/[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 2737 by jfenwick, Tue Nov 3 00:44:00 2009 UTC revision 2791 by jfenwick, Tue Dec 1 03:36:36 2009 UTC
# Line 109  namespace escript Line 109  namespace escript
109  namespace  namespace
110  {  {
111    
112    
113    // enabling this will print out when ever the maximum stacksize used by resolve increases
114    // it assumes _OPENMP is also in use
115    //#define LAZY_STACK_PROF
116    
117    
118    
119    #ifndef _OPENMP
120      #ifdef LAZY_STACK_PROF
121      #undef LAZY_STACK_PROF
122      #endif
123    #endif
124    
125    
126    #ifdef LAZY_STACK_PROF
127    std::vector<void*> stackstart(getNumberOfThreads());
128    std::vector<void*> stackend(getNumberOfThreads());
129    size_t maxstackuse=0;
130    #endif
131    
132  enum ES_opgroup  enum ES_opgroup
133  {  {
134     G_UNKNOWN,     G_UNKNOWN,
# Line 221  resultShape(DataAbstract_ptr left, ES_op Line 241  resultShape(DataAbstract_ptr left, ES_op
241          int rank=left->getRank();          int rank=left->getRank();
242          if (axis_offset<0 || axis_offset>rank)          if (axis_offset<0 || axis_offset>rank)
243          {          {
244                 throw DataException("Error - Data::transpose must have 0 <= axis_offset <= rank=" + rank);              stringstream e;
245              }              e << "Error - Data::transpose must have 0 <= axis_offset <= rank=" << rank;
246              for (int i=0; i<rank; i++)              throw DataException(e.str());
247            }
248            for (int i=0; i<rank; i++)
249          {          {
250             int index = (axis_offset+i)%rank;             int index = (axis_offset+i)%rank;
251                 sh.push_back(s[index]); // Append to new shape             sh.push_back(s[index]); // Append to new shape
252              }          }
253          return sh;          return sh;
254         }         }
255      break;      break;
# Line 306  SwapShape(DataAbstract_ptr left, const i Line 328  SwapShape(DataAbstract_ptr left, const i
328          throw DataException("Error - Data::swapaxes argument must have at least rank 2.");          throw DataException("Error - Data::swapaxes argument must have at least rank 2.");
329       }       }
330       if (axis0<0 || axis0>rank-1) {       if (axis0<0 || axis0>rank-1) {
331          throw DataException("Error - Data::swapaxes: axis0 must be between 0 and rank-1=" + rank-1);          stringstream e;
332            e << "Error - Data::swapaxes: axis0 must be between 0 and rank-1=" << (rank-1);
333            throw DataException(e.str());
334       }       }
335       if (axis1<0 || axis1>rank-1) {       if (axis1<0 || axis1>rank-1) {
336           throw DataException("Error - Data::swapaxes: axis1 must be between 0 and rank-1=" + rank-1);          stringstream e;
337            e << "Error - Data::swapaxes: axis1 must be between 0 and rank-1=" << (rank-1);
338            throw DataException(e.str());
339       }       }
340       if (axis0 == axis1) {       if (axis0 == axis1) {
341           throw DataException("Error - Data::swapaxes: axis indices must be different.");           throw DataException("Error - Data::swapaxes: axis indices must be different.");
# Line 398  GTPShape(DataAbstract_ptr left, DataAbst Line 424  GTPShape(DataAbstract_ptr left, DataAbst
424    return shape2;    return shape2;
425  }  }
426    
 // 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_REDUCTION:  
    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)+".");  
    }  
 }  
   
   
427  }   // end anonymous namespace  }   // end anonymous namespace
428    
429    
# Line 439  opToString(ES_optype op) Line 439  opToString(ES_optype op)
439    return ES_opstrings[op];    return ES_opstrings[op];
440  }  }
441    
 #ifdef LAZY_NODE_STORAGE  
442  void DataLazy::LazyNodeSetup()  void DataLazy::LazyNodeSetup()
443  {  {
444  #ifdef _OPENMP  #ifdef _OPENMP
# Line 456  void DataLazy::LazyNodeSetup() Line 455  void DataLazy::LazyNodeSetup()
455      m_sampleids[0]=-1;      m_sampleids[0]=-1;
456  #endif  // _OPENMP  #endif  // _OPENMP
457  }  }
 #endif   // LAZY_NODE_STORAGE  
458    
459    
460  // Creates an identity node  // Creates an identity node
461  DataLazy::DataLazy(DataAbstract_ptr p)  DataLazy::DataLazy(DataAbstract_ptr p)
462      : parent(p->getFunctionSpace(),p->getShape())      : parent(p->getFunctionSpace(),p->getShape())
 #ifdef LAZY_NODE_STORAGE  
463      ,m_sampleids(0),      ,m_sampleids(0),
464      m_samples(1)      m_samples(1)
 #endif  
465  {  {
466     if (p->isLazy())     if (p->isLazy())
467     {     {
# Line 507  DataLazy::DataLazy(DataAbstract_ptr left Line 503  DataLazy::DataLazy(DataAbstract_ptr left
503     }     }
504     m_readytype=lleft->m_readytype;     m_readytype=lleft->m_readytype;
505     m_left=lleft;     m_left=lleft;
    m_buffsRequired=calcBuffs(m_left, m_right,m_op); // yeah m_right will be null at this point  
506     m_samplesize=getNumDPPSample()*getNoValues();     m_samplesize=getNumDPPSample()*getNoValues();
    m_maxsamplesize=max(m_samplesize,m_left->getMaxSampleSize());  
507     m_children=m_left->m_children+1;     m_children=m_left->m_children+1;
508     m_height=m_left->m_height+1;     m_height=m_left->m_height+1;
 #ifdef LAZY_NODE_STORAGE  
509     LazyNodeSetup();     LazyNodeSetup();
 #endif  
510     SIZELIMIT     SIZELIMIT
511  }  }
512    
# Line 581  LAZYDEBUG(cout << "Right " << right.get( Line 573  LAZYDEBUG(cout << "Right " << right.get(
573      m_readytype='C';      m_readytype='C';
574     }     }
575     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);  
576     m_children=m_left->m_children+m_right->m_children+2;     m_children=m_left->m_children+m_right->m_children+2;
577     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  
578     LazyNodeSetup();     LazyNodeSetup();
 #endif  
579     SIZELIMIT     SIZELIMIT
580  LAZYDEBUG(cout << "(3)Lazy created with " << m_samplesize << endl;)  LAZYDEBUG(cout << "(3)Lazy created with " << m_samplesize << endl;)
581  }  }
# Line 651  DataLazy::DataLazy(DataAbstract_ptr left Line 639  DataLazy::DataLazy(DataAbstract_ptr left
639      m_readytype='C';      m_readytype='C';
640     }     }
641     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);  
642     m_children=m_left->m_children+m_right->m_children+2;     m_children=m_left->m_children+m_right->m_children+2;
643     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  
644     LazyNodeSetup();     LazyNodeSetup();
 #endif  
645     SIZELIMIT     SIZELIMIT
646  LAZYDEBUG(cout << "(4)Lazy created with " << m_samplesize << endl;)  LAZYDEBUG(cout << "(4)Lazy created with " << m_samplesize << endl;)
647  }  }
# Line 685  DataLazy::DataLazy(DataAbstract_ptr left Line 669  DataLazy::DataLazy(DataAbstract_ptr left
669     }     }
670     m_readytype=lleft->m_readytype;     m_readytype=lleft->m_readytype;
671     m_left=lleft;     m_left=lleft;
    m_buffsRequired=calcBuffs(m_left, m_right,m_op); // yeah m_right will be null at this point  
672     m_samplesize=getNumDPPSample()*getNoValues();     m_samplesize=getNumDPPSample()*getNoValues();
    m_maxsamplesize=max(m_samplesize,m_left->getMaxSampleSize());  
673     m_children=m_left->m_children+1;     m_children=m_left->m_children+1;
674     m_height=m_left->m_height+1;     m_height=m_left->m_height+1;
 #ifdef LAZY_NODE_STORAGE  
675     LazyNodeSetup();     LazyNodeSetup();
 #endif  
676     SIZELIMIT     SIZELIMIT
677  LAZYDEBUG(cout << "(5)Lazy created with " << m_samplesize << endl;)  LAZYDEBUG(cout << "(5)Lazy created with " << m_samplesize << endl;)
678  }  }
# Line 719  DataLazy::DataLazy(DataAbstract_ptr left Line 699  DataLazy::DataLazy(DataAbstract_ptr left
699     }     }
700     m_readytype=lleft->m_readytype;     m_readytype=lleft->m_readytype;
701     m_left=lleft;     m_left=lleft;
    m_buffsRequired=calcBuffs(m_left, m_right,m_op); // yeah m_right will be null at this point  
702     m_samplesize=getNumDPPSample()*getNoValues();     m_samplesize=getNumDPPSample()*getNoValues();
    m_maxsamplesize=max(m_samplesize,m_left->getMaxSampleSize());  
703     m_children=m_left->m_children+1;     m_children=m_left->m_children+1;
704     m_height=m_left->m_height+1;     m_height=m_left->m_height+1;
 #ifdef LAZY_NODE_STORAGE  
705     LazyNodeSetup();     LazyNodeSetup();
 #endif  
706     SIZELIMIT     SIZELIMIT
707  LAZYDEBUG(cout << "(6)Lazy created with " << m_samplesize << endl;)  LAZYDEBUG(cout << "(6)Lazy created with " << m_samplesize << endl;)
708  }  }
# Line 754  DataLazy::DataLazy(DataAbstract_ptr left Line 730  DataLazy::DataLazy(DataAbstract_ptr left
730     }     }
731     m_readytype=lleft->m_readytype;     m_readytype=lleft->m_readytype;
732     m_left=lleft;     m_left=lleft;
    m_buffsRequired=calcBuffs(m_left, m_right,m_op); // yeah m_right will be null at this point  
733     m_samplesize=getNumDPPSample()*getNoValues();     m_samplesize=getNumDPPSample()*getNoValues();
    m_maxsamplesize=max(m_samplesize,m_left->getMaxSampleSize());  
734     m_children=m_left->m_children+1;     m_children=m_left->m_children+1;
735     m_height=m_left->m_height+1;     m_height=m_left->m_height+1;
 #ifdef LAZY_NODE_STORAGE  
736     LazyNodeSetup();     LazyNodeSetup();
 #endif  
737     SIZELIMIT     SIZELIMIT
738  LAZYDEBUG(cout << "(7)Lazy created with " << m_samplesize << endl;)  LAZYDEBUG(cout << "(7)Lazy created with " << m_samplesize << endl;)
739  }  }
740    
741  DataLazy::~DataLazy()  DataLazy::~DataLazy()
742  {  {
 #ifdef LAZY_NODE_SETUP  
743     delete[] m_sampleids;     delete[] m_sampleids;
    delete[] m_samples;  
 #endif  
 }  
   
   
 int  
 DataLazy::getBuffsRequired() const  
 {  
     return m_buffsRequired;  
 }  
   
   
 size_t  
 DataLazy::getMaxSampleSize() const  
 {  
     return m_maxsamplesize;  
744  }  }
745    
746    
   
 size_t  
 DataLazy::getSampleBufferSize() const  
 {  
     return m_maxsamplesize*(max(1,m_buffsRequired));  
 }  
   
747  /*  /*
748    \brief Evaluates the expression using methods on Data.    \brief Evaluates the expression using methods on Data.
749    This does the work for the collapse method.    This does the work for the collapse method.
# Line 971  DataLazy::collapse() Line 919  DataLazy::collapse()
919    m_op=IDENTITY;    m_op=IDENTITY;
920  }  }
921    
 /*  
   \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->resolveVectorSample(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;  
 }  
   
922    
 /*  
   \brief Compute the value of the expression (reduction 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::resolveReduction(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->resolveVectorSample(v,offset,sampleNo,roffset);  
   double* result=&(v[offset]);  
   roffset=offset;  
   unsigned int ndpps=getNumDPPSample();  
   unsigned int psize=DataTypes::noValues(getShape());  
   switch (m_op)  
   {  
     case MINVAL:  
     {  
       for (unsigned int z=0;z<ndpps;++z)  
       {  
          FMin op;  
          *result=DataMaths::reductionOp(*vleft, m_left->getShape(), roffset, op, numeric_limits<double>::max());  
          roffset+=psize;  
          result++;  
       }  
     }  
     break;  
     case MAXVAL:  
     {  
       for (unsigned int z=0;z<ndpps;++z)  
       {  
          FMax op;  
          *result=DataMaths::reductionOp(*vleft, m_left->getShape(), roffset, op, numeric_limits<double>::max()*-1);  
          roffset+=psize;  
          result++;  
       }  
     }  
     break;  
     default:  
     throw DataException("Programmer error - resolveReduction can not resolve operator "+opToString(m_op)+".");  
   }  
   return &v;  
 }  
923    
924    
925    
 /*  
   \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->resolveVectorSample(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->resolveVectorSample(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;  
 }  
   
   
 /*  
   \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->resolveVectorSample(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;  
 }  
   
   
926    
927  #define PROC_OP(TYPE,X)                               \  #define PROC_OP(TYPE,X)                               \
928      for (int j=0;j<onumsteps;++j)\      for (int j=0;j<onumsteps;++j)\
# Line 1361  LAZYDEBUG(cout << " result=      " << re Line 940  LAZYDEBUG(cout << " result=      " << re
940        rroffset+=orightstep;\        rroffset+=orightstep;\
941      }      }
942    
 /*  
   \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->resolveVectorSample(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->resolveVectorSample(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->resolveVectorSample(v,gap,sampleNo,lroffset);  
   gap+=m_left->getMaxSampleSize();  
   
   
 LAZYDEBUG(cout << "Query right with offset=" << gap << endl;)  
   
   
   const ValueType* right=m_right->resolveVectorSample(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  
943    
944  // The result will be stored in m_samples  // The result will be stored in m_samples
945  // 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 1667  LAZYDEBUG(cout << "Resolve sample " << t Line 955  LAZYDEBUG(cout << "Resolve sample " << t
955    if (m_op==IDENTITY)      if (m_op==IDENTITY)  
956    {    {
957      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);  
 //     }  
958      roffset=m_id->getPointOffset(sampleNo, 0);      roffset=m_id->getPointOffset(sampleNo, 0);
959    #ifdef LAZY_STACK_PROF
960    int x;
961    if (&x<stackend[omp_get_thread_num()])
962    {
963           stackend[omp_get_thread_num()]=&x;
964    }
965    #endif
966      return &(vec);      return &(vec);
967    }    }
968    if (m_readytype!='E')    if (m_readytype!='E')
# Line 2238  LAZYDEBUG(cout << DataTypes::pointToStri Line 1528  LAZYDEBUG(cout << DataTypes::pointToStri
1528    roffset=offset;    roffset=offset;
1529    return &m_samples;    return &m_samples;
1530  }  }
 #endif //LAZY_NODE_STORAGE  
1531    
 /*  
   \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.  
1532    
   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.  
   
 // the roffset is the offset within the returned vector where the data begins  
1533  const DataTypes::ValueType*  const DataTypes::ValueType*
1534  DataLazy::resolveVectorSample(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);  
   case G_REDUCTION: return resolveReduction(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)  
1535  {  {
1536  #ifdef _OPENMP  #ifdef _OPENMP
1537      int tid=omp_get_thread_num();      int tid=omp_get_thread_num();
1538  #else  #else
1539      int tid=0;      int tid=0;
1540  #endif  #endif
1541  #ifdef LAZY_NODE_STORAGE  
1542      return resolveNodeSample(tid, sampleNo, roffset);  #ifdef LAZY_STACK_PROF
1543        stackstart[tid]=&tid;
1544        stackend[tid]=&tid;
1545        const DataTypes::ValueType* r=resolveNodeSample(tid, sampleNo, roffset);
1546        size_t d=(size_t)stackstart[tid]-(size_t)stackend[tid];
1547        #pragma omp critical
1548        if (d>maxstackuse)
1549        {
1550    cout << "Max resolve Stack use " << d << endl;
1551            maxstackuse=d;
1552        }
1553        return r;
1554  #else  #else
1555      return resolveVectorSample(bg.getBuffer(tid),bg.getOffset(tid),sampleNo,roffset);      return resolveNodeSample(tid, sampleNo, roffset);
1556  #endif  #endif
1557  }  }
1558    
# Line 2320  DataLazy::resolveToIdentity() Line 1563  DataLazy::resolveToIdentity()
1563  {  {
1564     if (m_op==IDENTITY)     if (m_op==IDENTITY)
1565      return;      return;
 #ifndef LAZY_NODE_STORAGE  
    DataReady_ptr p=resolveVectorWorker();  
 #else  
1566     DataReady_ptr p=resolveNodeWorker();     DataReady_ptr p=resolveNodeWorker();
 #endif  
1567     makeIdentity(p);     makeIdentity(p);
1568  }  }
1569    
# Line 2340  void DataLazy::makeIdentity(const DataRe Line 1579  void DataLazy::makeIdentity(const DataRe
1579     else if(p->isExpanded()) {m_readytype='E';}     else if(p->isExpanded()) {m_readytype='E';}
1580     else if (p->isTagged()) {m_readytype='T';}     else if (p->isTagged()) {m_readytype='T';}
1581     else {throw DataException("Unknown DataReady instance in convertToIdentity constructor.");}     else {throw DataException("Unknown DataReady instance in convertToIdentity constructor.");}
    m_buffsRequired=1;  
1582     m_samplesize=p->getNumDPPSample()*p->getNoValues();     m_samplesize=p->getNumDPPSample()*p->getNoValues();
    m_maxsamplesize=m_samplesize;  
1583     m_left.reset();     m_left.reset();
1584     m_right.reset();     m_right.reset();
1585  }  }
# Line 2355  DataLazy::resolve() Line 1592  DataLazy::resolve()
1592      return m_id;      return m_id;
1593  }  }
1594    
 #ifdef LAZY_NODE_STORAGE  
   
1595  // This version of resolve uses storage in each node to hold results  // This version of resolve uses storage in each node to hold results
1596  DataReady_ptr  DataReady_ptr
1597  DataLazy::resolveNodeWorker()  DataLazy::resolveNodeWorker()
# Line 2378  DataLazy::resolveNodeWorker() Line 1613  DataLazy::resolveNodeWorker()
1613    int totalsamples=getNumSamples();    int totalsamples=getNumSamples();
1614    const ValueType* res=0;   // Storage for answer    const ValueType* res=0;   // Storage for answer
1615  LAZYDEBUG(cout << "Total number of samples=" <<totalsamples << endl;)  LAZYDEBUG(cout << "Total number of samples=" <<totalsamples << endl;)
1616    #pragma omp parallel for private(sample,res) schedule(static)    #pragma omp parallel private(sample,res)
   for (sample=0;sample<totalsamples;++sample)  
1617    {    {
1618      size_t roffset=0;      size_t roffset=0;
1619    #ifdef LAZY_STACK_PROF
1620        stackstart[omp_get_thread_num()]=&roffset;
1621        stackend[omp_get_thread_num()]=&roffset;
1622    #endif
1623        #pragma omp for schedule(static)
1624        for (sample=0;sample<totalsamples;++sample)
1625        {
1626            roffset=0;
1627  #ifdef _OPENMP  #ifdef _OPENMP
1628      res=resolveNodeSample(omp_get_thread_num(),sample,roffset);              res=resolveNodeSample(omp_get_thread_num(),sample,roffset);
1629  #else  #else
1630      res=resolveNodeSample(0,sample,roffset);              res=resolveNodeSample(0,sample,roffset);
1631  #endif  #endif
1632  LAZYDEBUG(cout << "Sample #" << sample << endl;)  LAZYDEBUG(cout << "Sample #" << sample << endl;)
1633  LAZYDEBUG(cout << "Final res[" << roffset<< "]=" << (*res)[roffset] << (*res)[roffset]<< endl; )  LAZYDEBUG(cout << "Final res[" << roffset<< "]=" << (*res)[roffset] << (*res)[roffset]<< endl; )
1634      DataVector::size_type outoffset=result->getPointOffset(sample,0);              DataVector::size_type outoffset=result->getPointOffset(sample,0);
1635      memcpy(&(resvec[outoffset]),&((*res)[roffset]),m_samplesize*sizeof(DataVector::ElementType));              memcpy(&(resvec[outoffset]),&((*res)[roffset]),m_samplesize*sizeof(DataVector::ElementType));
1636    }      }
   return resptr;  
 }  
   
 #endif // LAZY_NODE_STORAGE  
   
 // To simplify the memory management, all threads operate on one large vector, rather than one each.  
 // Each sample is evaluated independently and copied into the result DataExpanded.  
 DataReady_ptr  
 DataLazy::resolveVectorWorker()  
 {  
   
 LAZYDEBUG(cout << "Sample size=" << m_samplesize << endl;)  
 LAZYDEBUG(cout << "Buffers=" << m_buffsRequired << endl;)  
   if (m_readytype!='E')     // if the whole sub-expression is Constant or Tagged, then evaluate it normally  
   {  
     collapse();  
1637    }    }
1638    if (m_op==IDENTITY)       // So a lazy expression of Constant or Tagged data will be returned here.  #ifdef LAZY_STACK_PROF
1639      for (int i=0;i<getNumberOfThreads();++i)
1640    {    {
1641      return m_id;      size_t r=((size_t)stackstart[i] - (size_t)stackend[i]);
1642    //  cout << i << " " << stackstart[i] << " .. " << stackend[i] << " = " <<  r << endl;
1643        if (r>maxstackuse)
1644        {
1645            maxstackuse=r;
1646        }
1647    }    }
1648      // from this point on we must have m_op!=IDENTITY and m_readytype=='E'    cout << "Max resolve Stack use=" << maxstackuse << endl;
   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;)  
   DataExpanded* result=new DataExpanded(getFunctionSpace(),getShape(),  ValueType(getNoValues()));  
   ValueType& resvec=result->getVectorRW();  
   DataReady_ptr resptr=DataReady_ptr(result);  
   int sample;  
   size_t outoffset;     // offset in the output data  
   int totalsamples=getNumSamples();  
   const ValueType* res=0;   // Vector storing the answer  
   size_t resoffset=0;       // where in the vector to find the answer  
 LAZYDEBUG(cout << "Total number of samples=" <<totalsamples << endl;)  
   #pragma omp parallel for private(sample,resoffset,outoffset,res) schedule(static)  
   for (sample=0;sample<totalsamples;++sample)  
   {  
 LAZYDEBUG(cout << "################################# " << sample << endl;)  
 #ifdef _OPENMP  
     res=resolveVectorSample(v,threadbuffersize*omp_get_thread_num(),sample,resoffset);  
 #else  
     res=resolveVectorSample(v,0,sample,resoffset);   // res would normally be v, but not if its a single IDENTITY op.  
1649  #endif  #endif
 LAZYDEBUG(cerr << "-------------------------------- " << endl;)  
 LAZYDEBUG(cerr<< "Copying sample#" << sample << endl;)  
     outoffset=result->getPointOffset(sample,0);  
 LAZYDEBUG(cerr << "offset=" << outoffset << " from offset=" << resoffset << " " << m_samplesize << " doubles" << endl;)  
     for (unsigned int i=0;i<m_samplesize;++i,++outoffset,++resoffset)   // copy values into the output vector  
     {  
 LAZYDEBUG(cerr << "outoffset=" << outoffset << " resoffset=" << resoffset << " " << (*res)[resoffset]<< endl;)  
     resvec[outoffset]=(*res)[resoffset];  
     }  
 LAZYDEBUG(cerr << DataTypes::pointToString(resvec,getShape(),outoffset-m_samplesize+DataTypes::noValues(getShape()),"Final result:") << endl;)  
 LAZYDEBUG(cerr << "*********************************" << endl;)  
   }  
1650    return resptr;    return resptr;
1651  }  }
1652    
# Line 2459  std::string Line 1654  std::string
1654  DataLazy::toString() const  DataLazy::toString() const
1655  {  {
1656    ostringstream oss;    ostringstream oss;
1657    oss << "Lazy Data:";    oss << "Lazy Data: [depth=" << m_height<< "] ";
1658    if (escriptParams.getPRINT_LAZY_TREE()==0)    if (escriptParams.getPRINT_LAZY_TREE()==0)
1659    {    {
1660        intoString(oss);        intoString(oss);

Legend:
Removed from v.2737  
changed lines
  Added in v.2791

  ViewVC Help
Powered by ViewVC 1.1.26