historical commit dds 2.2.4

dds.cpp

@@ -1,5 +1,5 @@
 
-/* DDS 2.2.3   A bridge double dummy solver.				      */
+/* DDS 2.2.4   A bridge double dummy solver.				      */
 /* Copyright (C) 2006-2012 by Bo Haglund                                      */
 /* Cleanups and porting to Linux and MacOSX (C) 2006 by Alex Martelli         */
 /*								              */
@@ -42,10 +42,6 @@ int noOfThreads=MAXNOOFTHREADS;  /* The number of entries to the transposition t
 								 one entry per thread. */
 int noOfCores;			/* The number of processor cores, however cannot be higher than noOfThreads. */
 
-/*#if defined(_MSC_VER)
-CRITICAL_SECTION solv_crit;
-#endif*/
-
 #ifdef _MANAGED
 #pragma managed(push, off)
 #endif
@@ -59,14 +55,8 @@ extern "C" BOOL APIENTRY DllMain(HMODULE hModule,
 
   if (ul_reason_for_call==DLL_PROCESS_ATTACH) {
     InitStart(0, 0);
-/*#if defined(_MSC_VER)
-    InitializeCriticalSection(&solv_crit);
-#endif*/
   }
   else if (ul_reason_for_call==DLL_PROCESS_DETACH) {
-/*#if defined(_MSC_VER)
-    DeleteCriticalSection(&solv_crit);
-#endif*/
     for (k=0; k<noOfThreads; k++) {
       Wipe(k);
       if (localVar[k].pw[0])
@@ -139,7 +129,7 @@ extern "C" BOOL APIENTRY DllMain(HMODULE hModule,
 			   but preferable InitStart should be called outside
 					 SolveBoard like in DllMain for Windows. */
 
-  if ((thrId<0)||(thrId>=noOfThreads)) {
+  if ((thrId<0)||(thrId>=noOfThreads)) {	/* Fault corrected after suggestion by Dirk Willecke. */
     DumpInput(-15, dl, target, solutions, mode);
 	return -15;
   }
@@ -860,7 +850,8 @@ void InitStart(int gb_ram, int ncores) {
     MEMORYSTATUSEX statex;
 	statex.dwLength = sizeof (statex);
 
-    GlobalMemoryStatusEx (&statex);
+    GlobalMemoryStatusEx (&statex);	/* Using GlobalMemoryStatusEx instead of GlobalMemoryStatus
+					was suggested by Lorne Anderson. */
 
     pcmem=(unsigned long long)statex.ullTotalPhys/1024;
 
@@ -875,7 +866,7 @@ void InitStart(int gb_ram, int ncores) {
     noOfCores=Min(noOfThreads, (int)temp.dwNumberOfProcessors);
 
   #endif
-  #ifdef __linux__
+  #ifdef __linux__   /* The code for linux was suggested by Antony Lee. */
     FILE* fifo = popen("free -k | tail -n+3 | head -n1 | awk '{print $NF}'", "r");
     fscanf(fifo, "%ld", &pcmem);
     fclose(fifo);
@@ -965,9 +956,7 @@ void InitStart(int gb_ram, int ncores) {
     localVar[k].wmem=(WSIZE+1)*sizeof(struct winCardType);
     localVar[k].nmem=(NSIZE+1)*sizeof(struct nodeCardsType);
     localVar[k].lmem=(LSIZE+1)*sizeof(struct posSearchType);
-    /*localVar[k].maxIndex=(int)(
-	  localVar[k].maxmem-localVar[k].summem)/((WSIZE+1) * sizeof(struct winCardType));*/
-	 localVar[k].maxIndex=(int)(
+	localVar[k].maxIndex=(int)(
        (localVar[k].maxmem-localVar[k].summem)/((WSIZE+1) * sizeof(struct winCardType)));
 
     localVar[k].pw = (struct winCardType **)calloc(localVar[k].maxIndex+1, sizeof(struct winCardType *));
@@ -2372,19 +2361,7 @@ int QuickTricks(struct pos * posPoint, int hand,
   commPartner=FALSE;
   for (s=0; s<=3; s++) {
     if ((trump!=4)&&(trump!=s)) {
-     /*if ((posPoint->rankInSuit[hand][s]!=0)&&((posPoint->rankInSuit[lho[hand]][s]!=0)||
-	(posPoint->rankInSuit[lho[hand]][trump]==0))&&
-	((posPoint->rankInSuit[rho[hand]][s]!=0)||(posPoint->rankInSuit[rho[hand]][trump]==0))&&
-	(posPoint->rankInSuit[partner[hand]][s]==0)&&(posPoint->rankInSuit[partner[hand]][trump]>0)
-	&&((posPoint->winner[s].hand==lho[hand])||(posPoint->winner[s].hand==rho[hand]))&&
-	((posPoint->winner[trump].hand==lho[hand])||(posPoint->winner[trump].hand==rho[hand]))) {
-	commPartner=TRUE;
-        commSuit=s;
-        commRank=0;
-	ruff=TRUE;
-        break;
-      }
-      else*/ if (posPoint->winner[s].hand==partner[hand]) {
+      if (posPoint->winner[s].hand==partner[hand]) {
         /* Partner has winning card */
         if (posPoint->rankInSuit[hand][s]!=0) {
         /* Own hand has card in suit */
@@ -2465,16 +2442,7 @@ int QuickTricks(struct pos * posPoint, int hand,
     countPart=posPoint->length[partner[hand]][suit];
     opps=countLho | countRho;
 
-    /*if ((ruff)&&(suit==commSuit)) {
-      qtricks++;
-      if (qtricks>=cutoff)
-	return qtricks;
-      suit++;
-      if ((trump!=4) && (suit==trump))
-        suit++;
-      continue;
-    }
-    else*/ if (!opps && (countPart==0)) {
+    if (!opps && (countPart==0)) {
       if (countOwn==0) {
 	/* Continue with next suit. */
 	if ((trump!=4)&&(trump!=suit)) {
@@ -3466,7 +3434,7 @@ int MoveGen(struct pos * posPoint, int depth, int trump, struct movePlyType *mpl
         k--;
       }
       if (m!=1) {
-        if ((trump!=4)/*&&(posPoint->winner[trump].hand!=-1)*/) {
+        if ((trump!=4)&&(posPoint->winner[trump].rank!=0)) {
           for (k=0; k<=m-1; k++)
 	    mply->move[k].weight=WeightAllocTrump(posPoint,
               &(mply->move[k]), depth, ris, trump, thrId);
@@ -3516,7 +3484,7 @@ int MoveGen(struct pos * posPoint, int depth, int trump, struct movePlyType *mpl
     }
   }
 
-  if ((trump!=4)/*&&(posPoint->winner[trump].hand!=-1)*/) {
+  if ((trump!=4)&&(posPoint->winner[trump].rank!=0)) {
     for (k=0; k<=m-1; k++)
       mply->move[k].weight=WeightAllocTrump(posPoint,
           &(mply->move[k]), depth, 0/*ris*/, trump, thrId);
@@ -3544,7 +3512,7 @@ int WeightAllocNT(struct pos * posPoint, struct moveType * mp, int depth,
   unsigned short notVoidInSuit, int thrId) {
   int weight=0, k, l, kk, ll, suit, suitAdd=0, leadSuit;
   int suitWeightDelta, first, q;
-  int rRank;
+  int rRank, thirdBestHand;
   int suitBonus=0;
   int winMove=FALSE;
   unsigned short suitCount, suitCountLH, suitCountRH, aggr;
@@ -3560,6 +3528,7 @@ int WeightAllocNT(struct pos * posPoint, struct moveType * mp, int depth,
 
   switch (posPoint->handRelFirst) {
     case 0:
+      thirdBestHand=localVar[thrId].rel[aggr].absRank[3][suit].hand;
       suitCount=posPoint->length[q][suit];
       suitCountLH=posPoint->length[lho[q]][suit];
       suitCountRH=posPoint->length[rho[q]][suit];
@@ -3624,6 +3593,13 @@ int WeightAllocNT(struct pos * posPoint, struct moveType * mp, int depth,
           weight+=20/*17*//*14*/;
       }
       else {
+	if ((posPoint->secondBest[suit].hand==partner[first])&&(partner[first]==thirdBestHand))
+	  suitWeightDelta+=22/*20*//*10*/;
+	else if(((posPoint->secondBest[suit].hand==first)&&(partner[first]==thirdBestHand)&&
+	  (posPoint->length[partner[first]][suit]>1))||((posPoint->secondBest[suit].hand==partner[first])&&
+	  (first==thirdBestHand)&&(posPoint->length[partner[first]][suit]>1)))
+	   suitWeightDelta+=24/*20*//*10*/;
+
 	if (((suitCountLH==1)&&(posPoint->winner[suit].hand==lho[first]))
             ||((suitCountRH==1)&&(posPoint->winner[suit].hand==rho[first])))
           weight=suitWeightDelta+25/*23*//*22*/+rRank;
@@ -3826,7 +3802,7 @@ int WeightAllocNT(struct pos * posPoint, struct moveType * mp, int depth,
 int WeightAllocTrump(struct pos * posPoint, struct moveType * mp, int depth,
   unsigned short notVoidInSuit, int trump, int thrId) {
   int weight=0, k, l, kk, ll, suit, suitAdd=0, leadSuit;
-  int suitWeightDelta, first, q, rRank;
+  int suitWeightDelta, first, q, rRank, thirdBestHand;
   int suitBonus=0;
   int winMove=FALSE;
   unsigned short suitCount, suitCountLH, suitCountRH, aggr;
@@ -3842,6 +3818,7 @@ int WeightAllocTrump(struct pos * posPoint, struct moveType * mp, int depth,
 
   switch (posPoint->handRelFirst) {
     case 0:
+      thirdBestHand=localVar[thrId].rel[aggr].absRank[3][suit].hand;
       suitCount=posPoint->length[q][suit];
       suitCountLH=posPoint->length[lho[q]][suit];
       suitCountRH=posPoint->length[rho[q]][suit];
@@ -3988,6 +3965,14 @@ int WeightAllocTrump(struct pos * posPoint, struct moveType * mp, int depth,
           weight+=14/*15*//*12*//*11*/;
       }
       else {
+	if ((posPoint->secondBest[suit].hand==partner[first])&&(partner[first]==thirdBestHand))
+	  suitWeightDelta+=20/*22*/;
+	else if(((posPoint->secondBest[suit].hand==first)&&(partner[first]==thirdBestHand)&&
+	  (posPoint->length[partner[first]][suit]>1))||
+	  ((posPoint->secondBest[suit].hand==partner[first])&&
+	  (first==thirdBestHand)&&(posPoint->length[partner[first]][suit]>1)))
+	  suitWeightDelta+=20/*24*/;
+
         if (((suitCountLH==1)&&(posPoint->winner[suit].hand==lho[first]))
             ||((suitCountRH==1)&&(posPoint->winner[suit].hand==rho[first])))
           weight=suitWeightDelta+rRank-2;
@@ -5579,16 +5564,6 @@ DWORD CALLBACK SolveChunkDDtable (void *) {
   int thid;
   long j;
 
-  /*EnterCriticalSection(&solv_crit);
-  __try
-  {
-    threadIndex++;
-	thid=threadIndex;
-  }
-  __finally
-  {
-    LeaveCriticalSection(&solv_crit);
-  }*/
   thid=InterlockedIncrement(&threadIndex);
 
   while ((j=_InterlockedExchangeAdd(&current, chunk))<param.noOfBoards) {
@@ -5680,16 +5655,6 @@ DWORD CALLBACK SolveChunk (void *) {
   int thid;
   long j;
 
-  /*EnterCriticalSection(&solv_crit);
-  __try
-  {
-    threadIndex++;
-	thid=threadIndex;
-  }
-  __finally
-  {
-    LeaveCriticalSection(&solv_crit);
-  }*/
   thid=InterlockedIncrement(&threadIndex);
 
   while ((j=(InterlockedIncrement(&current)-1))<param.noOfBoards) {
@@ -5785,7 +5750,7 @@ int SolveAllBoards4(struct boards *bop, struct solvedBoards *solvedp) {
   for (i=0; i<MAXNOOFBOARDS; i++)
       solvedp->solvedBoard[i].cards=0;
 
-  omp_set_num_threads(noOfCores);
+  omp_set_num_threads(noOfCores);	/* Added after suggestion by Dirk Willecke. */
 
   #pragma omp parallel shared(bop, solvedp, chunk, fail) private(k)
   {
@@ -5827,7 +5792,7 @@ int SolveAllBoards1(struct boards *bop, struct solvedBoards *solvedp) {
   for (i=0; i<MAXNOOFBOARDS; i++)
     solvedp->solvedBoard[i].cards=0;
 
-  omp_set_num_threads(noOfCores);
+  omp_set_num_threads(noOfCores);	/* Added after suggestion by Dirk Willecke. */
 
   #pragma omp parallel shared(bop, solvedp, chunk, fail) private(k)
   {

dll.h

@@ -35,7 +35,7 @@
 
 /* end of portability-macros section */
 
-#define DDS_VERSION		20203	/* Version 2.2.3. Allowing for 2 digit
+#define DDS_VERSION		20204	/* Version 2.2.4. Allowing for 2 digit
 					minor versions */
 
 #define PBN
@@ -365,10 +365,6 @@ struct localVarType {
   int lenSetSize;  /* Index with range 0 to lenSetSizeLimit */
 };
 
-/*#if defined(_WIN32)
-extern CRITICAL_SECTION solv_crit;
-#endif*/
-
 extern int noOfThreads;
 extern int noOfCores;
 extern struct localVarType localVar[MAXNOOFTHREADS];
@@ -444,6 +440,11 @@ EXTERN_C DLLEXPORT int STDCALL SolveAllBoards(struct boardsPBN *bop, struct solv
 #endif
 #endif
 
+#ifdef __linux
+/*paul hide 2012-11-6 added the following as in previous version*/
+EXTERN_C void InitStart(int gb_ram, int ncores);   /* For usage with ctypes in Linux. */ 
+#endif
+
 void InitStart(int gb_ram, int ncores);
 void InitGame(int gameNo, int moveTreeFlag, int first, int handRelFirst, int thrId);
 void InitSearch(struct pos * posPoint, int depth,

readme.txt

@@ -1,13 +1,13 @@
-DDS 2.2.1,  Bo Haglund 2012-07-16
+DDS 2.2.4,  Bo Haglund 2012-11-07
 
 For Win32, DDS compiles with Visual C++ 2010 Express edition 
-and the Mingw port of gcc.
+and the TDM-GCC/Mingw port of gcc.
 
 When using Visual C++, the statement
 #include "stdafx.h" at the beginning of dds.cpp must be uncommented.
 
 When not using Visual C++, the compilation of DDS includes function CalcDDtable
-implemented using GCC/MingW OpenMP. 
+implemented using TDM-GCC/MingW OpenMP. 
 
 
 Linking with an application using DDS
@@ -24,20 +24,21 @@ Maximum number of threads
 The maximum number of simultaneous threads depends on the PC physical memory size:
 1 GB or less, max 2 threads.
 2 GB, max 4 threads.
-3 or 4 GB, max 8 threads.
+3 or 4 GB, max 16 threads.
 
-For Windows, allocating memory for the maximum number of simultaneous threads can 
+For e.g. Windows, allocating memory for the maximum number of simultaneous threads can 
 be done by reading out the physical memory size from Windows. This is done in the DDS.DLL.
 Another alternative is to provide the physical memory size as a parameter (gb_ram) in the
-InitStart call. This alternative needs to be used when the operating system is not Windows.
+InitStart call. This alternative needs to be used when the operating system does not support
+memory autoconfiguration.
 
 
 Setting the number of simultaneous threads when calling CalcDDtable.
 --------------------------------------------------------------------
 For Windows, this can be done either by reading out the number of processor cores from
 Windows and using this for setting the number of threads, or by supplying the number of
 threads (ncores) in InitStart. This latter alternative needs to be used when the operating 
-system is not Windows. 
+system does not support reading out of the number of processors. 
 
 
 Options at DDS compilation
@@ -54,7 +55,7 @@ The possible configurens thus are:
 2)  "PBN":  Support for SolveBoardPBN and CalcDDtablePBN.
 3) None of these definitions:  No support for any of the above 3 functions.
 
-Staying with the previous configuration might be needed when 2.2.2 is to replace an 
+Staying with the previous configuration might be needed when 2.2.4 is to replace an 
 older 2.x.y version, and the application using DDS cannot handle a changed interface.
 
 

release_notes.txt

@@ -371,12 +371,25 @@ 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 2.2.3 Mingw DLLs are about 5% faster than the 1.1.15 DLLs.
+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.