/[escript]/branches/clazy/escriptcore/src/DataLazy.cpp
ViewVC logotype

Diff of /branches/clazy/escriptcore/src/DataLazy.cpp

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

branches/schroedinger_upto1946/escript/src/DataLazy.cpp revision 1993 by phornby, Fri Nov 7 04:52:15 2008 UTC trunk/escript/src/DataLazy.cpp revision 2152 by jfenwick, Thu Dec 11 04:26:42 2008 UTC
# Line 28  Line 28 
28  #include "UnaryFuncs.h"     // for escript::fsign  #include "UnaryFuncs.h"     // for escript::fsign
29  #include "Utils.h"  #include "Utils.h"
30    
31    // #define LAZYDEBUG(X) X;
32    #define LAZYDEBUG(X)
33    
34  /*  /*
35  How does DataLazy work?  How does DataLazy work?
36  ~~~~~~~~~~~~~~~~~~~~~~~  ~~~~~~~~~~~~~~~~~~~~~~~
# Line 48  I will refer to individual DataLazy obje Line 51  I will refer to individual DataLazy obje
51  Each node also stores:  Each node also stores:
52  - m_readytype \in {'E','T','C','?'} ~ indicates what sort of DataReady would be produced if the expression was  - m_readytype \in {'E','T','C','?'} ~ indicates what sort of DataReady would be produced if the expression was
53      evaluated.      evaluated.
 - m_length ~ how many values would be stored in the answer if the expression was evaluated.  
54  - m_buffsrequired ~ the larged number of samples which would need to be kept simultaneously in order to  - m_buffsrequired ~ the larged number of samples which would need to be kept simultaneously in order to
55      evaluate the expression.      evaluate the expression.
56  - m_samplesize ~ the number of doubles stored in a sample.  - m_samplesize ~ the number of doubles stored in a sample.
# Line 70  The convention that I use, is that the r Line 72  The convention that I use, is that the r
72    
73  For expressions which evaluate to Constant or Tagged, there is a different evaluation method.  For expressions which evaluate to Constant or Tagged, there is a different evaluation method.
74  The collapse method invokes the (non-lazy) operations on the Data class to evaluate the expression.  The collapse method invokes the (non-lazy) operations on the Data class to evaluate the expression.
75    
76    To add a new operator you need to do the following (plus anything I might have forgotten - adding a new group for example):
77    1) Add to the ES_optype.
78    2) determine what opgroup your operation belongs to (X)
79    3) add a string for the op to the end of ES_opstrings
80    4) increase ES_opcount
81    5) add an entry (X) to opgroups
82    6) add an entry to the switch in collapseToReady
83    7) add an entry to resolveX
84  */  */
85    
86    
# Line 87  enum ES_opgroup Line 98  enum ES_opgroup
98     G_UNKNOWN,     G_UNKNOWN,
99     G_IDENTITY,     G_IDENTITY,
100     G_BINARY,        // pointwise operations with two arguments     G_BINARY,        // pointwise operations with two arguments
101     G_UNARY      // pointwise operations with one argument     G_UNARY,     // pointwise operations with one argument
102       G_UNARY_P,       // pointwise operations with one argument, requiring a parameter
103       G_NP1OUT,        // non-pointwise op with one output
104       G_NP1OUT_P,      // non-pointwise op with one output requiring a parameter
105       G_TENSORPROD     // general tensor product
106  };  };
107    
108    
# Line 98  string ES_opstrings[]={"UNKNOWN","IDENTI Line 113  string ES_opstrings[]={"UNKNOWN","IDENTI
113              "asin","acos","atan","sinh","cosh","tanh","erf",              "asin","acos","atan","sinh","cosh","tanh","erf",
114              "asinh","acosh","atanh",              "asinh","acosh","atanh",
115              "log10","log","sign","abs","neg","pos","exp","sqrt",              "log10","log","sign","abs","neg","pos","exp","sqrt",
116              "1/","where>0","where<0","where>=0","where<=0"};              "1/","where>0","where<0","where>=0","where<=0", "where<>0","where=0",
117  int ES_opcount=33;              "symmetric","nonsymmetric",
118                "prod",
119                "transpose", "trace"};
120    int ES_opcount=40;
121  ES_opgroup opgroups[]={G_UNKNOWN,G_IDENTITY,G_BINARY,G_BINARY,G_BINARY,G_BINARY, G_BINARY,  ES_opgroup opgroups[]={G_UNKNOWN,G_IDENTITY,G_BINARY,G_BINARY,G_BINARY,G_BINARY, G_BINARY,
122              G_UNARY,G_UNARY,G_UNARY, //10              G_UNARY,G_UNARY,G_UNARY, //10
123              G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,    // 17              G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,    // 17
124              G_UNARY,G_UNARY,G_UNARY,                    // 20              G_UNARY,G_UNARY,G_UNARY,                    // 20
125              G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,        // 28              G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY,    // 28
126              G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY};              G_UNARY,G_UNARY,G_UNARY,G_UNARY,G_UNARY, G_UNARY_P, G_UNARY_P,      // 35
127                G_NP1OUT,G_NP1OUT,
128                G_TENSORPROD,
129                G_NP1OUT_P, G_NP1OUT_P};
130  inline  inline
131  ES_opgroup  ES_opgroup
132  getOpgroup(ES_optype op)  getOpgroup(ES_optype op)
# Line 140  resultFS(DataAbstract_ptr left, DataAbst Line 161  resultFS(DataAbstract_ptr left, DataAbst
161  }  }
162    
163  // return the shape of the result of "left op right"  // return the shape of the result of "left op right"
164    // the shapes resulting from tensor product are more complex to compute so are worked out elsewhere
165  DataTypes::ShapeType  DataTypes::ShapeType
166  resultShape(DataAbstract_ptr left, DataAbstract_ptr right, ES_optype op)  resultShape(DataAbstract_ptr left, DataAbstract_ptr right, ES_optype op)
167  {  {
168      if (left->getShape()!=right->getShape())      if (left->getShape()!=right->getShape())
169      {      {
170        if (getOpgroup(op)!=G_BINARY)        if ((getOpgroup(op)!=G_BINARY) && (getOpgroup(op)!=G_NP1OUT))
171        {        {
172          throw DataException("Shapes not the name - shapes must match for (point)binary operations.");          throw DataException("Shapes not the name - shapes must match for (point)binary operations.");
173        }        }
# Line 162  resultShape(DataAbstract_ptr left, DataA Line 184  resultShape(DataAbstract_ptr left, DataA
184      return left->getShape();      return left->getShape();
185  }  }
186    
187  // determine the number of points in the result of "left op right"  // return the shape for "op left"
188  size_t  
189  resultLength(DataAbstract_ptr left, DataAbstract_ptr right, ES_optype op)  DataTypes::ShapeType
190    resultShape(DataAbstract_ptr left, ES_optype op)
191  {  {
192     switch (getOpgroup(op))      switch(op)
193     {      {
194     case G_BINARY: return left->getLength();          case TRANS:
195     case G_UNARY: return left->getLength();          return left->getShape();
196     default:      break;
197      throw DataException("Programmer Error - attempt to getLength() for operator "+opToString(op)+".");      case TRACE:
198     }          return DataTypes::scalarShape;
199        break;
200            default:
201        throw DataException("Programmer error - resultShape(left,op) can't compute shapes for operator "+opToString(op)+".");
202        }
203  }  }
204    
205    // determine the output shape for the general tensor product operation
206    // the additional parameters return information required later for the product
207    // the majority of this code is copy pasted from C_General_Tensor_Product
208    DataTypes::ShapeType
209    GTPShape(DataAbstract_ptr left, DataAbstract_ptr right, int axis_offset, int transpose, int& SL, int& SM, int& SR)
210    {
211        
212      // Get rank and shape of inputs
213      int rank0 = left->getRank();
214      int rank1 = right->getRank();
215      const DataTypes::ShapeType& shape0 = left->getShape();
216      const DataTypes::ShapeType& shape1 = right->getShape();
217    
218      // Prepare for the loops of the product and verify compatibility of shapes
219      int start0=0, start1=0;
220      if (transpose == 0)       {}
221      else if (transpose == 1)  { start0 = axis_offset; }
222      else if (transpose == 2)  { start1 = rank1-axis_offset; }
223      else              { throw DataException("DataLazy GeneralTensorProduct Constructor: Error - transpose should be 0, 1 or 2"); }
224    
225      if (rank0<axis_offset)
226      {
227        throw DataException("DataLazy GeneralTensorProduct Constructor: Error - rank of left < axisoffset");
228      }
229    
230      // Adjust the shapes for transpose
231      DataTypes::ShapeType tmpShape0(rank0);    // pre-sizing the vectors rather
232      DataTypes::ShapeType tmpShape1(rank1);    // than using push_back
233      for (int i=0; i<rank0; i++)   { tmpShape0[i]=shape0[(i+start0)%rank0]; }
234      for (int i=0; i<rank1; i++)   { tmpShape1[i]=shape1[(i+start1)%rank1]; }
235    
236      // Prepare for the loops of the product
237      SL=1, SM=1, SR=1;
238      for (int i=0; i<rank0-axis_offset; i++)   {
239        SL *= tmpShape0[i];
240      }
241      for (int i=rank0-axis_offset; i<rank0; i++)   {
242        if (tmpShape0[i] != tmpShape1[i-(rank0-axis_offset)]) {
243          throw DataException("C_GeneralTensorProduct: Error - incompatible shapes");
244        }
245        SM *= tmpShape0[i];
246      }
247      for (int i=axis_offset; i<rank1; i++)     {
248        SR *= tmpShape1[i];
249      }
250    
251      // Define the shape of the output (rank of shape is the sum of the loop ranges below)
252      DataTypes::ShapeType shape2(rank0+rank1-2*axis_offset);  
253      {         // block to limit the scope of out_index
254         int out_index=0;
255         for (int i=0; i<rank0-axis_offset; i++, ++out_index) { shape2[out_index]=tmpShape0[i]; } // First part of arg_0_Z
256         for (int i=axis_offset; i<rank1; i++, ++out_index)   { shape2[out_index]=tmpShape1[i]; } // Last part of arg_1_Z
257      }
258    
259      if (shape2.size()>ESCRIPT_MAX_DATA_RANK)
260      {
261         ostringstream os;
262         os << "C_GeneralTensorProduct: Error - Attempt to create a rank " << shape2.size() << " object. The maximum rank is " << ESCRIPT_MAX_DATA_RANK << ".";
263         throw DataException(os.str());
264      }
265    
266      return shape2;
267    }
268    
269    
270    // determine the number of points in the result of "left op right"
271    // note that determining the resultLength for G_TENSORPROD is more complex and will not be processed here
272    // size_t
273    // resultLength(DataAbstract_ptr left, DataAbstract_ptr right, ES_optype op)
274    // {
275    //    switch (getOpgroup(op))
276    //    {
277    //    case G_BINARY: return left->getLength();
278    //    case G_UNARY: return left->getLength();
279    //    case G_NP1OUT: return left->getLength();
280    //    default:
281    //  throw DataException("Programmer Error - attempt to getLength() for operator "+opToString(op)+".");
282    //    }
283    // }
284    
285  // determine the number of samples requires to evaluate an expression combining left and right  // determine the number of samples requires to evaluate an expression combining left and right
286    // NP1OUT needs an extra buffer because we can't write the answers over the top of the input.
287    // The same goes for G_TENSORPROD
288  int  int
289  calcBuffs(const DataLazy_ptr& left, const DataLazy_ptr& right, ES_optype op)  calcBuffs(const DataLazy_ptr& left, const DataLazy_ptr& right, ES_optype op)
290  {  {
# Line 183  calcBuffs(const DataLazy_ptr& left, cons Line 292  calcBuffs(const DataLazy_ptr& left, cons
292     {     {
293     case G_IDENTITY: return 1;     case G_IDENTITY: return 1;
294     case G_BINARY: return max(left->getBuffsRequired(),right->getBuffsRequired()+1);     case G_BINARY: return max(left->getBuffsRequired(),right->getBuffsRequired()+1);
295     case G_UNARY: return max(left->getBuffsRequired(),1);     case G_UNARY:
296       case G_UNARY_P: return max(left->getBuffsRequired(),1);
297       case G_NP1OUT: return 1+max(left->getBuffsRequired(),1);
298       case G_NP1OUT_P: return 1+max(left->getBuffsRequired(),1);
299       case G_TENSORPROD: return 1+max(left->getBuffsRequired(),right->getBuffsRequired()+1);
300     default:     default:
301      throw DataException("Programmer Error - attempt to calcBuffs() for operator "+opToString(op)+".");      throw DataException("Programmer Error - attempt to calcBuffs() for operator "+opToString(op)+".");
302     }     }
# Line 208  opToString(ES_optype op) Line 321  opToString(ES_optype op)
321    
322  DataLazy::DataLazy(DataAbstract_ptr p)  DataLazy::DataLazy(DataAbstract_ptr p)
323      : parent(p->getFunctionSpace(),p->getShape()),      : parent(p->getFunctionSpace(),p->getShape()),
324      m_op(IDENTITY)      m_op(IDENTITY),
325        m_axis_offset(0),
326        m_transpose(0),
327        m_SL(0), m_SM(0), m_SR(0)
328  {  {
329     if (p->isLazy())     if (p->isLazy())
330     {     {
# Line 225  DataLazy::DataLazy(DataAbstract_ptr p) Line 341  DataLazy::DataLazy(DataAbstract_ptr p)
341      else if (p->isTagged()) {m_readytype='T';}      else if (p->isTagged()) {m_readytype='T';}
342      else {throw DataException("Unknown DataReady instance in DataLazy constructor.");}      else {throw DataException("Unknown DataReady instance in DataLazy constructor.");}
343     }     }
    m_length=p->getLength();  
344     m_buffsRequired=1;     m_buffsRequired=1;
345     m_samplesize=getNumDPPSample()*getNoValues();     m_samplesize=getNumDPPSample()*getNoValues();
346  cout << "(1)Lazy created with " << m_samplesize << endl;     m_maxsamplesize=m_samplesize;
347    LAZYDEBUG(cout << "(1)Lazy created with " << m_samplesize << endl;)
348  }  }
349    
350    
# Line 236  cout << "(1)Lazy created with " << m_sam Line 352  cout << "(1)Lazy created with " << m_sam
352    
353  DataLazy::DataLazy(DataAbstract_ptr left, ES_optype op)  DataLazy::DataLazy(DataAbstract_ptr left, ES_optype op)
354      : parent(left->getFunctionSpace(),left->getShape()),      : parent(left->getFunctionSpace(),left->getShape()),
355      m_op(op)      m_op(op),
356        m_axis_offset(0),
357        m_transpose(0),
358        m_SL(0), m_SM(0), m_SR(0)
359  {  {
360     if (getOpgroup(op)!=G_UNARY)     if ((getOpgroup(op)!=G_UNARY) && (getOpgroup(op)!=G_NP1OUT))
361     {     {
362      throw DataException("Programmer error - constructor DataLazy(left, op) will only process UNARY operations.");      throw DataException("Programmer error - constructor DataLazy(left, op) will only process UNARY operations.");
363     }     }
364    
365     DataLazy_ptr lleft;     DataLazy_ptr lleft;
366     if (!left->isLazy())     if (!left->isLazy())
367     {     {
# Line 252  DataLazy::DataLazy(DataAbstract_ptr left Line 372  DataLazy::DataLazy(DataAbstract_ptr left
372      lleft=dynamic_pointer_cast<DataLazy>(left);      lleft=dynamic_pointer_cast<DataLazy>(left);
373     }     }
374     m_readytype=lleft->m_readytype;     m_readytype=lleft->m_readytype;
    m_length=left->getLength();  
375     m_left=lleft;     m_left=lleft;
376     m_buffsRequired=1;     m_buffsRequired=calcBuffs(m_left, m_right,m_op); // yeah m_right will be null at this point
377     m_samplesize=getNumDPPSample()*getNoValues();     m_samplesize=getNumDPPSample()*getNoValues();
378       m_maxsamplesize=max(m_samplesize,m_left->getMaxSampleSize());
379  }  }
380    
381    
382  // In this constructor we need to consider interpolation  // In this constructor we need to consider interpolation
383  DataLazy::DataLazy(DataAbstract_ptr left, DataAbstract_ptr right, ES_optype op)  DataLazy::DataLazy(DataAbstract_ptr left, DataAbstract_ptr right, ES_optype op)
384      : parent(resultFS(left,right,op), resultShape(left,right,op)),      : parent(resultFS(left,right,op), resultShape(left,right,op)),
385      m_op(op)      m_op(op),
386        m_SL(0), m_SM(0), m_SR(0)
387  {  {
388     if (getOpgroup(op)!=G_BINARY)     if ((getOpgroup(op)!=G_BINARY))
389     {     {
390      throw DataException("Programmer error - constructor DataLazy(left, right, op) will only process BINARY operations.");      throw DataException("Programmer error - constructor DataLazy(left, right, op) will only process BINARY operations.");
391     }     }
# Line 276  DataLazy::DataLazy(DataAbstract_ptr left Line 397  DataLazy::DataLazy(DataAbstract_ptr left
397      Data tmp(ltemp,fs);      Data tmp(ltemp,fs);
398      left=tmp.borrowDataPtr();      left=tmp.borrowDataPtr();
399     }     }
400     if (getFunctionSpace()!=right->getFunctionSpace())   // left needs to be interpolated     if (getFunctionSpace()!=right->getFunctionSpace())   // right needs to be interpolated
401       {
402        Data tmp(Data(right),getFunctionSpace());
403        right=tmp.borrowDataPtr();
404       }
405       left->operandCheck(*right);
406    
407       if (left->isLazy())          // the children need to be DataLazy. Wrap them in IDENTITY if required
408       {
409        m_left=dynamic_pointer_cast<DataLazy>(left);
410       }
411       else
412       {
413        m_left=DataLazy_ptr(new DataLazy(left));
414       }
415       if (right->isLazy())
416       {
417        m_right=dynamic_pointer_cast<DataLazy>(right);
418       }
419       else
420       {
421        m_right=DataLazy_ptr(new DataLazy(right));
422       }
423       char lt=m_left->m_readytype;
424       char rt=m_right->m_readytype;
425       if (lt=='E' || rt=='E')
426       {
427        m_readytype='E';
428       }
429       else if (lt=='T' || rt=='T')
430       {
431        m_readytype='T';
432       }
433       else
434       {
435        m_readytype='C';
436       }
437       m_samplesize=getNumDPPSample()*getNoValues();
438       m_maxsamplesize=max(max(m_samplesize,m_right->getMaxSampleSize()),m_left->getMaxSampleSize());  
439       m_buffsRequired=calcBuffs(m_left, m_right,m_op);
440    LAZYDEBUG(cout << "(3)Lazy created with " << m_samplesize << endl;)
441    }
442    
443    DataLazy::DataLazy(DataAbstract_ptr left, DataAbstract_ptr right, ES_optype op, int axis_offset, int transpose)
444        : parent(resultFS(left,right,op), GTPShape(left,right, axis_offset, transpose, m_SL,m_SM, m_SR)),
445        m_op(op),
446        m_axis_offset(axis_offset),
447        m_transpose(transpose)
448    {
449       if ((getOpgroup(op)!=G_TENSORPROD))
450       {
451        throw DataException("Programmer error - constructor DataLazy(left, right, op, ax, tr) will only process BINARY operations which require parameters.");
452       }
453       if ((transpose>2) || (transpose<0))
454       {
455        throw DataException("DataLazy GeneralTensorProduct constructor: Error - transpose should be 0, 1 or 2");
456       }
457       if (getFunctionSpace()!=left->getFunctionSpace())    // left needs to be interpolated
458       {
459        FunctionSpace fs=getFunctionSpace();
460        Data ltemp(left);
461        Data tmp(ltemp,fs);
462        left=tmp.borrowDataPtr();
463       }
464       if (getFunctionSpace()!=right->getFunctionSpace())   // right needs to be interpolated
465     {     {
466      Data tmp(Data(right),getFunctionSpace());      Data tmp(Data(right),getFunctionSpace());
467      right=tmp.borrowDataPtr();      right=tmp.borrowDataPtr();
# Line 313  DataLazy::DataLazy(DataAbstract_ptr left Line 498  DataLazy::DataLazy(DataAbstract_ptr left
498     {     {
499      m_readytype='C';      m_readytype='C';
500     }     }
501     m_length=resultLength(m_left,m_right,m_op);     m_samplesize=getNumDPPSample()*getNoValues();
502     m_samplesize=getNumDPPSample()*getNoValues();         m_maxsamplesize=max(max(m_samplesize,m_right->getMaxSampleSize()),m_left->getMaxSampleSize());  
503     m_buffsRequired=calcBuffs(m_left, m_right,m_op);     m_buffsRequired=calcBuffs(m_left, m_right,m_op);
504  cout << "(3)Lazy created with " << m_samplesize << endl;  LAZYDEBUG(cout << "(4)Lazy created with " << m_samplesize << endl;)
505    }
506    
507    
508    DataLazy::DataLazy(DataAbstract_ptr left, ES_optype op, int axis_offset)
509        : parent(left->getFunctionSpace(), resultShape(left,op)),
510        m_op(op),
511        m_axis_offset(axis_offset),
512        m_transpose(0),
513        m_tol(0)
514    {
515       if ((getOpgroup(op)!=G_NP1OUT_P))
516       {
517        throw DataException("Programmer error - constructor DataLazy(left, op, ax) will only process UNARY operations which require parameters.");
518       }
519       DataLazy_ptr lleft;
520       if (!left->isLazy())
521       {
522        lleft=DataLazy_ptr(new DataLazy(left));
523       }
524       else
525       {
526        lleft=dynamic_pointer_cast<DataLazy>(left);
527       }
528       m_readytype=lleft->m_readytype;
529       m_left=lleft;
530       m_buffsRequired=calcBuffs(m_left, m_right,m_op); // yeah m_right will be null at this point
531       m_samplesize=getNumDPPSample()*getNoValues();
532       m_maxsamplesize=max(m_samplesize,m_left->getMaxSampleSize());
533    LAZYDEBUG(cout << "(5)Lazy created with " << m_samplesize << endl;)
534  }  }
535    
536    DataLazy::DataLazy(DataAbstract_ptr left, ES_optype op, double tol)
537        : parent(left->getFunctionSpace(), left->getShape()),
538        m_op(op),
539        m_axis_offset(0),
540        m_transpose(0),
541        m_tol(tol)
542    {
543       if ((getOpgroup(op)!=G_UNARY_P))
544       {
545        throw DataException("Programmer error - constructor DataLazy(left, op, tol) will only process UNARY operations which require parameters.");
546       }
547       DataLazy_ptr lleft;
548       if (!left->isLazy())
549       {
550        lleft=DataLazy_ptr(new DataLazy(left));
551       }
552       else
553       {
554        lleft=dynamic_pointer_cast<DataLazy>(left);
555       }
556       m_readytype=lleft->m_readytype;
557       m_left=lleft;
558       m_buffsRequired=calcBuffs(m_left, m_right,m_op); // yeah m_right will be null at this point
559       m_samplesize=getNumDPPSample()*getNoValues();
560       m_maxsamplesize=max(m_samplesize,m_left->getMaxSampleSize());
561    LAZYDEBUG(cout << "(6)Lazy created with " << m_samplesize << endl;)
562    }
563    
564  DataLazy::~DataLazy()  DataLazy::~DataLazy()
565  {  {
# Line 332  DataLazy::getBuffsRequired() const Line 573  DataLazy::getBuffsRequired() const
573  }  }
574    
575    
576    size_t
577    DataLazy::getMaxSampleSize() const
578    {
579        return m_maxsamplesize;
580    }
581    
582  /*  /*
583    \brief Evaluates the expression using methods on Data.    \brief Evaluates the expression using methods on Data.
584    This does the work for the collapse method.    This does the work for the collapse method.
# Line 351  DataLazy::collapseToReady() Line 598  DataLazy::collapseToReady()
598    DataReady_ptr pleft=m_left->collapseToReady();    DataReady_ptr pleft=m_left->collapseToReady();
599    Data left(pleft);    Data left(pleft);
600    Data right;    Data right;
601    if (getOpgroup(m_op)==G_BINARY)    if ((getOpgroup(m_op)==G_BINARY) || (getOpgroup(m_op)==G_TENSORPROD))
602    {    {
603      right=Data(m_right->collapseToReady());      right=Data(m_right->collapseToReady());
604    }    }
# Line 450  DataLazy::collapseToReady() Line 697  DataLazy::collapseToReady()
697      case LEZ:      case LEZ:
698      result=left.whereNonPositive();      result=left.whereNonPositive();
699      break;      break;
700        case NEZ:
701        result=left.whereNonZero(m_tol);
702        break;
703        case EZ:
704        result=left.whereZero(m_tol);
705        break;
706        case SYM:
707        result=left.symmetric();
708        break;
709        case NSYM:
710        result=left.nonsymmetric();
711        break;
712        case PROD:
713        result=C_GeneralTensorProduct(left,right,m_axis_offset, m_transpose);
714        break;
715        case TRANS:
716        result=left.transpose(m_axis_offset);
717        break;
718        case TRACE:
719        result=left.trace(m_axis_offset);
720        break;
721      default:      default:
722      throw DataException("Programmer error - do not know how to resolve operator "+opToString(m_op)+".");      throw DataException("Programmer error - collapseToReady does not know how to resolve operator "+opToString(m_op)+".");
723    }    }
724    return result.borrowReadyPtr();    return result.borrowReadyPtr();
725  }  }
# Line 478  DataLazy::collapse() Line 746  DataLazy::collapse()
746  }  }
747    
748  /*  /*
749    \brief Compute the value of the expression (binary operation) for the given sample.    \brief Compute the value of the expression (unary operation) for the given sample.
750    \return Vector which stores the value of the subexpression for the given sample.    \return Vector which stores the value of the subexpression for the given sample.
751    \param v A vector to store intermediate results.    \param v A vector to store intermediate results.
752    \param offset Index in v to begin storing results.    \param offset Index in v to begin storing results.
# Line 533  DataLazy::resolveUnary(ValueType& v, siz Line 801  DataLazy::resolveUnary(ValueType& v, siz
801      tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::tanh);      tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::tanh);
802      break;      break;
803      case ERF:      case ERF:
804  #ifdef _WIN32  #if defined (_WIN32) && !defined(__INTEL_COMPILER)
805      throw DataException("Error - Data:: erf function is not supported on _WIN32 platforms.");      throw DataException("Error - Data:: erf function is not supported on _WIN32 platforms.");
806  #else  #else
807      tensor_unary_operation(m_samplesize, left, result, ::erf);      tensor_unary_operation(m_samplesize, left, result, ::erf);
808      break;      break;
809  #endif  #endif
810     case ASINH:     case ASINH:
811  #ifdef _WIN32  #if defined (_WIN32) && !defined(__INTEL_COMPILER)
812      tensor_unary_operation(m_samplesize, left, result, escript::asinh_substitute);      tensor_unary_operation(m_samplesize, left, result, escript::asinh_substitute);
813  #else  #else
814      tensor_unary_operation(m_samplesize, left, result, ::asinh);      tensor_unary_operation(m_samplesize, left, result, ::asinh);
815  #endif    #endif  
816      break;      break;
817     case ACOSH:     case ACOSH:
818  #ifdef _WIN32  #if defined (_WIN32) && !defined(__INTEL_COMPILER)
819      tensor_unary_operation(m_samplesize, left, result, escript::acosh_substitute);      tensor_unary_operation(m_samplesize, left, result, escript::acosh_substitute);
820  #else  #else
821      tensor_unary_operation(m_samplesize, left, result, ::acosh);      tensor_unary_operation(m_samplesize, left, result, ::acosh);
822  #endif    #endif  
823      break;      break;
824     case ATANH:     case ATANH:
825  #ifdef _WIN32  #if defined (_WIN32) && !defined(__INTEL_COMPILER)
826      tensor_unary_operation(m_samplesize, left, result, escript::atanh_substitute);      tensor_unary_operation(m_samplesize, left, result, escript::atanh_substitute);
827  #else  #else
828      tensor_unary_operation(m_samplesize, left, result, ::atanh);      tensor_unary_operation(m_samplesize, left, result, ::atanh);
# Line 601  DataLazy::resolveUnary(ValueType& v, siz Line 869  DataLazy::resolveUnary(ValueType& v, siz
869      case LEZ:      case LEZ:
870      tensor_unary_operation(m_samplesize, left, result, bind2nd(less_equal<double>(),0.0));      tensor_unary_operation(m_samplesize, left, result, bind2nd(less_equal<double>(),0.0));
871      break;      break;
872    // There are actually G_UNARY_P but I don't see a compelling reason to treat them differently
873        case NEZ:
874        tensor_unary_operation(m_samplesize, left, result, bind2nd(AbsGT(),m_tol));
875        break;
876        case EZ:
877        tensor_unary_operation(m_samplesize, left, result, bind2nd(AbsLTE(),m_tol));
878        break;
879    
880      default:      default:
881      throw DataException("Programmer error - resolveUnary can not resolve operator "+opToString(m_op)+".");      throw DataException("Programmer error - resolveUnary can not resolve operator "+opToString(m_op)+".");
# Line 612  DataLazy::resolveUnary(ValueType& v, siz Line 887  DataLazy::resolveUnary(ValueType& v, siz
887    
888    
889    
890    
891    /*
892      \brief Compute the value of the expression (unary operation) for the given sample.
893      \return Vector which stores the value of the subexpression for the given sample.
894      \param v A vector to store intermediate results.
895      \param offset Index in v to begin storing results.
896      \param sampleNo Sample number to evaluate.
897      \param roffset (output parameter) the offset in the return vector where the result begins.
898    
899      The return value will be an existing vector so do not deallocate it.
900      If the result is stored in v it should be stored at the offset given.
901      Everything from offset to the end of v should be considered available for this method to use.
902    */
903    DataTypes::ValueType*
904    DataLazy::resolveNP1OUT(ValueType& v, size_t offset, int sampleNo, size_t& roffset) const
905    {
906        // we assume that any collapsing has been done before we get here
907        // since we only have one argument we don't need to think about only
908        // processing single points.
909      if (m_readytype!='E')
910      {
911        throw DataException("Programmer error - resolveNP1OUT should only be called on expanded Data.");
912      }
913        // since we can't write the result over the input, we need a result offset further along
914      size_t subroffset=roffset+m_samplesize;
915      const ValueType* vleft=m_left->resolveSample(v,offset,sampleNo,subroffset);
916      roffset=offset;
917      switch (m_op)
918      {
919        case SYM:
920        DataMaths::symmetric(*vleft,m_left->getShape(),subroffset, v, getShape(), offset);
921        break;
922        case NSYM:
923        DataMaths::nonsymmetric(*vleft,m_left->getShape(),subroffset, v, getShape(), offset);
924        break;
925        default:
926        throw DataException("Programmer error - resolveNP1OUT can not resolve operator "+opToString(m_op)+".");
927      }
928      return &v;
929    }
930    
931    /*
932      \brief Compute the value of the expression (unary operation) for the given sample.
933      \return Vector which stores the value of the subexpression for the given sample.
934      \param v A vector to store intermediate results.
935      \param offset Index in v to begin storing results.
936      \param sampleNo Sample number to evaluate.
937      \param roffset (output parameter) the offset in the return vector where the result begins.
938    
939      The return value will be an existing vector so do not deallocate it.
940      If the result is stored in v it should be stored at the offset given.
941      Everything from offset to the end of v should be considered available for this method to use.
942    */
943    DataTypes::ValueType*
944    DataLazy::resolveNP1OUT_P(ValueType& v, size_t offset, int sampleNo, size_t& roffset) const
945    {
946        // we assume that any collapsing has been done before we get here
947        // since we only have one argument we don't need to think about only
948        // processing single points.
949      if (m_readytype!='E')
950      {
951        throw DataException("Programmer error - resolveNP1OUT_P should only be called on expanded Data.");
952      }
953        // since we can't write the result over the input, we need a result offset further along
954      size_t subroffset=roffset+m_samplesize;
955      const ValueType* vleft=m_left->resolveSample(v,offset,sampleNo,subroffset);
956      roffset=offset;
957      switch (m_op)
958      {
959        case TRACE:
960             DataMaths::trace(*vleft,m_left->getShape(),subroffset, v,getShape(),offset,m_axis_offset);
961        break;
962        case TRANS:
963             DataMaths::transpose(*vleft,m_left->getShape(),subroffset, v,getShape(),offset,m_axis_offset);
964        break;
965        default:
966        throw DataException("Programmer error - resolveNP1OUTP can not resolve operator "+opToString(m_op)+".");
967      }
968      return &v;
969    }
970    
971    
972  #define PROC_OP(TYPE,X)                               \  #define PROC_OP(TYPE,X)                               \
973      for (int i=0;i<steps;++i,resultp+=resultStep) \      for (int j=0;j<onumsteps;++j)\
974      { \      {\
975         tensor_binary_operation##TYPE(chunksize, &((*left)[lroffset]), &((*right)[rroffset]), resultp, X); \        for (int i=0;i<numsteps;++i,resultp+=resultStep) \
976         lroffset+=leftStep; \        { \
977         rroffset+=rightStep; \  LAZYDEBUG(cout << "[left,right]=[" << lroffset << "," << rroffset << "]" << endl;)\
978             tensor_binary_operation< TYPE >(chunksize, &((*left)[lroffset]), &((*right)[rroffset]), resultp, X); \
979             lroffset+=leftstep; \
980             rroffset+=rightstep; \
981          }\
982          lroffset+=oleftstep;\
983          rroffset+=orightstep;\
984      }      }
985    
986  /*  /*
# Line 644  DataLazy::resolveUnary(ValueType& v, siz Line 1007  DataLazy::resolveUnary(ValueType& v, siz
1007  DataTypes::ValueType*  DataTypes::ValueType*
1008  DataLazy::resolveBinary(ValueType& v,  size_t offset, int sampleNo, size_t& roffset) const  DataLazy::resolveBinary(ValueType& v,  size_t offset, int sampleNo, size_t& roffset) const
1009  {  {
1010  cout << "Resolve binary: " << toString() << endl;  LAZYDEBUG(cout << "Resolve binary: " << toString() << endl;)
1011    
1012    size_t lroffset=0, rroffset=0;    // offsets in the left and right result vectors    size_t lroffset=0, rroffset=0;    // offsets in the left and right result vectors
1013      // first work out which of the children are expanded      // first work out which of the children are expanded
1014    bool leftExp=(m_left->m_readytype=='E');    bool leftExp=(m_left->m_readytype=='E');
1015    bool rightExp=(m_right->m_readytype=='E');    bool rightExp=(m_right->m_readytype=='E');
1016    bool bigloops=((leftExp && rightExp) || (!leftExp && !rightExp)); // is processing in single step?    if (!leftExp && !rightExp)
1017    int steps=(bigloops?1:getNumDPPSample());    {
1018    size_t chunksize=(bigloops? m_samplesize : getNoValues());    // if bigloops, pretend the whole sample is a datapoint      throw DataException("Programmer Error - please use collapse if neither argument has type 'E'.");
1019    if (m_left->getRank()!=m_right->getRank())    // need to deal with scalar * ? ops    }
1020    {    bool leftScalar=(m_left->getRank()==0);
1021      EsysAssert((m_left->getRank()==0) || (m_right->getRank()==0), "Error - Ranks must match unless one is 0.");    bool rightScalar=(m_right->getRank()==0);
1022      steps=getNumDPPSample()*max(m_left->getNoValues(),m_right->getNoValues());    if ((m_left->getRank()!=m_right->getRank()) && (!leftScalar && !rightScalar))
1023      chunksize=1;    // for scalar    {
1024    }          throw DataException("resolveBinary - ranks of arguments must match unless one of them is scalar.");
1025    int leftStep=((leftExp && !rightExp)? m_right->getNoValues() : 0);    }
1026    int rightStep=((rightExp && !leftExp)? m_left->getNoValues() : 0);    size_t leftsize=m_left->getNoValues();
1027    int resultStep=max(leftStep,rightStep);   // only one (at most) should be !=0    size_t rightsize=m_right->getNoValues();
1028      size_t chunksize=1;           // how many doubles will be processed in one go
1029      int leftstep=0;       // how far should the left offset advance after each step
1030      int rightstep=0;
1031      int numsteps=0;       // total number of steps for the inner loop
1032      int oleftstep=0;  // the o variables refer to the outer loop
1033      int orightstep=0; // The outer loop is only required in cases where there is an extended scalar
1034      int onumsteps=1;
1035      
1036      bool LES=(leftExp && leftScalar); // Left is an expanded scalar
1037      bool RES=(rightExp && rightScalar);
1038      bool LS=(!leftExp && leftScalar); // left is a single scalar
1039      bool RS=(!rightExp && rightScalar);
1040      bool LN=(!leftExp && !leftScalar);    // left is a single non-scalar
1041      bool RN=(!rightExp && !rightScalar);
1042      bool LEN=(leftExp && !leftScalar);    // left is an expanded non-scalar
1043      bool REN=(rightExp && !rightScalar);
1044    
1045      if ((LES && RES) || (LEN && REN)) // both are Expanded scalars or both are expanded non-scalars
1046      {
1047        chunksize=m_left->getNumDPPSample()*leftsize;
1048        leftstep=0;
1049        rightstep=0;
1050        numsteps=1;
1051      }
1052      else if (LES || RES)
1053      {
1054        chunksize=1;
1055        if (LES)        // left is an expanded scalar
1056        {
1057            if (RS)
1058            {
1059               leftstep=1;
1060               rightstep=0;
1061               numsteps=m_left->getNumDPPSample();
1062            }
1063            else        // RN or REN
1064            {
1065               leftstep=0;
1066               oleftstep=1;
1067               rightstep=1;
1068               orightstep=(RN?-rightsize:0);
1069               numsteps=rightsize;
1070               onumsteps=m_left->getNumDPPSample();
1071            }
1072        }
1073        else        // right is an expanded scalar
1074        {
1075            if (LS)
1076            {
1077               rightstep=1;
1078               leftstep=0;
1079               numsteps=m_right->getNumDPPSample();
1080            }
1081            else
1082            {
1083               rightstep=0;
1084               orightstep=1;
1085               leftstep=1;
1086               oleftstep=(LN?-leftsize:0);
1087               numsteps=leftsize;
1088               onumsteps=m_right->getNumDPPSample();
1089            }
1090        }
1091      }
1092      else  // this leaves (LEN, RS), (LEN, RN) and their transposes
1093      {
1094        if (LEN)    // and Right will be a single value
1095        {
1096            chunksize=rightsize;
1097            leftstep=rightsize;
1098            rightstep=0;
1099            numsteps=m_left->getNumDPPSample();
1100            if (RS)
1101            {
1102               numsteps*=leftsize;
1103            }
1104        }
1105        else    // REN
1106        {
1107            chunksize=leftsize;
1108            rightstep=leftsize;
1109            leftstep=0;
1110            numsteps=m_right->getNumDPPSample();
1111            if (LS)
1112            {
1113               numsteps*=rightsize;
1114            }
1115        }
1116      }
1117    
1118      int resultStep=max(leftstep,rightstep);   // only one (at most) should be !=0
1119      // Get the values of sub-expressions      // Get the values of sub-expressions
1120    const ValueType* left=m_left->resolveSample(v,offset,sampleNo,lroffset);    const ValueType* left=m_left->resolveSample(v,offset,sampleNo,lroffset);
1121    const ValueType* right=m_right->resolveSample(v,offset+m_samplesize,sampleNo,rroffset); // Note    const ValueType* right=m_right->resolveSample(v,offset+m_samplesize,sampleNo,rroffset); // Note
1122      // the right child starts further along.      // the right child starts further along.
1123    LAZYDEBUG(cout << "Post sub calls in " << toString() << endl;)
1124    LAZYDEBUG(cout << "shapes=" << DataTypes::shapeToString(m_left->getShape()) << "," << DataTypes::shapeToString(m_right->getShape()) << endl;)
1125    LAZYDEBUG(cout << "chunksize=" << chunksize << endl << "leftstep=" << leftstep << " rightstep=" << rightstep;)
1126    LAZYDEBUG(cout << " numsteps=" << numsteps << endl << "oleftstep=" << oleftstep << " orightstep=" << orightstep;)
1127    LAZYDEBUG(cout << "onumsteps=" << onumsteps << endl;)
1128    LAZYDEBUG(cout << " DPPS=" << m_left->getNumDPPSample() << "," <<m_right->getNumDPPSample() << endl;)
1129    LAZYDEBUG(cout << "" << LS << RS << LN << RN << LES << RES <<LEN << REN <<   endl;)
1130    double* resultp=&(v[offset]);     // results are stored at the vector offset we recieved    double* resultp=&(v[offset]);     // results are stored at the vector offset we recieved
1131    switch(m_op)    switch(m_op)
1132    {    {
1133      case ADD:      case ADD:
1134          PROC_OP(/**/,plus<double>());          PROC_OP(NO_ARG,plus<double>());
1135      break;      break;
1136      case SUB:      case SUB:
1137      PROC_OP(/**/,minus<double>());      PROC_OP(NO_ARG,minus<double>());
1138      break;      break;
1139      case MUL:      case MUL:
1140      PROC_OP(/**/,multiplies<double>());      PROC_OP(NO_ARG,multiplies<double>());
1141      break;      break;
1142      case DIV:      case DIV:
1143      PROC_OP(/**/,divides<double>());      PROC_OP(NO_ARG,divides<double>());
1144      break;      break;
1145      case POW:      case POW:
1146         PROC_OP(<double (double,double)>,::pow);         PROC_OP(double (double,double),::pow);
1147      break;      break;
1148      default:      default:
1149      throw DataException("Programmer error - resolveBinary can not resolve operator "+opToString(m_op)+".");      throw DataException("Programmer error - resolveBinary can not resolve operator "+opToString(m_op)+".");
# Line 694  cout << "Resolve binary: " << toString() Line 1155  cout << "Resolve binary: " << toString()
1155    
1156    
1157  /*  /*
1158      \brief Compute the value of the expression (tensor product) for the given sample.
1159      \return Vector which stores the value of the subexpression for the given sample.
1160      \param v A vector to store intermediate results.
1161      \param offset Index in v to begin storing results.
1162      \param sampleNo Sample number to evaluate.
1163      \param roffset (output parameter) the offset in the return vector where the result begins.
1164    
1165      The return value will be an existing vector so do not deallocate it.
1166      If the result is stored in v it should be stored at the offset given.
1167      Everything from offset to the end of v should be considered available for this method to use.
1168    */
1169    // This method assumes that any subexpressions which evaluate to Constant or Tagged Data
1170    // have already been collapsed to IDENTITY. So we must have at least one expanded child.
1171    // unlike the other resolve helpers, we must treat these datapoints separately.
1172    DataTypes::ValueType*
1173    DataLazy::resolveTProd(ValueType& v,  size_t offset, int sampleNo, size_t& roffset) const
1174    {
1175    LAZYDEBUG(cout << "Resolve TensorProduct: " << toString() << endl;)
1176    
1177      size_t lroffset=0, rroffset=0;    // offsets in the left and right result vectors
1178        // first work out which of the children are expanded
1179      bool leftExp=(m_left->m_readytype=='E');
1180      bool rightExp=(m_right->m_readytype=='E');
1181      int steps=getNumDPPSample();
1182      int leftStep=((leftExp && !rightExp)? m_right->getNoValues() : 0);
1183      int rightStep=((rightExp && !leftExp)? m_left->getNoValues() : 0);
1184      int resultStep=max(leftStep,rightStep);   // only one (at most) should be !=0
1185        // Get the values of sub-expressions (leave a gap of one sample for the result).
1186      const ValueType* left=m_left->resolveSample(v,offset+m_samplesize,sampleNo,lroffset);
1187      const ValueType* right=m_right->resolveSample(v,offset+2*m_samplesize,sampleNo,rroffset);
1188      double* resultp=&(v[offset]);     // results are stored at the vector offset we recieved
1189      switch(m_op)
1190      {
1191        case PROD:
1192        for (int i=0;i<steps;++i,resultp+=resultStep)
1193        {
1194              const double *ptr_0 = &((*left)[lroffset]);
1195              const double *ptr_1 = &((*right)[rroffset]);
1196              matrix_matrix_product(m_SL, m_SM, m_SR, ptr_0, ptr_1, resultp, m_transpose);
1197          lroffset+=leftStep;
1198          rroffset+=rightStep;
1199        }
1200        break;
1201        default:
1202        throw DataException("Programmer error - resolveTProduct can not resolve operator "+opToString(m_op)+".");
1203      }
1204      roffset=offset;
1205      return &v;
1206    }
1207    
1208    
1209    
1210    /*
1211    \brief Compute the value of the expression for the given sample.    \brief Compute the value of the expression for the given sample.
1212    \return Vector which stores the value of the subexpression for the given sample.    \return Vector which stores the value of the subexpression for the given sample.
1213    \param v A vector to store intermediate results.    \param v A vector to store intermediate results.
# Line 712  cout << "Resolve binary: " << toString() Line 1226  cout << "Resolve binary: " << toString()
1226  const DataTypes::ValueType*  const DataTypes::ValueType*
1227  DataLazy::resolveSample(ValueType& v, size_t offset, int sampleNo, size_t& roffset)  DataLazy::resolveSample(ValueType& v, size_t offset, int sampleNo, size_t& roffset)
1228  {  {
1229  cout << "Resolve sample " << toString() << endl;  LAZYDEBUG(cout << "Resolve sample " << toString() << endl;)
1230      // collapse so we have a 'E' node or an IDENTITY for some other type      // collapse so we have a 'E' node or an IDENTITY for some other type
1231    if (m_readytype!='E' && m_op!=IDENTITY)    if (m_readytype!='E' && m_op!=IDENTITY)
1232    {    {
# Line 724  cout << "Resolve sample " << toString() Line 1238  cout << "Resolve sample " << toString()
1238      if (m_readytype=='C')      if (m_readytype=='C')
1239      {      {
1240      roffset=0;      roffset=0;
1241    LAZYDEBUG(cout << "Finish  sample " << toString() << endl;)
1242      return &(vec);      return &(vec);
1243      }      }
1244      roffset=m_id->getPointOffset(sampleNo, 0);      roffset=m_id->getPointOffset(sampleNo, 0);
1245    LAZYDEBUG(cout << "Finish  sample " << toString() << endl;)
1246      return &(vec);      return &(vec);
1247    }    }
1248    if (m_readytype!='E')    if (m_readytype!='E')
# Line 735  cout << "Resolve sample " << toString() Line 1251  cout << "Resolve sample " << toString()
1251    }    }
1252    switch (getOpgroup(m_op))    switch (getOpgroup(m_op))
1253    {    {
1254    case G_UNARY: return resolveUnary(v, offset,sampleNo,roffset);    case G_UNARY:
1255      case G_UNARY_P: return resolveUnary(v, offset,sampleNo,roffset);
1256    case G_BINARY: return resolveBinary(v, offset,sampleNo,roffset);    case G_BINARY: return resolveBinary(v, offset,sampleNo,roffset);
1257      case G_NP1OUT: return resolveNP1OUT(v, offset, sampleNo,roffset);
1258      case G_NP1OUT_P: return resolveNP1OUT_P(v, offset, sampleNo,roffset);
1259      case G_TENSORPROD: return resolveTProd(v,offset, sampleNo,roffset);
1260    default:    default:
1261      throw DataException("Programmer Error - resolveSample does not know how to process "+opToString(m_op)+".");      throw DataException("Programmer Error - resolveSample does not know how to process "+opToString(m_op)+".");
1262    }    }
1263    
1264  }  }
1265    
1266    
# Line 749  DataReady_ptr Line 1270  DataReady_ptr
1270  DataLazy::resolve()  DataLazy::resolve()
1271  {  {
1272    
1273  cout << "Sample size=" << m_samplesize << endl;  LAZYDEBUG(cout << "Sample size=" << m_samplesize << endl;)
1274  cout << "Buffers=" << m_buffsRequired << endl;  LAZYDEBUG(cout << "Buffers=" << m_buffsRequired << endl;)
1275    
1276    if (m_readytype!='E')     // if the whole sub-expression is Constant or Tagged, then evaluate it normally    if (m_readytype!='E')     // if the whole sub-expression is Constant or Tagged, then evaluate it normally
1277    {    {
# Line 761  cout << "Buffers=" << m_buffsRequired << Line 1282  cout << "Buffers=" << m_buffsRequired <<
1282      return m_id;      return m_id;
1283    }    }
1284      // from this point on we must have m_op!=IDENTITY and m_readytype=='E'      // from this point on we must have m_op!=IDENTITY and m_readytype=='E'
1285    size_t threadbuffersize=m_samplesize*(max(1,m_buffsRequired));    // Each thread needs to have enough    size_t threadbuffersize=m_maxsamplesize*(max(1,m_buffsRequired)); // Each thread needs to have enough
1286      // storage to evaluate its expression      // storage to evaluate its expression
1287    int numthreads=1;    int numthreads=1;
1288  #ifdef _OPENMP  #ifdef _OPENMP
1289    numthreads=getNumberOfThreads();    numthreads=getNumberOfThreads();
   int threadnum=0;  
1290  #endif  #endif
1291    ValueType v(numthreads*threadbuffersize);    ValueType v(numthreads*threadbuffersize);
1292  cout << "Buffer created with size=" << v.size() << endl;  LAZYDEBUG(cout << "Buffer created with size=" << v.size() << endl;)
1293    DataExpanded* result=new DataExpanded(getFunctionSpace(),getShape(),  ValueType(getNoValues()));    DataExpanded* result=new DataExpanded(getFunctionSpace(),getShape(),  ValueType(getNoValues()));
1294    ValueType& resvec=result->getVector();    ValueType& resvec=result->getVector();
1295    DataReady_ptr resptr=DataReady_ptr(result);    DataReady_ptr resptr=DataReady_ptr(result);
# Line 778  cout << "Buffer created with size=" << v Line 1298  cout << "Buffer created with size=" << v
1298    int totalsamples=getNumSamples();    int totalsamples=getNumSamples();
1299    const ValueType* res=0;   // Vector storing the answer    const ValueType* res=0;   // Vector storing the answer
1300    size_t resoffset=0;       // where in the vector to find the answer    size_t resoffset=0;       // where in the vector to find the answer
1301    #pragma omp parallel for private(sample,resoffset,outoffset,threadnum,res) schedule(static)  LAZYDEBUG(cout << "Total number of samples=" <<totalsamples << endl;)
1302      #pragma omp parallel for private(sample,resoffset,outoffset,res) schedule(static)
1303    for (sample=0;sample<totalsamples;++sample)    for (sample=0;sample<totalsamples;++sample)
1304    {    {
1305  cout << "################################# " << sample << endl;  LAZYDEBUG(cout << "################################# " << sample << endl;)
1306  #ifdef _OPENMP  #ifdef _OPENMP
1307      res=resolveSample(v,threadbuffersize*omp_get_thread_num(),sample,resoffset);      res=resolveSample(v,threadbuffersize*omp_get_thread_num(),sample,resoffset);
1308  #else  #else
1309      res=resolveSample(v,0,sample,resoffset);   // res would normally be v, but not if its a single IDENTITY op.      res=resolveSample(v,0,sample,resoffset);   // res would normally be v, but not if its a single IDENTITY op.
1310  #endif  #endif
1311  cerr << "-------------------------------- " << endl;  LAZYDEBUG(cerr << "-------------------------------- " << endl;)
1312      outoffset=result->getPointOffset(sample,0);      outoffset=result->getPointOffset(sample,0);
1313  cerr << "offset=" << outoffset << endl;  LAZYDEBUG(cerr << "offset=" << outoffset << endl;)
1314      for (unsigned int i=0;i<m_samplesize;++i,++outoffset,++resoffset)   // copy values into the output vector      for (unsigned int i=0;i<m_samplesize;++i,++outoffset,++resoffset)   // copy values into the output vector
1315      {      {
1316      resvec[outoffset]=(*res)[resoffset];      resvec[outoffset]=(*res)[resoffset];
1317      }      }
1318  cerr << "*********************************" << endl;  LAZYDEBUG(cerr << "*********************************" << endl;)
1319    }    }
1320    return resptr;    return resptr;
1321  }  }
# Line 841  DataLazy::intoString(ostringstream& oss) Line 1362  DataLazy::intoString(ostringstream& oss)
1362      oss << ')';      oss << ')';
1363      break;      break;
1364    case G_UNARY:    case G_UNARY:
1365      case G_UNARY_P:
1366      case G_NP1OUT:
1367      case G_NP1OUT_P:
1368      oss << opToString(m_op) << '(';      oss << opToString(m_op) << '(';
1369      m_left->intoString(oss);      m_left->intoString(oss);
1370      oss << ')';      oss << ')';
1371      break;      break;
1372      case G_TENSORPROD:
1373        oss << opToString(m_op) << '(';
1374        m_left->intoString(oss);
1375        oss << ", ";
1376        m_right->intoString(oss);
1377        oss << ')';
1378        break;
1379    default:    default:
1380      oss << "UNKNOWN";      oss << "UNKNOWN";
1381    }    }
# Line 858  DataLazy::deepCopy() Line 1389  DataLazy::deepCopy()
1389    case G_IDENTITY:  return new DataLazy(m_id->deepCopy()->getPtr());    case G_IDENTITY:  return new DataLazy(m_id->deepCopy()->getPtr());
1390    case G_UNARY: return new DataLazy(m_left->deepCopy()->getPtr(),m_op);    case G_UNARY: return new DataLazy(m_left->deepCopy()->getPtr(),m_op);
1391    case G_BINARY:    return new DataLazy(m_left->deepCopy()->getPtr(),m_right->deepCopy()->getPtr(),m_op);    case G_BINARY:    return new DataLazy(m_left->deepCopy()->getPtr(),m_right->deepCopy()->getPtr(),m_op);
1392      case G_NP1OUT: return new DataLazy(m_left->deepCopy()->getPtr(), m_right->deepCopy()->getPtr(),m_op);
1393      case G_TENSORPROD: return new DataLazy(m_left->deepCopy()->getPtr(), m_right->deepCopy()->getPtr(), m_op, m_axis_offset, m_transpose);
1394    default:    default:
1395      throw DataException("Programmer error - do not know how to deepcopy operator "+opToString(m_op)+".");      throw DataException("Programmer error - do not know how to deepcopy operator "+opToString(m_op)+".");
1396    }    }
1397  }  }
1398    
1399    
1400    // There is no single, natural interpretation of getLength on DataLazy.
1401    // Instances of DataReady can look at the size of their vectors.
1402    // For lazy though, it could be the size the data would be if it were resolved;
1403    // or it could be some function of the lengths of the DataReady instances which
1404    // form part of the expression.
1405    // Rather than have people making assumptions, I have disabled the method.
1406  DataTypes::ValueType::size_type  DataTypes::ValueType::size_type
1407  DataLazy::getLength() const  DataLazy::getLength() const
1408  {  {
1409    return m_length;    throw DataException("getLength() does not make sense for lazy data.");
1410  }  }
1411    
1412    

Legend:
Removed from v.1993  
changed lines
  Added in v.2152

  ViewVC Help
Powered by ViewVC 1.1.26