/[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

revision 2514 by jfenwick, Fri Jul 3 00:57:45 2009 UTC revision 4154 by jfenwick, Tue Jan 22 09:30:23 2013 UTC
# Line 1  Line 1 
1    
2  /*******************************************************  /*****************************************************************************
3  *  *
4  * Copyright (c) 2003-2008 by University of Queensland  * Copyright (c) 2003-2013 by University of Queensland
5  * Earth Systems Science Computational Center (ESSCC)  * http://www.uq.edu.au
 * http://www.uq.edu.au/esscc  
6  *  *
7  * Primary Business: Queensland, Australia  * Primary Business: Queensland, Australia
8  * Licensed under the Open Software License version 3.0  * Licensed under the Open Software License version 3.0
9  * http://www.opensource.org/licenses/osl-3.0.php  * http://www.opensource.org/licenses/osl-3.0.php
10  *  *
11  *******************************************************/  * Development until 2012 by Earth Systems Science Computational Center (ESSCC)
12    * Development since 2012 by School of Earth Sciences
13    *
14    *****************************************************************************/
15    
16    
17  #include "DataLazy.h"  #include "DataLazy.h"
18  #ifdef USE_NETCDF  #include "esysUtils/Esys_MPI.h"
 #include <netcdfcpp.h>  
 #endif  
 #ifdef PASO_MPI  
 #include <mpi.h>  
 #endif  
19  #ifdef _OPENMP  #ifdef _OPENMP
20  #include <omp.h>  #include <omp.h>
21  #endif  #endif
# Line 30  Line 27 
27    
28  #include "EscriptParams.h"  #include "EscriptParams.h"
29    
30    #ifdef USE_NETCDF
31    #include <netcdfcpp.h>
32    #endif
33    
34  #include <iomanip>      // for some fancy formatting in debug  #include <iomanip>      // for some fancy formatting in debug
35    
36  // #define LAZYDEBUG(X) if (privdebug){X;}  // #define LAZYDEBUG(X) if (privdebug){X;}
# Line 44  bool privdebug=false; Line 45  bool privdebug=false;
45    
46  // #define SIZELIMIT if ((m_height>escript::escriptParams.getTOO_MANY_LEVELS()) || (m_children>escript::escriptParams.getTOO_MANY_NODES())) {cerr << "\n!!!!!!! SIZE LIMIT EXCEEDED " << m_children << ";" << m_height << endl << toString() << endl;resolveToIdentity();}  // #define SIZELIMIT if ((m_height>escript::escriptParams.getTOO_MANY_LEVELS()) || (m_children>escript::escriptParams.getTOO_MANY_NODES())) {cerr << "\n!!!!!!! SIZE LIMIT EXCEEDED " << m_children << ";" << m_height << endl << toString() << endl;resolveToIdentity();}
47    
48  #define SIZELIMIT if ((m_height>escript::escriptParams.getTOO_MANY_LEVELS()) || (m_children>escript::escriptParams.getTOO_MANY_NODES())) {resolveToIdentity();}  // #define SIZELIMIT if ((m_height>escript::escriptParams.getTOO_MANY_LEVELS()) || (m_children>escript::escriptParams.getTOO_MANY_NODES())) {cerr << "SIZE LIMIT EXCEEDED " << m_height << endl;resolveToIdentity();}
49    
50    
51    #define SIZELIMIT if (m_height>escript::escriptParams.getTOO_MANY_LEVELS())  {if (escript::escriptParams.getLAZY_VERBOSE()){cerr << "SIZE LIMIT EXCEEDED height=" << m_height << endl;}resolveToIdentity();}
52    
53  /*  /*
54  How does DataLazy work?  How does DataLazy work?
55  ~~~~~~~~~~~~~~~~~~~~~~~  ~~~~~~~~~~~~~~~~~~~~~~~
# Line 109  namespace escript Line 112  namespace escript
112  namespace  namespace
113  {  {
114    
115    
116    // enabling this will print out when ever the maximum stacksize used by resolve increases
117    // it assumes _OPENMP is also in use
118    //#define LAZY_STACK_PROF
119    
120    
121    
122    #ifndef _OPENMP
123      #ifdef LAZY_STACK_PROF
124      #undef LAZY_STACK_PROF
125      #endif
126    #endif
127    
128    
129    #ifdef LAZY_STACK_PROF
130    std::vector<void*> stackstart(getNumberOfThreads());
131    std::vector<void*> stackend(getNumberOfThreads());
132    size_t maxstackuse=0;
133    #endif
134    
135  enum ES_opgroup  enum ES_opgroup
136  {  {
137     G_UNKNOWN,     G_UNKNOWN,
# Line 119  enum ES_opgroup Line 142  enum ES_opgroup
142     G_NP1OUT,        // non-pointwise op with one output     G_NP1OUT,        // non-pointwise op with one output
143     G_NP1OUT_P,      // non-pointwise op with one output requiring a parameter     G_NP1OUT_P,      // non-pointwise op with one output requiring a parameter
144     G_TENSORPROD,    // general tensor product     G_TENSORPROD,    // general tensor product
145     G_NP1OUT_2P      // non-pointwise op with one output requiring two params     G_NP1OUT_2P,     // non-pointwise op with one output requiring two params
146       G_REDUCTION,     // non-pointwise unary op with a scalar output
147       G_CONDEVAL
148  };  };
149    
150    
# Line 134  string ES_opstrings[]={"UNKNOWN","IDENTI Line 159  string ES_opstrings[]={"UNKNOWN","IDENTI
159              "symmetric","nonsymmetric",              "symmetric","nonsymmetric",
160              "prod",              "prod",
161              "transpose", "trace",              "transpose", "trace",
162              "swapaxes"};              "swapaxes",
163  int ES_opcount=41;              "minval", "maxval",
164                "condEval"};
165    int ES_opcount=44;
166  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,
167              G_UNARY,G_UNARY,G_UNARY, //10              G_UNARY,G_UNARY,G_UNARY, //10
168              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
# Line 145  ES_opgroup opgroups[]={G_UNKNOWN,G_IDENT Line 172  ES_opgroup opgroups[]={G_UNKNOWN,G_IDENT
172              G_NP1OUT,G_NP1OUT,              G_NP1OUT,G_NP1OUT,
173              G_TENSORPROD,              G_TENSORPROD,
174              G_NP1OUT_P, G_NP1OUT_P,              G_NP1OUT_P, G_NP1OUT_P,
175              G_NP1OUT_2P};              G_NP1OUT_2P,
176                G_REDUCTION, G_REDUCTION,
177                G_CONDEVAL};
178  inline  inline
179  ES_opgroup  ES_opgroup
180  getOpgroup(ES_optype op)  getOpgroup(ES_optype op)
# Line 190  resultShape(DataAbstract_ptr left, DataA Line 219  resultShape(DataAbstract_ptr left, DataA
219        {        {
220          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.");
221        }        }
222    
223        if (left->getRank()==0)   // we need to allow scalar * anything        if (left->getRank()==0)   // we need to allow scalar * anything
224        {        {
225          return right->getShape();          return right->getShape();
# Line 217  resultShape(DataAbstract_ptr left, ES_op Line 247  resultShape(DataAbstract_ptr left, ES_op
247          int rank=left->getRank();          int rank=left->getRank();
248          if (axis_offset<0 || axis_offset>rank)          if (axis_offset<0 || axis_offset>rank)
249          {          {
250                 throw DataException("Error - Data::transpose must have 0 <= axis_offset <= rank=" + rank);              stringstream e;
251              }              e << "Error - Data::transpose must have 0 <= axis_offset <= rank=" << rank;
252              for (int i=0; i<rank; i++)              throw DataException(e.str());
253            }
254            for (int i=0; i<rank; i++)
255          {          {
256             int index = (axis_offset+i)%rank;             int index = (axis_offset+i)%rank;
257                 sh.push_back(s[index]); // Append to new shape             sh.push_back(s[index]); // Append to new shape
258              }          }
259          return sh;          return sh;
260         }         }
261      break;      break;
# Line 302  SwapShape(DataAbstract_ptr left, const i Line 334  SwapShape(DataAbstract_ptr left, const i
334          throw DataException("Error - Data::swapaxes argument must have at least rank 2.");          throw DataException("Error - Data::swapaxes argument must have at least rank 2.");
335       }       }
336       if (axis0<0 || axis0>rank-1) {       if (axis0<0 || axis0>rank-1) {
337          throw DataException("Error - Data::swapaxes: axis0 must be between 0 and rank-1=" + rank-1);          stringstream e;
338            e << "Error - Data::swapaxes: axis0 must be between 0 and rank-1=" << (rank-1);
339            throw DataException(e.str());
340       }       }
341       if (axis1<0 || axis1>rank-1) {       if (axis1<0 || axis1>rank-1) {
342           throw DataException("Error - Data::swapaxes: axis1 must be between 0 and rank-1=" + rank-1);          stringstream e;
343            e << "Error - Data::swapaxes: axis1 must be between 0 and rank-1=" << (rank-1);
344            throw DataException(e.str());
345       }       }
346       if (axis0 == axis1) {       if (axis0 == axis1) {
347           throw DataException("Error - Data::swapaxes: axis indices must be different.");           throw DataException("Error - Data::swapaxes: axis indices must be different.");
# Line 394  GTPShape(DataAbstract_ptr left, DataAbst Line 430  GTPShape(DataAbstract_ptr left, DataAbst
430    return shape2;    return shape2;
431  }  }
432    
 // determine the number of samples requires to evaluate an expression combining left and right  
 // NP1OUT needs an extra buffer because we can't write the answers over the top of the input.  
 // The same goes for G_TENSORPROD  
 // It might seem that pointwise binary ops (G_BINARY) could be written over the top of the lefts.  
 // This would be true were it not for the possibility that the LHS could be a scalar which needs to be examined  
 // multiple times  
 int  
 calcBuffs(const DataLazy_ptr& left, const DataLazy_ptr& right, ES_optype op)  
 {  
    switch(getOpgroup(op))  
    {  
    case G_IDENTITY: return 1;  
    case G_BINARY: return 1+max(left->getBuffsRequired(),right->getBuffsRequired()+1);  
    case G_UNARY:  
    case G_UNARY_P: return max(left->getBuffsRequired(),1);  
    case G_NP1OUT: return 1+max(left->getBuffsRequired(),1);  
    case G_NP1OUT_P: return 1+max(left->getBuffsRequired(),1);  
    case G_TENSORPROD: return 1+max(left->getBuffsRequired(),right->getBuffsRequired()+1);  
    case G_NP1OUT_2P: return 1+max(left->getBuffsRequired(),1);  
    default:  
     throw DataException("Programmer Error - attempt to calcBuffs() for operator "+opToString(op)+".");  
    }  
 }  
   
   
433  }   // end anonymous namespace  }   // end anonymous namespace
434    
435    
# Line 434  opToString(ES_optype op) Line 445  opToString(ES_optype op)
445    return ES_opstrings[op];    return ES_opstrings[op];
446  }  }
447    
 #ifdef LAZY_NODE_STORAGE  
448  void DataLazy::LazyNodeSetup()  void DataLazy::LazyNodeSetup()
449  {  {
450  #ifdef _OPENMP  #ifdef _OPENMP
# Line 451  void DataLazy::LazyNodeSetup() Line 461  void DataLazy::LazyNodeSetup()
461      m_sampleids[0]=-1;      m_sampleids[0]=-1;
462  #endif  // _OPENMP  #endif  // _OPENMP
463  }  }
 #endif   // LAZY_NODE_STORAGE  
464    
465    
466  // Creates an identity node  // Creates an identity node
467  DataLazy::DataLazy(DataAbstract_ptr p)  DataLazy::DataLazy(DataAbstract_ptr p)
468      : parent(p->getFunctionSpace(),p->getShape())      : parent(p->getFunctionSpace(),p->getShape())
 #ifdef LAZY_NODE_STORAGE  
469      ,m_sampleids(0),      ,m_sampleids(0),
470      m_samples(1)      m_samples(1)
 #endif  
471  {  {
472     if (p->isLazy())     if (p->isLazy())
473     {     {
# Line 480  LAZYDEBUG(cout << "(1)Lazy created with Line 487  LAZYDEBUG(cout << "(1)Lazy created with
487  }  }
488    
489  DataLazy::DataLazy(DataAbstract_ptr left, ES_optype op)  DataLazy::DataLazy(DataAbstract_ptr left, ES_optype op)
490      : parent(left->getFunctionSpace(),left->getShape()),      : parent(left->getFunctionSpace(),(getOpgroup(op)!=G_REDUCTION)?left->getShape():DataTypes::scalarShape),
491      m_op(op),      m_op(op),
492      m_axis_offset(0),      m_axis_offset(0),
493      m_transpose(0),      m_transpose(0),
494      m_SL(0), m_SM(0), m_SR(0)      m_SL(0), m_SM(0), m_SR(0)
495  {  {
496     if ((getOpgroup(op)!=G_UNARY) && (getOpgroup(op)!=G_NP1OUT))     if ((getOpgroup(op)!=G_UNARY) && (getOpgroup(op)!=G_NP1OUT) && (getOpgroup(op)!=G_REDUCTION))
497     {     {
498      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.");
499     }     }
# Line 502  DataLazy::DataLazy(DataAbstract_ptr left Line 509  DataLazy::DataLazy(DataAbstract_ptr left
509     }     }
510     m_readytype=lleft->m_readytype;     m_readytype=lleft->m_readytype;
511     m_left=lleft;     m_left=lleft;
    m_buffsRequired=calcBuffs(m_left, m_right,m_op); // yeah m_right will be null at this point  
512     m_samplesize=getNumDPPSample()*getNoValues();     m_samplesize=getNumDPPSample()*getNoValues();
    m_maxsamplesize=max(m_samplesize,m_left->getMaxSampleSize());  
513     m_children=m_left->m_children+1;     m_children=m_left->m_children+1;
514     m_height=m_left->m_height+1;     m_height=m_left->m_height+1;
 #ifdef LAZY_NODE_STORAGE  
515     LazyNodeSetup();     LazyNodeSetup();
 #endif  
516     SIZELIMIT     SIZELIMIT
517  }  }
518    
# Line 576  LAZYDEBUG(cout << "Right " << right.get( Line 579  LAZYDEBUG(cout << "Right " << right.get(
579      m_readytype='C';      m_readytype='C';
580     }     }
581     m_samplesize=getNumDPPSample()*getNoValues();     m_samplesize=getNumDPPSample()*getNoValues();
    m_maxsamplesize=max(max(m_samplesize,m_right->getMaxSampleSize()),m_left->getMaxSampleSize());    
    m_buffsRequired=calcBuffs(m_left, m_right,m_op);  
582     m_children=m_left->m_children+m_right->m_children+2;     m_children=m_left->m_children+m_right->m_children+2;
583     m_height=max(m_left->m_height,m_right->m_height)+1;     m_height=max(m_left->m_height,m_right->m_height)+1;
 #ifdef LAZY_NODE_STORAGE  
584     LazyNodeSetup();     LazyNodeSetup();
 #endif  
585     SIZELIMIT     SIZELIMIT
586  LAZYDEBUG(cout << "(3)Lazy created with " << m_samplesize << endl;)  LAZYDEBUG(cout << "(3)Lazy created with " << m_samplesize << endl;)
587  }  }
# Line 646  DataLazy::DataLazy(DataAbstract_ptr left Line 645  DataLazy::DataLazy(DataAbstract_ptr left
645      m_readytype='C';      m_readytype='C';
646     }     }
647     m_samplesize=getNumDPPSample()*getNoValues();     m_samplesize=getNumDPPSample()*getNoValues();
    m_maxsamplesize=max(max(m_samplesize,m_right->getMaxSampleSize()),m_left->getMaxSampleSize());    
    m_buffsRequired=calcBuffs(m_left, m_right,m_op);  
648     m_children=m_left->m_children+m_right->m_children+2;     m_children=m_left->m_children+m_right->m_children+2;
649     m_height=max(m_left->m_height,m_right->m_height)+1;     m_height=max(m_left->m_height,m_right->m_height)+1;
 #ifdef LAZY_NODE_STORAGE  
650     LazyNodeSetup();     LazyNodeSetup();
 #endif  
651     SIZELIMIT     SIZELIMIT
652  LAZYDEBUG(cout << "(4)Lazy created with " << m_samplesize << endl;)  LAZYDEBUG(cout << "(4)Lazy created with " << m_samplesize << endl;)
653  }  }
# Line 680  DataLazy::DataLazy(DataAbstract_ptr left Line 675  DataLazy::DataLazy(DataAbstract_ptr left
675     }     }
676     m_readytype=lleft->m_readytype;     m_readytype=lleft->m_readytype;
677     m_left=lleft;     m_left=lleft;
    m_buffsRequired=calcBuffs(m_left, m_right,m_op); // yeah m_right will be null at this point  
678     m_samplesize=getNumDPPSample()*getNoValues();     m_samplesize=getNumDPPSample()*getNoValues();
    m_maxsamplesize=max(m_samplesize,m_left->getMaxSampleSize());  
679     m_children=m_left->m_children+1;     m_children=m_left->m_children+1;
680     m_height=m_left->m_height+1;     m_height=m_left->m_height+1;
 #ifdef LAZY_NODE_STORAGE  
681     LazyNodeSetup();     LazyNodeSetup();
 #endif  
682     SIZELIMIT     SIZELIMIT
683  LAZYDEBUG(cout << "(5)Lazy created with " << m_samplesize << endl;)  LAZYDEBUG(cout << "(5)Lazy created with " << m_samplesize << endl;)
684  }  }
# Line 714  DataLazy::DataLazy(DataAbstract_ptr left Line 705  DataLazy::DataLazy(DataAbstract_ptr left
705     }     }
706     m_readytype=lleft->m_readytype;     m_readytype=lleft->m_readytype;
707     m_left=lleft;     m_left=lleft;
    m_buffsRequired=calcBuffs(m_left, m_right,m_op); // yeah m_right will be null at this point  
708     m_samplesize=getNumDPPSample()*getNoValues();     m_samplesize=getNumDPPSample()*getNoValues();
    m_maxsamplesize=max(m_samplesize,m_left->getMaxSampleSize());  
709     m_children=m_left->m_children+1;     m_children=m_left->m_children+1;
710     m_height=m_left->m_height+1;     m_height=m_left->m_height+1;
 #ifdef LAZY_NODE_STORAGE  
711     LazyNodeSetup();     LazyNodeSetup();
 #endif  
712     SIZELIMIT     SIZELIMIT
713  LAZYDEBUG(cout << "(6)Lazy created with " << m_samplesize << endl;)  LAZYDEBUG(cout << "(6)Lazy created with " << m_samplesize << endl;)
714  }  }
# Line 749  DataLazy::DataLazy(DataAbstract_ptr left Line 736  DataLazy::DataLazy(DataAbstract_ptr left
736     }     }
737     m_readytype=lleft->m_readytype;     m_readytype=lleft->m_readytype;
738     m_left=lleft;     m_left=lleft;
    m_buffsRequired=calcBuffs(m_left, m_right,m_op); // yeah m_right will be null at this point  
739     m_samplesize=getNumDPPSample()*getNoValues();     m_samplesize=getNumDPPSample()*getNoValues();
    m_maxsamplesize=max(m_samplesize,m_left->getMaxSampleSize());  
740     m_children=m_left->m_children+1;     m_children=m_left->m_children+1;
741     m_height=m_left->m_height+1;     m_height=m_left->m_height+1;
 #ifdef LAZY_NODE_STORAGE  
742     LazyNodeSetup();     LazyNodeSetup();
 #endif  
743     SIZELIMIT     SIZELIMIT
744  LAZYDEBUG(cout << "(7)Lazy created with " << m_samplesize << endl;)  LAZYDEBUG(cout << "(7)Lazy created with " << m_samplesize << endl;)
745  }  }
746    
747  DataLazy::~DataLazy()  
748    namespace
749  {  {
 #ifdef LAZY_NODE_SETUP  
    delete[] m_sampleids;  
    delete[] m_samples;  
 #endif  
 }  
750    
751        inline int max3(int a, int b, int c)
752        {
753        int t=(a>b?a:b);
754        return (t>c?t:c);
755    
756  int      }
 DataLazy::getBuffsRequired() const  
 {  
     return m_buffsRequired;  
757  }  }
758    
759    DataLazy::DataLazy(DataAbstract_ptr mask, DataAbstract_ptr left, DataAbstract_ptr right/*, double tol*/)
760  size_t      : parent(left->getFunctionSpace(), left->getShape()),
761  DataLazy::getMaxSampleSize() const      m_op(CONDEVAL),
762        m_axis_offset(0),
763        m_transpose(0),
764        m_tol(0)
765  {  {
766      return m_maxsamplesize;  
767       DataLazy_ptr lmask;
768       DataLazy_ptr lleft;
769       DataLazy_ptr lright;
770       if (!mask->isLazy())
771       {
772        lmask=DataLazy_ptr(new DataLazy(mask));
773       }
774       else
775       {
776        lmask=dynamic_pointer_cast<DataLazy>(mask);
777       }
778       if (!left->isLazy())
779       {
780        lleft=DataLazy_ptr(new DataLazy(left));
781       }
782       else
783       {
784        lleft=dynamic_pointer_cast<DataLazy>(left);
785       }
786       if (!right->isLazy())
787       {
788        lright=DataLazy_ptr(new DataLazy(right));
789       }
790       else
791       {
792        lright=dynamic_pointer_cast<DataLazy>(right);
793       }
794       m_readytype=lmask->m_readytype;
795       if ((lleft->m_readytype!=lright->m_readytype) || (lmask->m_readytype!=lleft->m_readytype))
796       {
797        throw DataException("Programmer Error - condEval arguments must have the same readytype");
798       }
799       m_left=lleft;
800       m_right=lright;
801       m_mask=lmask;
802       m_samplesize=getNumDPPSample()*getNoValues();
803       m_children=m_left->m_children+m_right->m_children+m_mask->m_children+1;
804       m_height=max3(m_left->m_height,m_right->m_height,m_mask->m_height)+1;
805       LazyNodeSetup();
806       SIZELIMIT
807    LAZYDEBUG(cout << "(8)Lazy created with " << m_samplesize << endl;)
808  }  }
809    
810    
811    
812  size_t  DataLazy::~DataLazy()
 DataLazy::getSampleBufferSize() const  
813  {  {
814      return m_maxsamplesize*(max(1,m_buffsRequired));     delete[] m_sampleids;
815  }  }
816    
817    
818  /*  /*
819    \brief Evaluates the expression using methods on Data.    \brief Evaluates the expression using methods on Data.
820    This does the work for the collapse method.    This does the work for the collapse method.
# Line 933  DataLazy::collapseToReady() Line 957  DataLazy::collapseToReady()
957      case SWAP:      case SWAP:
958      result=left.swapaxes(m_axis_offset, m_transpose);      result=left.swapaxes(m_axis_offset, m_transpose);
959      break;      break;
960        case MINVAL:
961        result=left.minval();
962        break;
963        case MAXVAL:
964        result=left.minval();
965        break;
966      default:      default:
967      throw DataException("Programmer error - collapseToReady does not know how to resolve operator "+opToString(m_op)+".");      throw DataException("Programmer error - collapseToReady does not know how to resolve operator "+opToString(m_op)+".");
968    }    }
# Line 960  DataLazy::collapse() Line 990  DataLazy::collapse()
990    m_op=IDENTITY;    m_op=IDENTITY;
991  }  }
992    
 /*  
   \brief Compute the value of the expression (unary operation) for the given sample.  
   \return Vector which stores the value of the subexpression for the given sample.  
   \param v A vector to store intermediate results.  
   \param offset Index in v to begin storing results.  
   \param sampleNo Sample number to evaluate.  
   \param roffset (output parameter) the offset in the return vector where the result begins.  
   
   The return value will be an existing vector so do not deallocate it.  
   If the result is stored in v it should be stored at the offset given.  
   Everything from offset to the end of v should be considered available for this method to use.  
 */  
 DataTypes::ValueType*  
 DataLazy::resolveUnary(ValueType& v, size_t offset, int sampleNo, size_t& roffset) const  
 {  
     // we assume that any collapsing has been done before we get here  
     // since we only have one argument we don't need to think about only  
     // processing single points.  
   if (m_readytype!='E')  
   {  
     throw DataException("Programmer error - resolveUnary should only be called on expanded Data.");  
   }  
   const ValueType* vleft=m_left->resolveSample(v,offset,sampleNo,roffset);  
   const double* left=&((*vleft)[roffset]);  
   double* result=&(v[offset]);  
   roffset=offset;  
   switch (m_op)  
   {  
     case SIN:    
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::sin);  
     break;  
     case COS:  
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::cos);  
     break;  
     case TAN:  
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::tan);  
     break;  
     case ASIN:  
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::asin);  
     break;  
     case ACOS:  
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::acos);  
     break;  
     case ATAN:  
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::atan);  
     break;  
     case SINH:  
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::sinh);  
     break;  
     case COSH:  
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::cosh);  
     break;  
     case TANH:  
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::tanh);  
     break;  
     case ERF:  
 #if defined (_WIN32) && !defined(__INTEL_COMPILER)  
     throw DataException("Error - Data:: erf function is not supported on _WIN32 platforms.");  
 #else  
     tensor_unary_operation(m_samplesize, left, result, ::erf);  
     break;  
 #endif  
    case ASINH:  
 #if defined (_WIN32) && !defined(__INTEL_COMPILER)  
     tensor_unary_operation(m_samplesize, left, result, escript::asinh_substitute);  
 #else  
     tensor_unary_operation(m_samplesize, left, result, ::asinh);  
 #endif    
     break;  
    case ACOSH:  
 #if defined (_WIN32) && !defined(__INTEL_COMPILER)  
     tensor_unary_operation(m_samplesize, left, result, escript::acosh_substitute);  
 #else  
     tensor_unary_operation(m_samplesize, left, result, ::acosh);  
 #endif    
     break;  
    case ATANH:  
 #if defined (_WIN32) && !defined(__INTEL_COMPILER)  
     tensor_unary_operation(m_samplesize, left, result, escript::atanh_substitute);  
 #else  
     tensor_unary_operation(m_samplesize, left, result, ::atanh);  
 #endif    
     break;  
     case LOG10:  
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::log10);  
     break;  
     case LOG:  
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::log);  
     break;  
     case SIGN:  
     tensor_unary_operation(m_samplesize, left, result, escript::fsign);  
     break;  
     case ABS:  
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::fabs);  
     break;  
     case NEG:  
     tensor_unary_operation(m_samplesize, left, result, negate<double>());  
     break;  
     case POS:  
     // it doesn't mean anything for delayed.  
     // it will just trigger a deep copy of the lazy object  
     throw DataException("Programmer error - POS not supported for lazy data.");  
     break;  
     case EXP:  
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::exp);  
     break;  
     case SQRT:  
     tensor_unary_operation<double (*)(double)>(m_samplesize, left, result, ::sqrt);  
     break;  
     case RECIP:  
     tensor_unary_operation(m_samplesize, left, result, bind1st(divides<double>(),1.));  
     break;  
     case GZ:  
     tensor_unary_operation(m_samplesize, left, result, bind2nd(greater<double>(),0.0));  
     break;  
     case LZ:  
     tensor_unary_operation(m_samplesize, left, result, bind2nd(less<double>(),0.0));  
     break;  
     case GEZ:  
     tensor_unary_operation(m_samplesize, left, result, bind2nd(greater_equal<double>(),0.0));  
     break;  
     case LEZ:  
     tensor_unary_operation(m_samplesize, left, result, bind2nd(less_equal<double>(),0.0));  
     break;  
 // There are actually G_UNARY_P but I don't see a compelling reason to treat them differently  
     case NEZ:  
     tensor_unary_operation(m_samplesize, left, result, bind2nd(AbsGT(),m_tol));  
     break;  
     case EZ:  
     tensor_unary_operation(m_samplesize, left, result, bind2nd(AbsLTE(),m_tol));  
     break;  
   
     default:  
     throw DataException("Programmer error - resolveUnary can not resolve operator "+opToString(m_op)+".");  
   }  
   return &v;  
 }  
   
   
   
   
   
   
 /*  
   \brief Compute the value of the expression (unary operation) for the given sample.  
   \return Vector which stores the value of the subexpression for the given sample.  
   \param v A vector to store intermediate results.  
   \param offset Index in v to begin storing results.  
   \param sampleNo Sample number to evaluate.  
   \param roffset (output parameter) the offset in the return vector where the result begins.  
   
   The return value will be an existing vector so do not deallocate it.  
   If the result is stored in v it should be stored at the offset given.  
   Everything from offset to the end of v should be considered available for this method to use.  
 */  
 DataTypes::ValueType*  
 DataLazy::resolveNP1OUT(ValueType& v, size_t offset, int sampleNo, size_t& roffset) const  
 {  
     // we assume that any collapsing has been done before we get here  
     // since we only have one argument we don't need to think about only  
     // processing single points.  
   if (m_readytype!='E')  
   {  
     throw DataException("Programmer error - resolveNP1OUT should only be called on expanded Data.");  
   }  
     // since we can't write the result over the input, we need a result offset further along  
   size_t subroffset=roffset+m_samplesize;  
 LAZYDEBUG(cerr << "subroffset=" << subroffset << endl;)  
   const ValueType* vleft=m_left->resolveSample(v,offset+m_left->m_samplesize,sampleNo,subroffset);  
   roffset=offset;  
   size_t loop=0;  
   size_t numsteps=(m_readytype=='E')?getNumDPPSample():1;  
   size_t step=getNoValues();  
   switch (m_op)  
   {  
     case SYM:  
     for (loop=0;loop<numsteps;++loop)  
     {  
         DataMaths::symmetric(*vleft,m_left->getShape(),subroffset, v, getShape(), offset);  
         subroffset+=step;  
         offset+=step;  
     }  
     break;  
     case NSYM:  
     for (loop=0;loop<numsteps;++loop)  
     {  
         DataMaths::nonsymmetric(*vleft,m_left->getShape(),subroffset, v, getShape(), offset);  
         subroffset+=step;  
         offset+=step;  
     }  
     break;  
     default:  
     throw DataException("Programmer error - resolveNP1OUT can not resolve operator "+opToString(m_op)+".");  
   }  
   return &v;  
 }  
   
 /*  
   \brief Compute the value of the expression (unary operation) for the given sample.  
   \return Vector which stores the value of the subexpression for the given sample.  
   \param v A vector to store intermediate results.  
   \param offset Index in v to begin storing results.  
   \param sampleNo Sample number to evaluate.  
   \param roffset (output parameter) the offset in the return vector where the result begins.  
   
   The return value will be an existing vector so do not deallocate it.  
   If the result is stored in v it should be stored at the offset given.  
   Everything from offset to the end of v should be considered available for this method to use.  
 */  
 DataTypes::ValueType*  
 DataLazy::resolveNP1OUT_P(ValueType& v, size_t offset, int sampleNo, size_t& roffset) const  
 {  
     // we assume that any collapsing has been done before we get here  
     // since we only have one argument we don't need to think about only  
     // processing single points.  
   if (m_readytype!='E')  
   {  
     throw DataException("Programmer error - resolveNP1OUT_P should only be called on expanded Data.");  
   }  
     // since we can't write the result over the input, we need a result offset further along  
   size_t subroffset;  
   const ValueType* vleft=m_left->resolveSample(v,offset+m_left->m_samplesize,sampleNo,subroffset);  
 LAZYDEBUG(cerr << "srcsamplesize=" << offset+m_left->m_samplesize << " beg=" << subroffset << endl;)  
 LAZYDEBUG(cerr << "Offset for 5800=" << getPointOffset(5800/getNumDPPSample(),5800%getNumDPPSample()) << endl;)  
   roffset=offset;  
   size_t loop=0;  
   size_t numsteps=(m_readytype=='E')?getNumDPPSample():1;  
   size_t outstep=getNoValues();  
   size_t instep=m_left->getNoValues();  
 LAZYDEBUG(cerr << "instep=" << instep << " outstep=" << outstep<< " numsteps=" << numsteps << endl;)  
   switch (m_op)  
   {  
     case TRACE:  
     for (loop=0;loop<numsteps;++loop)  
     {  
 size_t zz=sampleNo*getNumDPPSample()+loop;  
 if (zz==5800)  
 {  
 LAZYDEBUG(cerr << "point=" <<  zz<< endl;)  
 LAZYDEBUG(cerr << "Input to  trace=" << DataTypes::pointToString(*vleft,m_left->getShape(),subroffset,"") << endl;)  
 LAZYDEBUG(cerr << "Offset for point=" << getPointOffset(5800/getNumDPPSample(),5800%getNumDPPSample()) << " vs ";)  
 LAZYDEBUG(cerr << subroffset << endl;)  
 LAZYDEBUG(cerr << "output=" << offset << endl;)  
 }  
             DataMaths::trace(*vleft,m_left->getShape(),subroffset, v ,getShape(),offset,m_axis_offset);  
 if (zz==5800)  
 {  
 LAZYDEBUG(cerr << "Result of trace=" << DataTypes::pointToString(v,getShape(),offset,"") << endl;)  
 }  
         subroffset+=instep;  
         offset+=outstep;  
     }  
     break;  
     case TRANS:  
     for (loop=0;loop<numsteps;++loop)  
     {  
             DataMaths::transpose(*vleft,m_left->getShape(),subroffset, v,getShape(),offset,m_axis_offset);  
         subroffset+=instep;  
         offset+=outstep;  
     }  
     break;  
     default:  
     throw DataException("Programmer error - resolveNP1OUTP can not resolve operator "+opToString(m_op)+".");  
   }  
   return &v;  
 }  
993    
994    
 /*  
   \brief Compute the value of the expression (unary operation with int params) for the given sample.  
   \return Vector which stores the value of the subexpression for the given sample.  
   \param v A vector to store intermediate results.  
   \param offset Index in v to begin storing results.  
   \param sampleNo Sample number to evaluate.  
   \param roffset (output parameter) the offset in the return vector where the result begins.  
   
   The return value will be an existing vector so do not deallocate it.  
   If the result is stored in v it should be stored at the offset given.  
   Everything from offset to the end of v should be considered available for this method to use.  
 */  
 DataTypes::ValueType*  
 DataLazy::resolveNP1OUT_2P(ValueType& v, size_t offset, int sampleNo, size_t& roffset) const  
 {  
     // we assume that any collapsing has been done before we get here  
     // since we only have one argument we don't need to think about only  
     // processing single points.  
   if (m_readytype!='E')  
   {  
     throw DataException("Programmer error - resolveNP1OUT_2P should only be called on expanded Data.");  
   }  
     // since we can't write the result over the input, we need a result offset further along  
   size_t subroffset;  
   const ValueType* vleft=m_left->resolveSample(v,offset+m_left->m_samplesize,sampleNo,subroffset);  
 LAZYDEBUG(cerr << "srcsamplesize=" << offset+m_left->m_samplesize << " beg=" << subroffset << endl;)  
 LAZYDEBUG(cerr << "Offset for 5800=" << getPointOffset(5800/getNumDPPSample(),5800%getNumDPPSample()) << endl;)  
   roffset=offset;  
   size_t loop=0;  
   size_t numsteps=(m_readytype=='E')?getNumDPPSample():1;  
   size_t outstep=getNoValues();  
   size_t instep=m_left->getNoValues();  
 LAZYDEBUG(cerr << "instep=" << instep << " outstep=" << outstep<< " numsteps=" << numsteps << endl;)  
   switch (m_op)  
   {  
     case SWAP:  
     for (loop=0;loop<numsteps;++loop)  
     {  
             DataMaths::swapaxes(*vleft,m_left->getShape(),subroffset, v,getShape(),offset,m_axis_offset, m_transpose);  
         subroffset+=instep;  
         offset+=outstep;  
     }  
     break;  
     default:  
     throw DataException("Programmer error - resolveNP1OUT2P can not resolve operator "+opToString(m_op)+".");  
   }  
   return &v;  
 }  
995    
996    
997    
# Line 1295  LAZYDEBUG(cout << " result=      " << re Line 1011  LAZYDEBUG(cout << " result=      " << re
1011        rroffset+=orightstep;\        rroffset+=orightstep;\
1012      }      }
1013    
 /*  
   \brief Compute the value of the expression (binary operation) for the given sample.  
   \return Vector which stores the value of the subexpression for the given sample.  
   \param v A vector to store intermediate results.  
   \param offset Index in v to begin storing results.  
   \param sampleNo Sample number to evaluate.  
   \param roffset (output parameter) the offset in the return vector where the result begins.  
   
   The return value will be an existing vector so do not deallocate it.  
   If the result is stored in v it should be stored at the offset given.  
   Everything from offset to the end of v should be considered available for this method to use.  
 */  
 // This method assumes that any subexpressions which evaluate to Constant or Tagged Data  
 // have already been collapsed to IDENTITY. So we must have at least one expanded child.  
 // If both children are expanded, then we can process them in a single operation (we treat  
 // the whole sample as one big datapoint.  
 // If one of the children is not expanded, then we need to treat each point in the sample  
 // individually.  
 // There is an additional complication when scalar operations are considered.  
 // For example, 2+Vector.  
 // In this case each double within the point is treated individually  
 DataTypes::ValueType*  
 DataLazy::resolveBinary(ValueType& v,  size_t offset, int sampleNo, size_t& roffset) const  
 {  
 LAZYDEBUG(cout << "Resolve binary: " << toString() << endl;)  
   
   size_t lroffset=0, rroffset=0;    // offsets in the left and right result vectors  
     // first work out which of the children are expanded  
   bool leftExp=(m_left->m_readytype=='E');  
   bool rightExp=(m_right->m_readytype=='E');  
   if (!leftExp && !rightExp)  
   {  
     throw DataException("Programmer Error - please use collapse if neither argument has type 'E'.");  
   }  
   bool leftScalar=(m_left->getRank()==0);  
   bool rightScalar=(m_right->getRank()==0);  
   if ((m_left->getRank()!=m_right->getRank()) && (!leftScalar && !rightScalar))  
   {  
     throw DataException("resolveBinary - ranks of arguments must match unless one of them is scalar.");  
   }  
   size_t leftsize=m_left->getNoValues();  
   size_t rightsize=m_right->getNoValues();  
   size_t chunksize=1;           // how many doubles will be processed in one go  
   int leftstep=0;       // how far should the left offset advance after each step  
   int rightstep=0;  
   int numsteps=0;       // total number of steps for the inner loop  
   int oleftstep=0;  // the o variables refer to the outer loop  
   int orightstep=0; // The outer loop is only required in cases where there is an extended scalar  
   int onumsteps=1;  
     
   bool LES=(leftExp && leftScalar); // Left is an expanded scalar  
   bool RES=(rightExp && rightScalar);  
   bool LS=(!leftExp && leftScalar); // left is a single scalar  
   bool RS=(!rightExp && rightScalar);  
   bool LN=(!leftExp && !leftScalar);    // left is a single non-scalar  
   bool RN=(!rightExp && !rightScalar);  
   bool LEN=(leftExp && !leftScalar);    // left is an expanded non-scalar  
   bool REN=(rightExp && !rightScalar);  
   
   if ((LES && RES) || (LEN && REN)) // both are Expanded scalars or both are expanded non-scalars  
   {  
     chunksize=m_left->getNumDPPSample()*leftsize;  
     leftstep=0;  
     rightstep=0;  
     numsteps=1;  
   }  
   else if (LES || RES)  
   {  
     chunksize=1;  
     if (LES)        // left is an expanded scalar  
     {  
         if (RS)  
         {  
            leftstep=1;  
            rightstep=0;  
            numsteps=m_left->getNumDPPSample();  
         }  
         else        // RN or REN  
         {  
            leftstep=0;  
            oleftstep=1;  
            rightstep=1;  
            orightstep=(RN ? -(int)rightsize : 0);  
            numsteps=rightsize;  
            onumsteps=m_left->getNumDPPSample();  
         }  
     }  
     else        // right is an expanded scalar  
     {  
         if (LS)  
         {  
            rightstep=1;  
            leftstep=0;  
            numsteps=m_right->getNumDPPSample();  
         }  
         else  
         {  
            rightstep=0;  
            orightstep=1;  
            leftstep=1;  
            oleftstep=(LN ? -(int)leftsize : 0);  
            numsteps=leftsize;  
            onumsteps=m_right->getNumDPPSample();  
         }  
     }  
   }  
   else  // this leaves (LEN, RS), (LEN, RN) and their transposes  
   {  
     if (LEN)    // and Right will be a single value  
     {  
         chunksize=rightsize;  
         leftstep=rightsize;  
         rightstep=0;  
         numsteps=m_left->getNumDPPSample();  
         if (RS)  
         {  
            numsteps*=leftsize;  
         }  
     }  
     else    // REN  
     {  
         chunksize=leftsize;  
         rightstep=leftsize;  
         leftstep=0;  
         numsteps=m_right->getNumDPPSample();  
         if (LS)  
         {  
            numsteps*=rightsize;  
         }  
     }  
   }  
   
   int resultStep=max(leftstep,rightstep);   // only one (at most) should be !=0  
     // Get the values of sub-expressions  
   const ValueType* left=m_left->resolveSample(v,offset+getMaxSampleSize(),sampleNo,lroffset);   // see note on  
     // calcBufss for why we can't put offset as the 2nd param above  
   const ValueType* right=m_right->resolveSample(v,offset+2*getMaxSampleSize(),sampleNo,rroffset); // Note  
     // the right child starts further along.  
 LAZYDEBUG(cout << "Post sub calls in " << toString() << endl;)  
 LAZYDEBUG(cout << "shapes=" << DataTypes::shapeToString(m_left->getShape()) << "," << DataTypes::shapeToString(m_right->getShape()) << endl;)  
 LAZYDEBUG(cout << "chunksize=" << chunksize << endl << "leftstep=" << leftstep << " rightstep=" << rightstep;)  
 LAZYDEBUG(cout << " numsteps=" << numsteps << endl << "oleftstep=" << oleftstep << " orightstep=" << orightstep;)  
 LAZYDEBUG(cout << "onumsteps=" << onumsteps << endl;)  
 LAZYDEBUG(cout << " DPPS=" << m_left->getNumDPPSample() << "," <<m_right->getNumDPPSample() << endl;)  
 LAZYDEBUG(cout << "" << LS << RS << LN << RN << LES << RES <<LEN << REN <<   endl;)  
   
   
   double* resultp=&(v[offset]);     // results are stored at the vector offset we recieved  
   switch(m_op)  
   {  
     case ADD:  
         PROC_OP(NO_ARG,plus<double>());  
     break;  
     case SUB:  
     PROC_OP(NO_ARG,minus<double>());  
     break;  
     case MUL:  
     PROC_OP(NO_ARG,multiplies<double>());  
     break;  
     case DIV:  
     PROC_OP(NO_ARG,divides<double>());  
     break;  
     case POW:  
        PROC_OP(double (double,double),::pow);  
     break;  
     default:  
     throw DataException("Programmer error - resolveBinary can not resolve operator "+opToString(m_op)+".");  
   }  
   roffset=offset;  
   return &v;  
 }  
   
   
   
 /*  
   \brief Compute the value of the expression (tensor product) for the given sample.  
   \return Vector which stores the value of the subexpression for the given sample.  
   \param v A vector to store intermediate results.  
   \param offset Index in v to begin storing results.  
   \param sampleNo Sample number to evaluate.  
   \param roffset (output parameter) the offset in the return vector where the result begins.  
   
   The return value will be an existing vector so do not deallocate it.  
   If the result is stored in v it should be stored at the offset given.  
   Everything from offset to the end of v should be considered available for this method to use.  
 */  
 // This method assumes that any subexpressions which evaluate to Constant or Tagged Data  
 // have already been collapsed to IDENTITY. So we must have at least one expanded child.  
 // unlike the other resolve helpers, we must treat these datapoints separately.  
 DataTypes::ValueType*  
 DataLazy::resolveTProd(ValueType& v,  size_t offset, int sampleNo, size_t& roffset) const  
 {  
 LAZYDEBUG(cout << "Resolve TensorProduct: " << toString()  << " to offset " << offset<< endl;)  
   
   size_t lroffset=0, rroffset=0;    // offsets in the left and right result vectors  
     // first work out which of the children are expanded  
   bool leftExp=(m_left->m_readytype=='E');  
   bool rightExp=(m_right->m_readytype=='E');  
   int steps=getNumDPPSample();  
 /*  int leftStep=((leftExp && !rightExp)? m_right->getNoValues() : 0);  
   int rightStep=((rightExp && !leftExp)? m_left->getNoValues() : 0);*/  
   int leftStep=(leftExp? m_left->getNoValues() : 0);        // do not have scalars as input to this method  
   int rightStep=(rightExp?m_right->getNoValues() : 0);  
   
   int resultStep=getNoValues();  
     // Get the values of sub-expressions (leave a gap of one sample for the result).  
   int gap=offset+m_samplesize;    
   
 LAZYDEBUG(cout << "Query left with offset=" << gap << endl;)  
   
   const ValueType* left=m_left->resolveSample(v,gap,sampleNo,lroffset);  
   gap+=m_left->getMaxSampleSize();  
   
   
 LAZYDEBUG(cout << "Query right with offset=" << gap << endl;)  
   
   
   const ValueType* right=m_right->resolveSample(v,gap,sampleNo,rroffset);  
   
 LAZYDEBUG(cerr << "[Left shape]=" << DataTypes::shapeToString(m_left->getShape()) << "\n[Right shape]=" << DataTypes::shapeToString(m_right->getShape()) << " result=" <<DataTypes::shapeToString(getShape()) <<  endl;  
 cout << getNoValues() << endl;)  
 LAZYDEBUG(cerr << "Result of left=";)  
 LAZYDEBUG(cerr << "[" << lroffset << " .. " << lroffset+m_left->getNoValues() << "]" << endl;  
   
 for (int i=lroffset, limit=lroffset+(leftExp?m_left->getNoValues()*m_left->getNumDPPSample():m_left->getNoValues());i<limit;++i)  
 {  
 cout << "[" << setw(2) << i-lroffset << "] " << setw(10) << (*left)[i] << " ";  
 if (i%4==0) cout << endl;  
 })  
 LAZYDEBUG(cerr << "\nResult of right=" << endl;)  
 LAZYDEBUG(  
 for (int i=rroffset, limit=rroffset+(rightExp?m_right->getNoValues()*m_right->getNumDPPSample():m_right->getNoValues());i<limit;++i)  
 {  
 cerr << "[" <<  setw(2)<< i-rroffset << "] " << setw(10) << (*right)[i] << " ";  
 if (i%4==0) cout << endl;  
 }  
 cerr << endl;  
 )  
 LAZYDEBUG(cerr << "Post sub calls: " << toString() << endl;)  
 LAZYDEBUG(cout << "LeftExp=" << leftExp << " rightExp=" << rightExp << endl;)  
 LAZYDEBUG(cout << "LeftR=" << m_left->getRank() << " rightExp=" << m_right->getRank() << endl;)  
 LAZYDEBUG(cout << "LeftSize=" << m_left->getNoValues() << " RightSize=" << m_right->getNoValues() << endl;)  
 LAZYDEBUG(cout << "m_samplesize=" << m_samplesize << endl;)  
 LAZYDEBUG(cout << "outputshape=" << DataTypes::shapeToString(getShape()) << endl;)  
 LAZYDEBUG(cout << "DPPS=" << m_right->getNumDPPSample() <<"."<<endl;)  
   
   double* resultp=&(v[offset]);     // results are stored at the vector offset we recieved  
   switch(m_op)  
   {  
     case PROD:  
     for (int i=0;i<steps;++i,resultp+=resultStep)  
     {  
   
 LAZYDEBUG(cout << "lroffset=" << lroffset << "rroffset=" << rroffset << endl;)  
 LAZYDEBUG(cout << "l*=" << left << " r*=" << right << endl;)  
 LAZYDEBUG(cout << "m_SL=" << m_SL << " m_SM=" << m_SM << " m_SR=" << m_SR << endl;)  
   
           const double *ptr_0 = &((*left)[lroffset]);  
           const double *ptr_1 = &((*right)[rroffset]);  
   
 LAZYDEBUG(cout << DataTypes::pointToString(*left, m_left->getShape(),lroffset,"LEFT") << endl;)  
 LAZYDEBUG(cout << DataTypes::pointToString(*right,m_right->getShape(),rroffset, "RIGHT") << endl;)  
   
           matrix_matrix_product(m_SL, m_SM, m_SR, ptr_0, ptr_1, resultp, m_transpose);  
   
 LAZYDEBUG(cout << "Results--\n";  
 {  
   DataVector dv(getNoValues());  
 for (int z=0;z<getNoValues();++z)  
 {  
   cout << "[" << setw(2) << z<< "] " << setw(10) << resultp[z] << " ";  
   if (z%4==0) cout << endl;  
   dv[z]=resultp[z];  
 }  
 cout << endl << DataTypes::pointToString(dv,getShape(),0,"RESLT");  
 cout << "\nWritten to: " << resultp << " resultStep=" << resultStep << endl;  
 }  
 )  
       lroffset+=leftStep;  
       rroffset+=rightStep;  
     }  
     break;  
     default:  
     throw DataException("Programmer error - resolveTProduct can not resolve operator "+opToString(m_op)+".");  
   }  
   roffset=offset;  
   return &v;  
 }  
   
   
 #ifdef LAZY_NODE_STORAGE  
1014    
1015  // The result will be stored in m_samples  // The result will be stored in m_samples
1016  // The return value is a pointer to the DataVector, offset is the offset within the return value  // The return value is a pointer to the DataVector, offset is the offset within the return value
# Line 1601  LAZYDEBUG(cout << "Resolve sample " << t Line 1026  LAZYDEBUG(cout << "Resolve sample " << t
1026    if (m_op==IDENTITY)      if (m_op==IDENTITY)  
1027    {    {
1028      const ValueType& vec=m_id->getVectorRO();      const ValueType& vec=m_id->getVectorRO();
 //     if (m_readytype=='C')  
 //     {  
 //  roffset=0;      // all samples read from the same position  
 //  return &(m_samples);  
 //     }  
1029      roffset=m_id->getPointOffset(sampleNo, 0);      roffset=m_id->getPointOffset(sampleNo, 0);
1030    #ifdef LAZY_STACK_PROF
1031    int x;
1032    if (&x<stackend[omp_get_thread_num()])
1033    {
1034           stackend[omp_get_thread_num()]=&x;
1035    }
1036    #endif
1037      return &(vec);      return &(vec);
1038    }    }
1039    if (m_readytype!='E')    if (m_readytype!='E')
# Line 1619  LAZYDEBUG(cout << "Resolve sample " << t Line 1046  LAZYDEBUG(cout << "Resolve sample " << t
1046      return &(m_samples);        // sample is already resolved      return &(m_samples);        // sample is already resolved
1047    }    }
1048    m_sampleids[tid]=sampleNo;    m_sampleids[tid]=sampleNo;
1049    
1050    switch (getOpgroup(m_op))    switch (getOpgroup(m_op))
1051    {    {
1052    case G_UNARY:    case G_UNARY:
# Line 1628  LAZYDEBUG(cout << "Resolve sample " << t Line 1056  LAZYDEBUG(cout << "Resolve sample " << t
1056    case G_NP1OUT_P: return resolveNodeNP1OUT_P(tid, sampleNo, roffset);    case G_NP1OUT_P: return resolveNodeNP1OUT_P(tid, sampleNo, roffset);
1057    case G_TENSORPROD: return resolveNodeTProd(tid, sampleNo, roffset);    case G_TENSORPROD: return resolveNodeTProd(tid, sampleNo, roffset);
1058    case G_NP1OUT_2P: return resolveNodeNP1OUT_2P(tid, sampleNo, roffset);    case G_NP1OUT_2P: return resolveNodeNP1OUT_2P(tid, sampleNo, roffset);
1059      case G_REDUCTION: return resolveNodeReduction(tid, sampleNo, roffset);
1060      case G_CONDEVAL: return resolveNodeCondEval(tid, sampleNo, roffset);
1061    default:    default:
1062      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)+".");
1063    }    }
# Line 1766  DataLazy::resolveNodeUnary(int tid, int Line 1196  DataLazy::resolveNodeUnary(int tid, int
1196    
1197    
1198  const DataTypes::ValueType*  const DataTypes::ValueType*
1199    DataLazy::resolveNodeReduction(int tid, int sampleNo, size_t& roffset)
1200    {
1201        // we assume that any collapsing has been done before we get here
1202        // since we only have one argument we don't need to think about only
1203        // processing single points.
1204        // we will also know we won't get identity nodes
1205      if (m_readytype!='E')
1206      {
1207        throw DataException("Programmer error - resolveUnary should only be called on expanded Data.");
1208      }
1209      if (m_op==IDENTITY)
1210      {
1211        throw DataException("Programmer error - resolveNodeUnary should not be called on identity nodes.");
1212      }
1213      size_t loffset=0;
1214      const DataTypes::ValueType* leftres=m_left->resolveNodeSample(tid, sampleNo, loffset);
1215    
1216      roffset=m_samplesize*tid;
1217      unsigned int ndpps=getNumDPPSample();
1218      unsigned int psize=DataTypes::noValues(m_left->getShape());
1219      double* result=&(m_samples[roffset]);
1220      switch (m_op)
1221      {
1222        case MINVAL:
1223        {
1224          for (unsigned int z=0;z<ndpps;++z)
1225          {
1226            FMin op;
1227            *result=DataMaths::reductionOp(*leftres, m_left->getShape(), loffset, op, numeric_limits<double>::max());
1228            loffset+=psize;
1229            result++;
1230          }
1231        }
1232        break;
1233        case MAXVAL:
1234        {
1235          for (unsigned int z=0;z<ndpps;++z)
1236          {
1237          FMax op;
1238          *result=DataMaths::reductionOp(*leftres, m_left->getShape(), loffset, op, numeric_limits<double>::max()*-1);
1239          loffset+=psize;
1240          result++;
1241          }
1242        }
1243        break;
1244        default:
1245        throw DataException("Programmer error - resolveUnary can not resolve operator "+opToString(m_op)+".");
1246      }
1247      return &(m_samples);
1248    }
1249    
1250    const DataTypes::ValueType*
1251  DataLazy::resolveNodeNP1OUT(int tid, int sampleNo, size_t& roffset)  DataLazy::resolveNodeNP1OUT(int tid, int sampleNo, size_t& roffset)
1252  {  {
1253      // we assume that any collapsing has been done before we get here      // we assume that any collapsing has been done before we get here
# Line 1894  DataLazy::resolveNodeNP1OUT_2P(int tid, Line 1376  DataLazy::resolveNodeNP1OUT_2P(int tid,
1376    return &m_samples;    return &m_samples;
1377  }  }
1378    
1379    const DataTypes::ValueType*
1380    DataLazy::resolveNodeCondEval(int tid, int sampleNo, size_t& roffset)
1381    {
1382      if (m_readytype!='E')
1383      {
1384        throw DataException("Programmer error - resolveNodeCondEval should only be called on expanded Data.");
1385      }
1386      if (m_op!=CONDEVAL)
1387      {
1388        throw DataException("Programmer error - resolveNodeCondEval should only be called on CONDEVAL nodes.");
1389      }
1390      size_t subroffset;
1391    
1392      const ValueType* maskres=m_mask->resolveNodeSample(tid, sampleNo, subroffset);
1393      const ValueType* srcres=0;
1394      if ((*maskres)[subroffset]>0)
1395      {
1396        srcres=m_left->resolveNodeSample(tid, sampleNo, subroffset);
1397      }
1398      else
1399      {
1400        srcres=m_right->resolveNodeSample(tid, sampleNo, subroffset);
1401      }
1402    
1403      // Now we need to copy the result
1404    
1405      roffset=m_samplesize*tid;
1406      for (int i=0;i<m_samplesize;++i)
1407      {
1408        m_samples[roffset+i]=(*srcres)[subroffset+i];  
1409      }
1410    
1411      return &m_samples;
1412    }
1413    
1414  // This method assumes that any subexpressions which evaluate to Constant or Tagged Data  // This method assumes that any subexpressions which evaluate to Constant or Tagged Data
1415  // have already been collapsed to IDENTITY. So we must have at least one expanded child.  // have already been collapsed to IDENTITY. So we must have at least one expanded child.
# Line 2119  LAZYDEBUG(cout << DataTypes::pointToStri Line 1634  LAZYDEBUG(cout << DataTypes::pointToStri
1634    roffset=offset;    roffset=offset;
1635    return &m_samples;    return &m_samples;
1636  }  }
 #endif //LAZY_NODE_STORAGE  
   
 /*  
   \brief Compute the value of the expression for the given sample.  
   \return Vector which stores the value of the subexpression for the given sample.  
   \param v A vector to store intermediate results.  
   \param offset Index in v to begin storing results.  
   \param sampleNo Sample number to evaluate.  
   \param roffset (output parameter) the offset in the return vector where the result begins.  
1637    
   The return value will be an existing vector so do not deallocate it.  
 */  
 // the vector and the offset are a place where the method could write its data if it wishes  
 // it is not obligated to do so. For example, if it has its own storage already, it can use that.  
 // Hence the return value to indicate where the data is actually stored.  
 // Regardless, the storage should be assumed to be used, even if it isn't.  
1638    
 // the roffset is the offset within the returned vector where the data begins  
1639  const DataTypes::ValueType*  const DataTypes::ValueType*
1640  DataLazy::resolveSample(ValueType& v, size_t offset, int sampleNo, size_t& roffset)  DataLazy::resolveSample(int sampleNo, size_t& roffset)
 {  
 LAZYDEBUG(cout << "Resolve sample " << toString() << endl;)  
     // collapse so we have a 'E' node or an IDENTITY for some other type  
   if (m_readytype!='E' && m_op!=IDENTITY)  
   {  
     collapse();  
   }  
   if (m_op==IDENTITY)    
   {  
     const ValueType& vec=m_id->getVectorRO();  
     if (m_readytype=='C')  
     {  
     roffset=0;  
 LAZYDEBUG(cout << "Finish  sample " << toString() << endl;)  
     return &(vec);  
     }  
     roffset=m_id->getPointOffset(sampleNo, 0);  
 LAZYDEBUG(cout << "Finish  sample " << toString() << endl;)  
     return &(vec);  
   }  
   if (m_readytype!='E')  
   {  
     throw DataException("Programmer Error - Collapse did not produce an expanded node.");  
   }  
   switch (getOpgroup(m_op))  
   {  
   case G_UNARY:  
   case G_UNARY_P: return resolveUnary(v, offset,sampleNo,roffset);  
   case G_BINARY: return resolveBinary(v, offset,sampleNo,roffset);  
   case G_NP1OUT: return resolveNP1OUT(v, offset, sampleNo,roffset);  
   case G_NP1OUT_P: return resolveNP1OUT_P(v, offset, sampleNo,roffset);  
   case G_TENSORPROD: return resolveTProd(v,offset, sampleNo,roffset);  
   case G_NP1OUT_2P: return resolveNP1OUT_2P(v, offset, sampleNo, roffset);  
   default:  
     throw DataException("Programmer Error - resolveSample does not know how to process "+opToString(m_op)+".");  
   }  
   
 }  
   
 const DataTypes::ValueType*  
 DataLazy::resolveSample(BufferGroup& bg, int sampleNo, size_t& roffset)  
1641  {  {
1642  #ifdef _OPENMP  #ifdef _OPENMP
1643      int tid=omp_get_thread_num();      int tid=omp_get_thread_num();
1644  #else  #else
1645      int tid=0;      int tid=0;
1646  #endif  #endif
1647      return resolveSample(bg.getBuffer(tid),bg.getOffset(tid),sampleNo,roffset);  
1648    #ifdef LAZY_STACK_PROF
1649        stackstart[tid]=&tid;
1650        stackend[tid]=&tid;
1651        const DataTypes::ValueType* r=resolveNodeSample(tid, sampleNo, roffset);
1652        size_t d=(size_t)stackstart[tid]-(size_t)stackend[tid];
1653        #pragma omp critical
1654        if (d>maxstackuse)
1655        {
1656    cout << "Max resolve Stack use " << d << endl;
1657            maxstackuse=d;
1658        }
1659        return r;
1660    #else
1661        return resolveNodeSample(tid, sampleNo, roffset);
1662    #endif
1663  }  }
1664    
1665    
# Line 2196  DataLazy::resolveToIdentity() Line 1669  DataLazy::resolveToIdentity()
1669  {  {
1670     if (m_op==IDENTITY)     if (m_op==IDENTITY)
1671      return;      return;
 #ifndef LAZY_NODE_STORAGE  
    DataReady_ptr p=resolveVectorWorker();  
 #else  
1672     DataReady_ptr p=resolveNodeWorker();     DataReady_ptr p=resolveNodeWorker();
 #endif  
1673     makeIdentity(p);     makeIdentity(p);
1674  }  }
1675    
# Line 2216  void DataLazy::makeIdentity(const DataRe Line 1685  void DataLazy::makeIdentity(const DataRe
1685     else if(p->isExpanded()) {m_readytype='E';}     else if(p->isExpanded()) {m_readytype='E';}
1686     else if (p->isTagged()) {m_readytype='T';}     else if (p->isTagged()) {m_readytype='T';}
1687     else {throw DataException("Unknown DataReady instance in convertToIdentity constructor.");}     else {throw DataException("Unknown DataReady instance in convertToIdentity constructor.");}
    m_buffsRequired=1;  
1688     m_samplesize=p->getNumDPPSample()*p->getNoValues();     m_samplesize=p->getNumDPPSample()*p->getNoValues();
    m_maxsamplesize=m_samplesize;  
1689     m_left.reset();     m_left.reset();
1690     m_right.reset();     m_right.reset();
1691  }  }
# Line 2231  DataLazy::resolve() Line 1698  DataLazy::resolve()
1698      return m_id;      return m_id;
1699  }  }
1700    
 #ifdef LAZY_NODE_STORAGE  
1701    
1702  // This version of resolve uses storage in each node to hold results  /* This is really a static method but I think that caused problems in windows */
1703  DataReady_ptr  void
1704  DataLazy::resolveNodeWorker()  DataLazy::resolveGroupWorker(std::vector<DataLazy*>& dats)
1705  {  {
1706    if (m_readytype!='E')     // if the whole sub-expression is Constant or Tagged, then evaluate it normally    if (dats.empty())
1707    {    {
1708      collapse();      return;
1709    }    }
1710    if (m_op==IDENTITY)       // So a lazy expression of Constant or Tagged data will be returned here.    vector<DataLazy*> work;
1711      FunctionSpace fs=dats[0]->getFunctionSpace();
1712      bool match=true;
1713      for (int i=dats.size()-1;i>=0;--i)
1714    {    {
1715      return m_id;      if (dats[i]->m_readytype!='E')
1716        {
1717            dats[i]->collapse();
1718        }
1719        if (dats[i]->m_op!=IDENTITY)
1720        {
1721            work.push_back(dats[i]);
1722            if (fs!=dats[i]->getFunctionSpace())
1723            {
1724                match=false;
1725            }
1726        }
1727    }    }
1728      // from this point on we must have m_op!=IDENTITY and m_readytype=='E'    if (work.empty())
   DataExpanded* result=new DataExpanded(getFunctionSpace(),getShape(),  ValueType(getNoValues()));  
   ValueType& resvec=result->getVectorRW();  
   DataReady_ptr resptr=DataReady_ptr(result);  
   
   int sample;  
   int totalsamples=getNumSamples();  
   const ValueType* res=0;   // Storage for answer  
 LAZYDEBUG(cout << "Total number of samples=" <<totalsamples << endl;)  
   #pragma omp parallel for private(sample,res) schedule(static)  
   for (sample=0;sample<totalsamples;++sample)  
1729    {    {
1730      size_t roffset=0;      return;     // no work to do
1731      }
1732      if (match)    // all functionspaces match.  Yes I realise this is overly strict
1733      {     // it is possible that dats[0] is one of the objects which we discarded and
1734            // all the other functionspaces match.
1735        vector<DataExpanded*> dep;
1736        vector<ValueType*> vecs;
1737        for (int i=0;i<work.size();++i)
1738        {
1739            dep.push_back(new DataExpanded(fs,work[i]->getShape(), ValueType(work[i]->getNoValues())));
1740            vecs.push_back(&(dep[i]->getVectorRW()));
1741        }
1742        int totalsamples=work[0]->getNumSamples();
1743        const ValueType* res=0; // Storage for answer
1744        int sample;
1745        #pragma omp parallel private(sample, res)
1746        {
1747            size_t roffset=0;
1748            #pragma omp for schedule(static)
1749            for (sample=0;sample<totalsamples;++sample)
1750            {
1751            roffset=0;
1752            int j;
1753            for (j=work.size()-1;j>=0;--j)
1754            {
1755  #ifdef _OPENMP  #ifdef _OPENMP
1756      res=resolveNodeSample(omp_get_thread_num(),sample,roffset);                  res=work[j]->resolveNodeSample(omp_get_thread_num(),sample,roffset);
1757  #else  #else
1758      res=resolveNodeSample(0,sample,roffset);                  res=work[j]->resolveNodeSample(0,sample,roffset);
1759  #endif  #endif
1760  LAZYDEBUG(cout << "Sample #" << sample << endl;)                  DataVector::size_type outoffset=dep[j]->getPointOffset(sample,0);
1761  LAZYDEBUG(cout << "Final res[" << roffset<< "]=" << (*res)[roffset] << (*res)[roffset]<< endl; )                  memcpy(&((*vecs[j])[outoffset]),&((*res)[roffset]),work[j]->m_samplesize*sizeof(DataVector::ElementType));
1762      DataVector::size_type outoffset=result->getPointOffset(sample,0);          }
1763      memcpy(&(resvec[outoffset]),&((*res)[roffset]),m_samplesize*sizeof(DataVector::ElementType));          }
1764        }
1765        // Now we need to load the new results as identity ops into the lazy nodes
1766        for (int i=work.size()-1;i>=0;--i)
1767        {
1768            work[i]->makeIdentity(boost::dynamic_pointer_cast<DataReady>(dep[i]->getPtr()));
1769        }
1770      }
1771      else  // functionspaces do not match
1772      {
1773        for (int i=0;i<work.size();++i)
1774        {
1775            work[i]->resolveToIdentity();
1776        }
1777    }    }
   return resptr;  
1778  }  }
1779    
 #endif // LAZY_NODE_STORAGE  
1780    
1781  // To simplify the memory management, all threads operate on one large vector, rather than one each.  
1782  // Each sample is evaluated independently and copied into the result DataExpanded.  // This version of resolve uses storage in each node to hold results
1783  DataReady_ptr  DataReady_ptr
1784  DataLazy::resolveVectorWorker()  DataLazy::resolveNodeWorker()
1785  {  {
   
 LAZYDEBUG(cout << "Sample size=" << m_samplesize << endl;)  
 LAZYDEBUG(cout << "Buffers=" << m_buffsRequired << endl;)  
1786    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
1787    {    {
1788      collapse();      collapse();
# Line 2290  LAZYDEBUG(cout << "Buffers=" << m_buffsR Line 1792  LAZYDEBUG(cout << "Buffers=" << m_buffsR
1792      return m_id;      return m_id;
1793    }    }
1794      // 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'
   size_t threadbuffersize=m_maxsamplesize*(max(1,m_buffsRequired)); // Each thread needs to have enough  
     // storage to evaluate its expression  
   int numthreads=1;  
 #ifdef _OPENMP  
   numthreads=omp_get_max_threads();  
 #endif  
   ValueType v(numthreads*threadbuffersize);  
 LAZYDEBUG(cout << "Buffer created with size=" << v.size() << endl;)  
1795    DataExpanded* result=new DataExpanded(getFunctionSpace(),getShape(),  ValueType(getNoValues()));    DataExpanded* result=new DataExpanded(getFunctionSpace(),getShape(),  ValueType(getNoValues()));
1796    ValueType& resvec=result->getVectorRW();    ValueType& resvec=result->getVectorRW();
1797    DataReady_ptr resptr=DataReady_ptr(result);    DataReady_ptr resptr=DataReady_ptr(result);
1798    
1799    int sample;    int sample;
   size_t outoffset;     // offset in the output data  
1800    int totalsamples=getNumSamples();    int totalsamples=getNumSamples();
1801    const ValueType* res=0;   // Vector storing the answer    const ValueType* res=0;   // Storage for answer
   size_t resoffset=0;       // where in the vector to find the answer  
1802  LAZYDEBUG(cout << "Total number of samples=" <<totalsamples << endl;)  LAZYDEBUG(cout << "Total number of samples=" <<totalsamples << endl;)
1803    #pragma omp parallel for private(sample,resoffset,outoffset,res) schedule(static)    #pragma omp parallel private(sample,res)
   for (sample=0;sample<totalsamples;++sample)  
1804    {    {
1805  LAZYDEBUG(cout << "################################# " << sample << endl;)      size_t roffset=0;
1806    #ifdef LAZY_STACK_PROF
1807        stackstart[omp_get_thread_num()]=&roffset;
1808        stackend[omp_get_thread_num()]=&roffset;
1809    #endif
1810        #pragma omp for schedule(static)
1811        for (sample=0;sample<totalsamples;++sample)
1812        {
1813            roffset=0;
1814  #ifdef _OPENMP  #ifdef _OPENMP
1815      res=resolveSample(v,threadbuffersize*omp_get_thread_num(),sample,resoffset);              res=resolveNodeSample(omp_get_thread_num(),sample,roffset);
1816  #else  #else
1817      res=resolveSample(v,0,sample,resoffset);   // res would normally be v, but not if its a single IDENTITY op.              res=resolveNodeSample(0,sample,roffset);
1818  #endif  #endif
1819  LAZYDEBUG(cerr << "-------------------------------- " << endl;)  LAZYDEBUG(cout << "Sample #" << sample << endl;)
1820  LAZYDEBUG(cerr<< "Copying sample#" << sample << endl;)  LAZYDEBUG(cout << "Final res[" << roffset<< "]=" << (*res)[roffset] << (*res)[roffset]<< endl; )
1821      outoffset=result->getPointOffset(sample,0);              DataVector::size_type outoffset=result->getPointOffset(sample,0);
1822  LAZYDEBUG(cerr << "offset=" << outoffset << " from offset=" << resoffset << " " << m_samplesize << " doubles" << endl;)              memcpy(&(resvec[outoffset]),&((*res)[roffset]),m_samplesize*sizeof(DataVector::ElementType));
1823      for (unsigned int i=0;i<m_samplesize;++i,++outoffset,++resoffset)   // copy values into the output vector      }
1824      {    }
1825  LAZYDEBUG(cerr << "outoffset=" << outoffset << " resoffset=" << resoffset << " " << (*res)[resoffset]<< endl;)  #ifdef LAZY_STACK_PROF
1826      resvec[outoffset]=(*res)[resoffset];    for (int i=0;i<getNumberOfThreads();++i)
1827      }    {
1828  LAZYDEBUG(cerr << DataTypes::pointToString(resvec,getShape(),outoffset-m_samplesize+DataTypes::noValues(getShape()),"Final result:") << endl;)      size_t r=((size_t)stackstart[i] - (size_t)stackend[i]);
1829  LAZYDEBUG(cerr << "*********************************" << endl;)  //  cout << i << " " << stackstart[i] << " .. " << stackend[i] << " = " <<  r << endl;
1830        if (r>maxstackuse)
1831        {
1832            maxstackuse=r;
1833        }
1834    }    }
1835      cout << "Max resolve Stack use=" << maxstackuse << endl;
1836    #endif
1837    return resptr;    return resptr;
1838  }  }
1839    
# Line 2335  std::string Line 1841  std::string
1841  DataLazy::toString() const  DataLazy::toString() const
1842  {  {
1843    ostringstream oss;    ostringstream oss;
1844    oss << "Lazy Data:";    oss << "Lazy Data: [depth=" << m_height<< "] ";
1845    intoString(oss);    switch (escriptParams.getLAZY_STR_FMT())
1846      {
1847      case 1:   // tree format
1848        oss << endl;
1849        intoTreeString(oss,"");
1850        break;
1851      case 2:   // just the depth
1852        break;
1853      default:
1854        intoString(oss);
1855        break;
1856      }
1857    return oss.str();    return oss.str();
1858  }  }
1859    
# Line 2377  DataLazy::intoString(ostringstream& oss) Line 1894  DataLazy::intoString(ostringstream& oss)
1894    case G_UNARY_P:    case G_UNARY_P:
1895    case G_NP1OUT:    case G_NP1OUT:
1896    case G_NP1OUT_P:    case G_NP1OUT_P:
1897      case G_REDUCTION:
1898      oss << opToString(m_op) << '(';      oss << opToString(m_op) << '(';
1899      m_left->intoString(oss);      m_left->intoString(oss);
1900      oss << ')';      oss << ')';
# Line 2394  DataLazy::intoString(ostringstream& oss) Line 1912  DataLazy::intoString(ostringstream& oss)
1912      oss << ", " << m_axis_offset << ", " << m_transpose;      oss << ", " << m_axis_offset << ", " << m_transpose;
1913      oss << ')';      oss << ')';
1914      break;      break;
1915      case G_CONDEVAL:
1916        oss << opToString(m_op)<< '(' ;
1917        m_mask->intoString(oss);
1918        oss << " ? ";
1919        m_left->intoString(oss);
1920        oss << " : ";
1921        m_right->intoString(oss);
1922        oss << ')';
1923        break;
1924    default:    default:
1925      oss << "UNKNOWN";      oss << "UNKNOWN";
1926    }    }
1927  }  }
1928    
1929    
1930    void
1931    DataLazy::intoTreeString(ostringstream& oss, string indent) const
1932    {
1933      oss << '[' << m_rank << ':' << setw(3) << m_samplesize << "] " << indent;
1934      switch (getOpgroup(m_op))
1935      {
1936      case G_IDENTITY:
1937        if (m_id->isExpanded())
1938        {
1939           oss << "E";
1940        }
1941        else if (m_id->isTagged())
1942        {
1943          oss << "T";
1944        }
1945        else if (m_id->isConstant())
1946        {
1947          oss << "C";
1948        }
1949        else
1950        {
1951          oss << "?";
1952        }
1953        oss << '@' << m_id.get() << endl;
1954        break;
1955      case G_BINARY:
1956        oss << opToString(m_op) << endl;
1957        indent+='.';
1958        m_left->intoTreeString(oss, indent);
1959        m_right->intoTreeString(oss, indent);
1960        break;
1961      case G_UNARY:
1962      case G_UNARY_P:
1963      case G_NP1OUT:
1964      case G_NP1OUT_P:
1965      case G_REDUCTION:
1966        oss << opToString(m_op) << endl;
1967        indent+='.';
1968        m_left->intoTreeString(oss, indent);
1969        break;
1970      case G_TENSORPROD:
1971        oss << opToString(m_op) << endl;
1972        indent+='.';
1973        m_left->intoTreeString(oss, indent);
1974        m_right->intoTreeString(oss, indent);
1975        break;
1976      case G_NP1OUT_2P:
1977        oss << opToString(m_op) << ", " << m_axis_offset << ", " << m_transpose<< endl;
1978        indent+='.';
1979        m_left->intoTreeString(oss, indent);
1980        break;
1981      default:
1982        oss << "UNKNOWN";
1983      }
1984    }
1985    
1986    
1987  DataAbstract*  DataAbstract*
1988  DataLazy::deepCopy()  DataLazy::deepCopy()
1989  {  {
1990    switch (getOpgroup(m_op))    switch (getOpgroup(m_op))
1991    {    {
1992    case G_IDENTITY:  return new DataLazy(m_id->deepCopy()->getPtr());    case G_IDENTITY:  return new DataLazy(m_id->deepCopy()->getPtr());
1993    case G_UNARY: return new DataLazy(m_left->deepCopy()->getPtr(),m_op);    case G_UNARY:
1994      case G_REDUCTION:      return new DataLazy(m_left->deepCopy()->getPtr(),m_op);
1995      case G_UNARY_P:   return new DataLazy(m_left->deepCopy()->getPtr(), m_op, m_tol);
1996    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);
1997    case G_NP1OUT: return new DataLazy(m_left->deepCopy()->getPtr(), m_right->deepCopy()->getPtr(),m_op);    case G_NP1OUT: return new DataLazy(m_left->deepCopy()->getPtr(), m_right->deepCopy()->getPtr(),m_op);
1998    case G_TENSORPROD: return new DataLazy(m_left->deepCopy()->getPtr(), m_right->deepCopy()->getPtr(), m_op, m_axis_offset, m_transpose);    case G_TENSORPROD: return new DataLazy(m_left->deepCopy()->getPtr(), m_right->deepCopy()->getPtr(), m_op, m_axis_offset, m_transpose);
1999      case G_NP1OUT_P:   return new DataLazy(m_left->deepCopy()->getPtr(),m_op,  m_axis_offset);
2000      case G_NP1OUT_2P:  return new DataLazy(m_left->deepCopy()->getPtr(), m_op, m_axis_offset, m_transpose);
2001    default:    default:
2002      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)+".");
2003    }    }
2004  }  }
2005    
2006    
2007    
2008  // There is no single, natural interpretation of getLength on DataLazy.  // There is no single, natural interpretation of getLength on DataLazy.
2009  // Instances of DataReady can look at the size of their vectors.  // Instances of DataReady can look at the size of their vectors.
2010  // For lazy though, it could be the size the data would be if it were resolved;  // For lazy though, it could be the size the data would be if it were resolved;

Legend:
Removed from v.2514  
changed lines
  Added in v.4154

  ViewVC Help
Powered by ViewVC 1.1.26