C++ Teuchos::ptrFromRef方法代码示例

void DefaultIdentityLinearOp<Scalar>::applyImpl(
  const EOpTransp M_trans,
  const MultiVectorBase<Scalar> &X,
  const Ptr<MultiVectorBase<Scalar> > &Y,
  const Scalar alpha,
  const Scalar beta
  ) const
{
  using Teuchos::tuple;
  using Teuchos::ptrFromRef;
#ifdef TEUCHOS_DEBUG
  THYRA_ASSERT_LINEAR_OP_MULTIVEC_APPLY_SPACES(
    "DefaultIdentityLinearOp<Scalar>::apply(...)", *this, M_trans, X, &*Y
    );
#endif // TEUCHOS_DEBUG  
  Thyra::linear_combination<Scalar>(
    tuple<Scalar>(alpha)(),
    tuple<Ptr<const MultiVectorBase<Scalar> > >(ptrFromRef(X))(),
    beta, Y
    );
}

void panzer::GatherTangent_BlockedTpetra<EvalT, TRAITS,S,LO,GO,NodeT>::
evaluateFields(typename TRAITS::EvalData workset)
{
   using Teuchos::RCP;
   using Teuchos::ArrayRCP;
   using Teuchos::ptrFromRef;
   using Teuchos::rcp_dynamic_cast;

   using Thyra::VectorBase;
   using Thyra::SpmdVectorBase;
   using Thyra::ProductVectorBase;

   // If blockedContainer_ was not initialized, then no global evaluation data
   // container was set, in which case this evaluator becomes a no-op
   if (blockedContainer_ == Teuchos::null)
     return;

   Teuchos::FancyOStream out(Teuchos::rcpFromRef(std::cout));
   out.setShowProcRank(true);
   out.setOutputToRootOnly(-1);

   std::vector<std::pair<int,GO> > GIDs;
   std::vector<LO> LIDs;

   // for convenience pull out some objects from workset
   std::string blockId = this->wda(workset).block_id;
   const std::vector<std::size_t> & localCellIds = this->wda(workset).cell_local_ids;

   Teuchos::RCP<ProductVectorBase<double> > x;
   if (useTimeDerivativeSolutionVector_)
     x = rcp_dynamic_cast<ProductVectorBase<double> >(blockedContainer_->get_dxdt());
   else
     x = rcp_dynamic_cast<ProductVectorBase<double> >(blockedContainer_->get_x());

   // gather operation for each cell in workset
   for(std::size_t worksetCellIndex=0;worksetCellIndex<localCellIds.size();++worksetCellIndex) {
      LO cellLocalId = localCellIds[worksetCellIndex];

      gidIndexer_->getElementGIDs(cellLocalId,GIDs,blockId);

      // caculate the local IDs for this element
      LIDs.resize(GIDs.size());
      for(std::size_t i=0;i<GIDs.size();i++) {
         // used for doing local ID lookups
         RCP<const MapType> x_map = blockedContainer_->getMapForBlock(GIDs[i].first);

         LIDs[i] = x_map->getLocalElement(GIDs[i].second);
      }

      // loop over the fields to be gathered
      Teuchos::ArrayRCP<const double> local_x;
      for (std::size_t fieldIndex=0; fieldIndex<gatherFields_.size();fieldIndex++) {
         int fieldNum = fieldIds_[fieldIndex];
         int indexerId = gidIndexer_->getFieldBlock(fieldNum);

         // grab local data for inputing
         RCP<SpmdVectorBase<double> > block_x = rcp_dynamic_cast<SpmdVectorBase<double> >(x->getNonconstVectorBlock(indexerId));
         block_x->getLocalData(ptrFromRef(local_x));

         const std::vector<int> & elmtOffset = gidIndexer_->getGIDFieldOffsets(blockId,fieldNum);

         // loop over basis functions and fill the fields
         for(std::size_t basis=0;basis<elmtOffset.size();basis++) {
            int offset = elmtOffset[basis];
            int lid = LIDs[offset];

            (gatherFields_[fieldIndex])(worksetCellIndex,basis) = local_x[lid];
         }
      }
   }
}

void panzer::ScatterResidual_BlockedEpetra<panzer::Traits::Jacobian, Traits,LO,GO>::
evaluateFields(typename Traits::EvalData workset)
{ 
   using Teuchos::RCP;
   using Teuchos::ArrayRCP;
   using Teuchos::ptrFromRef;
   using Teuchos::rcp_dynamic_cast;

   using Thyra::VectorBase;
   using Thyra::SpmdVectorBase;
   using Thyra::ProductVectorBase;
   using Thyra::BlockedLinearOpBase;

   typedef BlockedEpetraLinearObjContainer BLOC;

   std::vector<std::pair<int,GO> > GIDs;
   std::vector<LO> LIDs;
   std::vector<double> jacRow;

   // for convenience pull out some objects from workset
   std::string blockId = workset.block_id;
   const std::vector<std::size_t> & localCellIds = workset.cell_local_ids;

   RCP<const BLOC> blockedContainer = blockedContainer_;

   RCP<ProductVectorBase<double> > r = rcp_dynamic_cast<ProductVectorBase<double> >(blockedContainer->get_f());
   Teuchos::RCP<BlockedLinearOpBase<double> > Jac = rcp_dynamic_cast<BlockedLinearOpBase<double> >(blockedContainer->get_A());

   int numFieldBlocks = globalIndexer_->getNumFieldBlocks();
   std::vector<int> blockOffsets(numFieldBlocks+1); // number of fields, plus a sentinnel
   for(int blk=0;blk<numFieldBlocks;blk++) {
      int blockOffset = globalIndexer_->getBlockGIDOffset(blockId,blk);
      blockOffsets[blk] = blockOffset;
   }

   boost::unordered_map<std::pair<int,int>,Teuchos::RCP<Epetra_CrsMatrix> > jacEpetraBlocks;

   // NOTE: A reordering of these loops will likely improve performance
   //       The "getGIDFieldOffsets" may be expensive.  However the
   //       "getElementGIDs" can be cheaper. However the lookup for LIDs
   //       may be more expensive!

   // scatter operation for each cell in workset
   for(std::size_t worksetCellIndex=0;worksetCellIndex<localCellIds.size();++worksetCellIndex) {
      std::size_t cellLocalId = localCellIds[worksetCellIndex];

      globalIndexer_->getElementGIDs(cellLocalId,GIDs,blockId); 

      // caculate the local IDs for this element
      LIDs.resize(GIDs.size());
      for(std::size_t i=0;i<GIDs.size();i++) {
         // used for doing local ID lookups
         RCP<const Epetra_Map> r_map = blockedContainer->getMapForBlock(GIDs[i].first);

         LIDs[i] = r_map->LID(GIDs[i].second);
      }

      // loop over each field to be scattered
      Teuchos::ArrayRCP<double> local_r;
      for(std::size_t fieldIndex = 0; fieldIndex < scatterFields_.size(); fieldIndex++) {
         int fieldNum = fieldIds_[fieldIndex];
         int blockRowIndex = globalIndexer_->getFieldBlock(fieldNum);

         // grab local data for inputing
         if(r!=Teuchos::null) {
            RCP<SpmdVectorBase<double> > block_r = rcp_dynamic_cast<SpmdVectorBase<double> >(r->getNonconstVectorBlock(blockRowIndex));
            block_r->getNonconstLocalData(ptrFromRef(local_r));
         }

         const std::vector<int> & elmtOffset = globalIndexer_->getGIDFieldOffsets(blockId,fieldNum);
        
         // loop over the basis functions (currently they are nodes)
         for(std::size_t rowBasisNum = 0; rowBasisNum < elmtOffset.size(); rowBasisNum++) {
            const ScalarT & scatterField = (scatterFields_[fieldIndex])(worksetCellIndex,rowBasisNum);
            int rowOffset = elmtOffset[rowBasisNum];
            int r_lid = LIDs[rowOffset];
    
            // Sum residual
            if(local_r!=Teuchos::null)
               local_r[r_lid] += (scatterField.val());

            blockOffsets[numFieldBlocks] = scatterField.size(); // add the sentinel
    
            // loop over the sensitivity indices: all DOFs on a cell
            jacRow.resize(scatterField.size());
            
            for(int sensIndex=0;sensIndex<scatterField.size();++sensIndex) {
               jacRow[sensIndex] = scatterField.fastAccessDx(sensIndex);
            }
    
            for(int blockColIndex=0;blockColIndex<numFieldBlocks;blockColIndex++) {
               int start = blockOffsets[blockColIndex];
               int end = blockOffsets[blockColIndex+1];

               if(end-start<=0) 
                  continue;

               // check hash table for jacobian sub block
               std::pair<int,int> blockIndex = std::make_pair(blockRowIndex,blockColIndex);
               Teuchos::RCP<Epetra_CrsMatrix> subJac = jacEpetraBlocks[blockIndex];
//.........这里部分代码省略.........

void panzer::GatherSolution_BlockedTpetra<panzer::Traits::Jacobian, TRAITS,S,LO,GO,NodeT>::
evaluateFields(typename TRAITS::EvalData workset)
{
   using Teuchos::RCP;
   using Teuchos::ArrayRCP;
   using Teuchos::ptrFromRef;
   using Teuchos::rcp_dynamic_cast;

   using Thyra::VectorBase;
   using Thyra::SpmdVectorBase;
   using Thyra::ProductVectorBase;

   std::vector<std::pair<int,GO> > GIDs;
   std::vector<LO> LIDs;

   // for convenience pull out some objects from workset
   std::string blockId = this->wda(workset).block_id;
   const std::vector<std::size_t> & localCellIds = this->wda(workset).cell_local_ids;

   double seed_value = 0.0;
   Teuchos::RCP<ProductVectorBase<double> > x;
   if (useTimeDerivativeSolutionVector_) {
     x = rcp_dynamic_cast<ProductVectorBase<double> >(blockedContainer_->get_dxdt());
     seed_value = workset.alpha;
   }
   else {
     x = rcp_dynamic_cast<ProductVectorBase<double> >(blockedContainer_->get_x());
     seed_value = workset.beta;
   }

   // turn off sensitivies: this may be faster if we don't expand the term
   // but I suspect not because anywhere it is used the full complement of
   // sensitivies will be needed anyway.
   if(disableSensitivities_)
      seed_value = 0.0;

   // NOTE: A reordering of these loops will likely improve performance
   //       The "getGIDFieldOffsets may be expensive.  However the
   //       "getElementGIDs" can be cheaper. However the lookup for LIDs
   //       may be more expensive!

   // gather operation for each cell in workset
   for(std::size_t worksetCellIndex=0;worksetCellIndex<localCellIds.size();++worksetCellIndex) {
      LO cellLocalId = localCellIds[worksetCellIndex];

      gidIndexer_->getElementGIDs(cellLocalId,GIDs,blockId);

      // caculate the local IDs for this element
      LIDs.resize(GIDs.size());
      for(std::size_t i=0;i<GIDs.size();i++) {
         // used for doing local ID lookups
         RCP<const MapType> x_map = blockedContainer_->getMapForBlock(GIDs[i].first);

         LIDs[i] = x_map->getLocalElement(GIDs[i].second);
      }

      // loop over the fields to be gathered
      Teuchos::ArrayRCP<const double> local_x;
      for(std::size_t fieldIndex=0;
          fieldIndex<gatherFields_.size();fieldIndex++) {
         int fieldNum = fieldIds_[fieldIndex];
         int indexerId = gidIndexer_->getFieldBlock(fieldNum);

         // grab local data for inputing
         RCP<SpmdVectorBase<double> > block_x = rcp_dynamic_cast<SpmdVectorBase<double> >(x->getNonconstVectorBlock(indexerId));
         block_x->getLocalData(ptrFromRef(local_x));

         const std::vector<int> & elmtOffset = gidIndexer_->getGIDFieldOffsets(blockId,fieldNum);

         // loop over basis functions and fill the fields
         for(std::size_t basis=0;basis<elmtOffset.size();basis++) {
            int offset = elmtOffset[basis];
            int lid = LIDs[offset];

            // set the value and seed the FAD object
            (gatherFields_[fieldIndex])(worksetCellIndex,basis) = ScalarT(GIDs.size(), local_x[lid]);
            (gatherFields_[fieldIndex])(worksetCellIndex,basis).fastAccessDx(offset) = seed_value;
         }
      }
   }
}

void panzer::ScatterDirichletResidual_BlockedTpetra<panzer::Traits::Jacobian, TRAITS,LO,GO,NodeT>::
evaluateFields(typename TRAITS::EvalData workset)
{
    using Teuchos::RCP;
    using Teuchos::ArrayRCP;
    using Teuchos::ptrFromRef;
    using Teuchos::rcp_dynamic_cast;

    using Thyra::VectorBase;
    using Thyra::SpmdVectorBase;
    using Thyra::ProductVectorBase;
    using Thyra::BlockedLinearOpBase;

    std::vector<std::pair<int,GO> > GIDs;
    std::vector<LO> LIDs;

    // for convenience pull out some objects from workset
    std::string blockId = this->wda(workset).block_id;
    const std::vector<std::size_t> & localCellIds = this->wda(workset).cell_local_ids;

    RCP<ProductVectorBase<double> > r = rcp_dynamic_cast<ProductVectorBase<double> >(blockedContainer_->get_f());
    Teuchos::RCP<BlockedLinearOpBase<double> > Jac = rcp_dynamic_cast<BlockedLinearOpBase<double> >(blockedContainer_->get_A());

    int numFieldBlocks = globalIndexer_->getNumFieldBlocks();
    std::vector<int> blockOffsets(numFieldBlocks+1); // number of fields, plus a sentinnel
    for(int blk=0; blk<numFieldBlocks; blk++) {
        int blockOffset = globalIndexer_->getBlockGIDOffset(blockId,blk);
        blockOffsets[blk] = blockOffset;
    }

    std::unordered_map<std::pair<int,int>,Teuchos::RCP<CrsMatrixType>,panzer::pair_hash> jacTpetraBlocks;

    // NOTE: A reordering of these loops will likely improve performance
    //       The "getGIDFieldOffsets may be expensive.  However the
    //       "getElementGIDs" can be cheaper. However the lookup for LIDs
    //       may be more expensive!

    // scatter operation for each cell in workset
    for(std::size_t worksetCellIndex=0; worksetCellIndex<localCellIds.size(); ++worksetCellIndex) {
        std::size_t cellLocalId = localCellIds[worksetCellIndex];

        globalIndexer_->getElementGIDs(cellLocalId,GIDs);
        blockOffsets[numFieldBlocks] = GIDs.size();

        // caculate the local IDs for this element
        LIDs.resize(GIDs.size());
        for(std::size_t i=0; i<GIDs.size(); i++) {
            // used for doing local ID lookups
            RCP<const MapType> r_map = blockedContainer_->getMapForBlock(GIDs[i].first);

            LIDs[i] = r_map->getLocalElement(GIDs[i].second);
        }

        // loop over each field to be scattered
        Teuchos::ArrayRCP<double> local_r, local_dc;
        for(std::size_t fieldIndex = 0; fieldIndex < scatterFields_.size(); fieldIndex++) {
            int fieldNum = fieldIds_[fieldIndex];
            int blockRowIndex = globalIndexer_->getFieldBlock(fieldNum);

            RCP<SpmdVectorBase<double> > dc = rcp_dynamic_cast<SpmdVectorBase<double> >(dirichletCounter_->getNonconstVectorBlock(blockRowIndex));
            dc->getNonconstLocalData(ptrFromRef(local_dc));

            // grab local data for inputing
            RCP<SpmdVectorBase<double> > block_r = rcp_dynamic_cast<SpmdVectorBase<double> >(r->getNonconstVectorBlock(blockRowIndex));
            block_r->getNonconstLocalData(ptrFromRef(local_r));

            // this call "should" get the right ordering according to the Intrepid basis
            const std::pair<std::vector<int>,std::vector<int> > & indicePair
                = globalIndexer_->getGIDFieldOffsets_closure(blockId,fieldNum, side_subcell_dim_, local_side_id_);
            const std::vector<int> & elmtOffset = indicePair.first;
            const std::vector<int> & basisIdMap = indicePair.second;

            // loop over basis functions
            for(std::size_t basis=0; basis<elmtOffset.size(); basis++) {
                int offset = elmtOffset[basis];
                int lid = LIDs[offset];
                if(lid<0) // not on this processor
                    continue;

                int basisId = basisIdMap[basis];

                if (checkApplyBC_)
                    if (!applyBC_[fieldIndex](worksetCellIndex,basisId))
                        continue;

                // zero out matrix row
                for(int blockColIndex=0; blockColIndex<numFieldBlocks; blockColIndex++) {
                    int start = blockOffsets[blockColIndex];
                    int end = blockOffsets[blockColIndex+1];

                    if(end-start<=0)
                        continue;

                    // check hash table for jacobian sub block
                    std::pair<int,int> blockIndex = std::make_pair(blockRowIndex,blockColIndex);
                    Teuchos::RCP<CrsMatrixType> subJac = jacTpetraBlocks[blockIndex];

                    // if you didn't find one before, add it to the hash table
                    if(subJac==Teuchos::null) {
                        Teuchos::RCP<Thyra::LinearOpBase<double> > tOp = Jac->getNonconstBlock(blockIndex.first,blockIndex.second);
//.........这里部分代码省略.........

void panzer::GatherSolution_BlockedEpetra<panzer::Traits::Residual, TRAITS,LO,GO>::
evaluateFields(typename TRAITS::EvalData workset)
{ 
   using Teuchos::RCP;
   using Teuchos::ArrayRCP;
   using Teuchos::ptrFromRef;
   using Teuchos::rcp_dynamic_cast;

   using Thyra::VectorBase;
   using Thyra::SpmdVectorBase;
   using Thyra::ProductVectorBase;

   typedef BlockedEpetraLinearObjContainer BLOC;

   Teuchos::FancyOStream out(Teuchos::rcpFromRef(std::cout));
   out.setShowProcRank(true);
   out.setOutputToRootOnly(-1);

   std::vector<std::pair<int,GO> > GIDs;
   std::vector<int> LIDs;
 
   // for convenience pull out some objects from workset
   std::string blockId = this->wda(workset).block_id;
   const std::vector<std::size_t> & localCellIds = this->wda(workset).cell_local_ids;

   Teuchos::RCP<ProductVectorBase<double> > x;
   if (useTimeDerivativeSolutionVector_)
     x = rcp_dynamic_cast<ProductVectorBase<double> >(blockedContainer_->get_dxdt());
   else
     x = rcp_dynamic_cast<ProductVectorBase<double> >(blockedContainer_->get_x()); 

   // gather operation for each cell in workset
   for(std::size_t worksetCellIndex=0;worksetCellIndex<localCellIds.size();++worksetCellIndex) {
      LO cellLocalId = localCellIds[worksetCellIndex];
 
      gidIndexer_->getElementGIDs(cellLocalId,GIDs,blockId); 
 
      // caculate the local IDs for this element
      LIDs.resize(GIDs.size());
      for(std::size_t i=0;i<GIDs.size();i++) {
         // used for doing local ID lookups
         RCP<const Epetra_Map> x_map = blockedContainer_->getMapForBlock(GIDs[i].first);

         LIDs[i] = x_map->LID(GIDs[i].second);
         // TEUCHOS_ASSERT(LIDs[i]>=0);
      }

      // loop over the fields to be gathered
      Teuchos::ArrayRCP<const double> local_x;
      for (std::size_t fieldIndex=0; fieldIndex<gatherFields_.size();fieldIndex++) {
         int fieldNum = fieldIds_[fieldIndex];
         int indexerId = gidIndexer_->getFieldBlock(fieldNum);

         // grab local data for inputing
         RCP<SpmdVectorBase<double> > block_x = rcp_dynamic_cast<SpmdVectorBase<double> >(x->getNonconstVectorBlock(indexerId));
         block_x->getLocalData(ptrFromRef(local_x));

         const std::vector<int> & elmtOffset = gidIndexer_->getGIDFieldOffsets(blockId,fieldNum);
 
         // loop over basis functions and fill the fields
         for(std::size_t basis=0;basis<elmtOffset.size();basis++) {
            int offset = elmtOffset[basis];
            int lid = LIDs[offset];

            // TEUCHOS_ASSERT(indexerId==GIDs[offset].first);
            // TEUCHOS_ASSERT(lid<local_x.size() && lid>=0);

            (gatherFields_[fieldIndex])(worksetCellIndex,basis) = local_x[lid];
         }
      }
   }
}

Teuchos::RCP<const Epetra_Map>
Thyra::get_Epetra_Map(const VectorSpaceBase<double>& vs_in,
  const RCP<const Epetra_Comm>& comm)
{

  using Teuchos::rcpFromRef;
  using Teuchos::rcpFromPtr;
  using Teuchos::rcp_dynamic_cast;
  using Teuchos::ptrFromRef;
  using Teuchos::ptr_dynamic_cast;

  const Ptr<const VectorSpaceBase<double> > vs_ptr = ptrFromRef(vs_in);
  
  const Ptr<const SpmdVectorSpaceBase<double> > spmd_vs =
    ptr_dynamic_cast<const SpmdVectorSpaceBase<double> >(vs_ptr);

  const Ptr<const ProductVectorSpaceBase<double> > &prod_vs = 
    ptr_dynamic_cast<const ProductVectorSpaceBase<double> >(vs_ptr);

  TEUCHOS_TEST_FOR_EXCEPTION( is_null(spmd_vs) && is_null(prod_vs), std::logic_error,
    "Error, the concrete VectorSpaceBase object of type "
    +Teuchos::demangleName(typeid(vs_in).name())+" does not support the"
    " SpmdVectorSpaceBase or the ProductVectorSpaceBase interfaces!" );

  const int numBlocks = (nonnull(prod_vs) ? prod_vs->numBlocks() : 1);

  // Get an array of SpmdVectorBase objects for the blocks
  
  Array<RCP<const SpmdVectorSpaceBase<double> > > spmd_vs_blocks;
  if (nonnull(prod_vs)) {
    for (int block_i = 0; block_i < numBlocks; ++block_i) {
      const RCP<const SpmdVectorSpaceBase<double> > spmd_vs_i =
        rcp_dynamic_cast<const SpmdVectorSpaceBase<double> >(
          prod_vs->getBlock(block_i), true);
      spmd_vs_blocks.push_back(spmd_vs_i);
    }
  }
  else {
    spmd_vs_blocks.push_back(rcpFromPtr(spmd_vs));
  }
  
  // Find the number of local elements, summed over all blocks

  int myLocalElements = 0;
  for (int block_i = 0; block_i < numBlocks; ++block_i) {
    myLocalElements += spmd_vs_blocks[block_i]->localSubDim();
  }
  
  // Find the GIDs owned by this processor, taken from all blocks
  
  int count=0;
  int blockOffset = 0;
  Array<int> myGIDs(myLocalElements);
  for (int block_i = 0; block_i < numBlocks; ++block_i) {
    const RCP<const SpmdVectorSpaceBase<double> > spmd_vs_i = spmd_vs_blocks[block_i];
    const int lowGIDInBlock = spmd_vs_i->localOffset();
    const int numLocalElementsInBlock = spmd_vs_i->localSubDim();
    for (int i=0; i < numLocalElementsInBlock; ++i, ++count) {
      myGIDs[count] = blockOffset + lowGIDInBlock + i;
    }
    blockOffset += spmd_vs_i->dim();
  }
  
  const int globalDim = vs_in.dim();

  return Teuchos::rcp(
    new Epetra_Map(globalDim, myLocalElements, &(myGIDs[0]), 0, *comm));

}

void panzer::ScatterDirichletResidual_BlockedEpetra<panzer::Traits::Jacobian, Traits,LO,GO>::
evaluateFields(typename Traits::EvalData workset)
{ 
   using Teuchos::RCP;
   using Teuchos::ArrayRCP;
   using Teuchos::ptrFromRef;
   using Teuchos::rcp_dynamic_cast;

   using Thyra::VectorBase;
   using Thyra::SpmdVectorBase;
   using Thyra::ProductVectorBase;
   using Thyra::BlockedLinearOpBase;

   typedef BlockedEpetraLinearObjContainer BLOC;

   std::vector<std::pair<int,GO> > GIDs;
   std::vector<int> LIDs;
 
   // for convenience pull out some objects from workset
   std::string blockId = workset.block_id;
   const std::vector<std::size_t> & localCellIds = workset.cell_local_ids;

   RCP<const BLOC> blockedContainer = blockedContainer_;
   RCP<ProductVectorBase<double> > r = rcp_dynamic_cast<ProductVectorBase<double> >(blockedContainer->get_f());
   Teuchos::RCP<BlockedLinearOpBase<double> > Jac = rcp_dynamic_cast<BlockedLinearOpBase<double> >(blockedContainer->get_A());

   int numFieldBlocks = globalIndexer_->getNumFieldBlocks();
   std::vector<int> blockOffsets(numFieldBlocks+1); // number of fields, plus a sentinnel
   for(int blk=0;blk<numFieldBlocks;blk++) {
      int blockOffset = globalIndexer_->getBlockGIDOffset(blockId,blk);
      blockOffsets[blk] = blockOffset;
   }

   boost::unordered_map<std::pair<int,int>,Teuchos::RCP<Epetra_CrsMatrix> > jacEpetraBlocks;

   // NOTE: A reordering of these loops will likely improve performance
   //       The "getGIDFieldOffsets may be expensive.  However the
   //       "getElementGIDs" can be cheaper. However the lookup for LIDs
   //       may be more expensive!

   // scatter operation for each cell in workset
   for(std::size_t worksetCellIndex=0;worksetCellIndex<localCellIds.size();++worksetCellIndex) {
      std::size_t cellLocalId = localCellIds[worksetCellIndex];

      globalIndexer_->getElementGIDs(cellLocalId,GIDs); 
      blockOffsets[numFieldBlocks] = GIDs.size();

      // caculate the local IDs for this element
      LIDs.resize(GIDs.size());
      for(std::size_t i=0;i<GIDs.size();i++) {
         // used for doing local ID lookups
         RCP<const Epetra_Map> r_map = blockedContainer->getMapForBlock(GIDs[i].first);

         LIDs[i] = r_map->LID(GIDs[i].second);
      }

      std::vector<bool> is_owned(GIDs.size(), false);
      globalIndexer_->ownedIndices(GIDs,is_owned);

      // loop over each field to be scattered
      Teuchos::ArrayRCP<double> local_r, local_dc;
      for(std::size_t fieldIndex = 0; fieldIndex < scatterFields_.size(); fieldIndex++) {
         int fieldNum = fieldIds_[fieldIndex];
         int blockRowIndex = globalIndexer_->getFieldBlock(fieldNum);

         RCP<SpmdVectorBase<double> > dc = rcp_dynamic_cast<SpmdVectorBase<double> >(dirichletCounter_->getNonconstVectorBlock(blockRowIndex));
         dc->getNonconstLocalData(ptrFromRef(local_dc));

         // grab local data for inputing
         RCP<SpmdVectorBase<double> > block_r = rcp_dynamic_cast<SpmdVectorBase<double> >(r->getNonconstVectorBlock(blockRowIndex));
         block_r->getNonconstLocalData(ptrFromRef(local_r));
   
         // this call "should" get the right ordering accordint to the Intrepid basis
         const std::pair<std::vector<int>,std::vector<int> > & indicePair 
               = globalIndexer_->getGIDFieldOffsets_closure(blockId,fieldNum, side_subcell_dim_, local_side_id_);
         const std::vector<int> & elmtOffset = indicePair.first;
         const std::vector<int> & basisIdMap = indicePair.second;
   
         // loop over basis functions
         for(std::size_t basis=0;basis<elmtOffset.size();basis++) {
            int offset = elmtOffset[basis];
            int lid = LIDs[offset];
            if(lid<0) // not on this processor
               continue;

            int basisId = basisIdMap[basis];

	    if (checkApplyBC_)
	      if (!applyBC_[fieldIndex](worksetCellIndex,basisId))
		continue;

            // zero out matrix row
            for(int blockColIndex=0;blockColIndex<numFieldBlocks;blockColIndex++) {
               int start = blockOffsets[blockColIndex];
               int end = blockOffsets[blockColIndex+1];

               if(end-start<=0) 
                  continue;

               // check hash table for jacobian sub block
//.........这里部分代码省略.........