by DDS to detect PBN text coded card distributions with less than
    52 cards.
3)  DDS hanged when a faulty card distribution was provided in the CalcDDtable call.
    This fault was not present in the openMP alternative though.

2.2.0 is 10-15% faster than 2.1.2. 


Release Notes DDS 1.1.13
------------------------
Problem that has been fixed in 1.1.13:
DDS used a deficient function in Windows to detect RAM memory size, which
could cause too few threads to be allocated. DDS now uses a correct Windows
function to detect the RAM size. 

1.1.13 is 5-10% faster than 1.1.12.


Release Notes DDS 2.2.1
----------------------- 
Problem fixed in 2.2.1:
Fault in using the Windows function to detect RAM memory size that could
cause DDS termination.

Clarifying that DDS 2.2.1 is a 32-bit DLL not capable of using more than 4 GB.
Max threads are 8. 


Release Notes DDS 1.1.14
------------------------
Problem fixed in 1.1.14:
Fault in using the Windows function to detect RAM memory size that could
cause DDS termination.


Release Notes DDS 2.2.2
----------------------- 
DDS 2.2.2 is mainly a fault correction release.
Problems fixed:
1. An uninitialized variable used in move generation required that the compiler
initialized it by setting it to zero. This is true for Visual C++ but might
perhaps have caused a problem with some other compiler.
2. The rel table missed initialization for absRank and relRank when all cards in
a suit are missing. 

DDS 2.2.2 is slightly faster than 2.2.1, about 4%.


Release Notes DDS 1.1.15
------------------------
Problem fixed in 1.1.15:
The rel table missed initialization for absRank and relRank when all cards in
a suit are missing.

DDS 1.1.15 is about 7% faster than 1.1.14.


Release Notes DDS 2.2.3
----------------------- 
DDS 2.2.3 add Mingw compiled DLLs and makes a number of code improvements.
Maximum number of threads is increased to 16.
The 2.2.3 Mingw DLLs are about 5% faster than the 2.2.2 DLLs. 

Release Notes DDS 1.1.16
------------------------
DDS 1.1.16 add Mingw compiled DLLs and makes a number of code improvements.
The 1.1.16 Mingw DLLs are about 5% faster than the 1.1.15 DLLs.


Release Notes DDS 2.2.4
-----------------------
A newer port from GCC to Windows, the TDM-GCC 4.7.1 version is used. It gives
faster DLLs and only requires one supporting DLL, which supports OpenMP.
Also, the move ordering has been slighly improved.
A total speed increase of 4% is achieved compared to 2.2.3.


Release Notes DDS 1.1.17
------------------------
A newer port from GCC to Windows, the TDM-GCC 4.7.1 version is used. It gives
faster DLLs and does not require any supporting DLLs.
Also, the move ordering has been slighly improved.
A total speed increase of 4% is achieved compared to 1.1.16.


Release Notes DDS 1.1.18
------------------------
The main news is that this release is licensed under Apache 2.0 instead
of GPL for the previous releases.
There is a slight change of the move ordering compared to 1.1.17.


Release Notes DDS 2.2.4
-----------------------
A newer port from GCC to Windows, the TDM-GCC 4.7.1 version is used. It gives
faster DLLs and only requires one supporting DLL, which supports OpenMP.
Also, the move ordering has been slighly improved.
A total speed increase of 4% is achieved compared to 2.2.3.


Release Notes DDS 2.3.0
-----------------------
The main news is the support of calculation of par scores / contracts.
The Mingw DLL now uses the Win API instead of OpenMP so no
supporting DLL for pthreads is now needed.


Release Notes DDS 2.4.0
-----------------------
Fixing a bug at calculation of par score contracts.
Adding the new functions CalcAllTables and CalcAllTablesPBN.
Numerous fixes for better adaptation to Linux.
Cleaning the code to remove some warnings given at code compilation.
For Windows users, adding readable version information of the DDS DLL file. 


Release Notes DDS 2.4.1
-----------------------  
Updates of functions CalcAllTables and CalcAllTablesPBN for improved performance. 
 

Release Notes DDS 2.4.2
-----------------------  
Fixing bugs in CalcAllTables, CalcAllTablesPBN and CalcPar.


Release Notes DDS 2.4.3
-----------------------  
Problem pointed out by John Goacher with missing par contract has been corrected.
Tuned code for increased performance, especially in function MoveGen and in
the functions WeightAlloc.


Release Notes DDS 2.5.0
-----------------------  
Added function SolveAllChunks.

Release Notes DDS 2.5.1
-----------------------  
The Windows DDS DLL always uses the Windows API for implementing the multithread
functions (i.e. OpenMP is not used for Windows).

Release Notes DDS 2.5.2
-----------------------  
A fault in the par calculation has been fixed. The fault could cause 
presentation of wrongly extra par contracts in cases were the sides had different 
vulnerability.