Intrepid2_Data.hpp

Go to the documentation of this file.

//
//  Intrepid2_Data.hpp
//  QuadraturePerformance
//
//  Created by Roberts, Nathan V on 8/24/20.
//
 
#ifndef Intrepid2_Data_h
#define Intrepid2_Data_h
 
#include "Intrepid2_ArgExtractor.hpp"
#include "Intrepid2_DataDimensionInfo.hpp"
#include "Intrepid2_DataFunctors.hpp"
#include "Intrepid2_DataVariationType.hpp"
#include "Intrepid2_ScalarView.hpp"
#include "Intrepid2_Utils.hpp"

 
namespace Intrepid2 {

template<class DataScalar,typename DeviceType>
class ZeroView {
public:
  static ScalarView<DataScalar,DeviceType> zeroView()
  {
    static ScalarView<DataScalar,DeviceType> zeroView = ScalarView<DataScalar,DeviceType>("zero",1);
    static bool havePushedFinalizeHook = false;
    if (!havePushedFinalizeHook)
    {
      Kokkos::push_finalize_hook( [=] {
        zeroView = ScalarView<DataScalar,DeviceType>();
      });
      havePushedFinalizeHook = true;
    }
    return zeroView;
  }
};

  template<class DataScalar,typename DeviceType>
  class Data {
  public:
    using value_type      = DataScalar;
    using execution_space = typename DeviceType::execution_space;
    
    using reference_type       = typename ScalarView<DataScalar,DeviceType>::reference_type;
    using const_reference_type = typename ScalarView<const DataScalar,DeviceType>::reference_type;
  private:
    ordinal_type dataRank_;
    Kokkos::View<DataScalar*,       DeviceType> data1_;  // the rank 1 data that is explicitly stored
    Kokkos::View<DataScalar**,      DeviceType> data2_;  // the rank 2 data that is explicitly stored
    Kokkos::View<DataScalar***,     DeviceType> data3_;  // the rank 3 data that is explicitly stored
    Kokkos::View<DataScalar****,    DeviceType> data4_;  // the rank 4 data that is explicitly stored
    Kokkos::View<DataScalar*****,   DeviceType> data5_;  // the rank 5 data that is explicitly stored
    Kokkos::View<DataScalar******,  DeviceType> data6_;  // the rank 6 data that is explicitly stored
    Kokkos::View<DataScalar*******, DeviceType> data7_;  // the rank 7 data that is explicitly stored
    Kokkos::Array<int,7> extents_;                     // logical extents in each dimension
    Kokkos::Array<DataVariationType,7> variationType_; // for each dimension, whether the data varies in that dimension
    Kokkos::Array<int,7> variationModulus_;            // for each dimension, a value by which indices should be modulused (only used when variationType_ is MODULAR)
    int blockPlusDiagonalLastNonDiagonal_ = -1;        // last row/column that is part of the non-diagonal part of the matrix indicated by BLOCK_PLUS_DIAGONAL (if any dimensions are thus marked)
    
    bool hasNontrivialModulusUNUSED_;  // this is a little nutty, but having this UNUSED member variable improves performance, probably by shifting the alignment of underlyingMatchesLogical_.  This is true with nvcc; it may also be true with Apple clang
    bool underlyingMatchesLogical_;   // if true, this Data object has the same rank and extent as the underlying view
    Kokkos::Array<ordinal_type,7> activeDims_;
    int numActiveDims_; // how many of the 7 entries are actually filled in
    
    ordinal_type rank_;
    
    // we use (const_)reference_type as the return for operator() for performance reasons, especially significant when using Sacado types
    using return_type = const_reference_type;
    
    ScalarView<DataScalar,DeviceType> zeroView_; // one-entry (zero); used to allow getEntry() to return 0 for off-diagonal entries in BLOCK_PLUS_DIAGONAL
    
    KOKKOS_INLINE_FUNCTION
    static int blockPlusDiagonalNumNondiagonalEntries(const int &lastNondiagonal)
    {
      return (lastNondiagonal + 1) * (lastNondiagonal + 1);
    }
    
    KOKKOS_INLINE_FUNCTION
    static int blockPlusDiagonalBlockEntryIndex(const int &lastNondiagonal, const int &numNondiagonalEntries, const int &i, const int &j)
    {
      return i * (lastNondiagonal + 1) + j;
    }
    
    KOKKOS_INLINE_FUNCTION
    static int blockPlusDiagonalDiagonalEntryIndex(const int &lastNondiagonal, const int &numNondiagonalEntries, const int &i)
    {
      return i - (lastNondiagonal + 1) + numNondiagonalEntries;
    }
    
    KOKKOS_INLINE_FUNCTION
    int getUnderlyingViewExtent(const int &dim) const
    {
      switch (dataRank_)
      {
        case 1: return data1_.extent_int(dim);
        case 2: return data2_.extent_int(dim);
        case 3: return data3_.extent_int(dim);
        case 4: return data4_.extent_int(dim);
        case 5: return data5_.extent_int(dim);
        case 6: return data6_.extent_int(dim);
        case 7: return data7_.extent_int(dim);
        default: return -1;
      }
    }
    
    void setActiveDims()
    {
      zeroView_ = ZeroView<DataScalar,DeviceType>::zeroView(); // one-entry (zero); used to allow getEntry() to return 0 for off-diagonal entries in BLOCK_PLUS_DIAGONAL
      // check that rank is compatible with the claimed extents:
      for (int d=rank_; d<7; d++)
      {
        INTREPID2_TEST_FOR_EXCEPTION(extents_[d] > 1, std::invalid_argument, "Nominal extents may not be > 1 in dimensions beyond the rank of the container");
      }
      
      numActiveDims_ = 0;
      int blockPlusDiagonalCount = 0;
      underlyingMatchesLogical_ = true;
      for (ordinal_type i=0; i<7; i++)
      {
        if (variationType_[i] == GENERAL)
        {
          if (extents_[i] != 0)
          {
            variationModulus_[i] = extents_[i];
          }
          else
          {
            variationModulus_[i] = 1;
          }
          activeDims_[numActiveDims_] = i;
          numActiveDims_++;
        }
        else if (variationType_[i] == MODULAR)
        {
          underlyingMatchesLogical_ = false;
          if (extents_[i] != getUnderlyingViewExtent(numActiveDims_))
          {
            const int dataExtent = getUnderlyingViewExtent(numActiveDims_);
            const int logicalExtent = extents_[i];
            const int modulus = dataExtent;
            
            INTREPID2_TEST_FOR_EXCEPTION( dataExtent * (logicalExtent / dataExtent) != logicalExtent, std::invalid_argument, "data extent must evenly divide logical extent");
            
            variationModulus_[i] = modulus;
          }
          else
          {
            variationModulus_[i] = extents_[i];
          }
          activeDims_[numActiveDims_] = i;
          numActiveDims_++;
        }
        else if (variationType_[i] == BLOCK_PLUS_DIAGONAL)
        {
          underlyingMatchesLogical_ = false;
          blockPlusDiagonalCount++;
          if (blockPlusDiagonalCount == 1) // first dimension thus marked --> active
          {
            
#ifdef HAVE_INTREPID2_DEBUG
            const int numNondiagonalEntries = blockPlusDiagonalNumNondiagonalEntries(blockPlusDiagonalLastNonDiagonal_);
            const int dataExtent = getUnderlyingViewExtent(numActiveDims_); // flat storage of all matrix entries
            const int logicalExtent = extents_[i];
            const int numDiagonalEntries    = logicalExtent - (blockPlusDiagonalLastNonDiagonal_ + 1);
            const int expectedDataExtent = numNondiagonalEntries + numDiagonalEntries;
            INTREPID2_TEST_FOR_EXCEPTION(dataExtent != expectedDataExtent, std::invalid_argument, ("BLOCK_PLUS_DIAGONAL data extent of " + std::to_string(dataExtent) + " does not match expected based on blockPlusDiagonalLastNonDiagonal setting of " + std::to_string(blockPlusDiagonalLastNonDiagonal_)).c_str());
#endif
            
            activeDims_[numActiveDims_] = i;
            numActiveDims_++;
          }
          variationModulus_[i] = getUnderlyingViewExtent(numActiveDims_);
          INTREPID2_TEST_FOR_EXCEPTION(variationType_[i+1] != BLOCK_PLUS_DIAGONAL, std::invalid_argument, "BLOCK_PLUS_DIAGONAL ranks must be contiguous");
          i++; // skip over the next BLOCK_PLUS_DIAGONAL
          variationModulus_[i] = 1; // trivial modulus (should not ever be used)
          INTREPID2_TEST_FOR_EXCEPTION(blockPlusDiagonalCount > 1, std::invalid_argument, "BLOCK_PLUS_DIAGONAL can only apply to two ranks");
        }
        else // CONSTANT
        {
          if (i < rank_)
          {
            underlyingMatchesLogical_ = false;
          }
          variationModulus_[i] = 1; // trivial modulus
        }
      }
      
      if (rank_ != dataRank_)
      {
        underlyingMatchesLogical_ = false;
      }
      
      for (int d=numActiveDims_; d<7; d++)
      {
        // for *inactive* dims, the activeDims_ map just is the identity
        // (this allows getEntry() to work even when the logical rank of the Data object is lower than that of the underlying View.  This can happen for gradients in 1D.)
        activeDims_[d] = d;
      }
      for (int d=0; d<7; d++)
      {
        INTREPID2_TEST_FOR_EXCEPTION(variationModulus_[d] == 0, std::logic_error, "variationModulus should not ever be 0");
      }
    }
 
    public:
    template<class UnaryOperator>
    void applyOperator(UnaryOperator unaryOperator)
    {
      using ExecutionSpace = typename DeviceType::execution_space;
      
      switch (dataRank_)
      {
        case 1:
        {
          const int dataRank = 1;
          auto view = getUnderlyingView<dataRank>();
          
          const int dataExtent = this->getDataExtent(0);
          Kokkos::RangePolicy<ExecutionSpace> policy(ExecutionSpace(),0,dataExtent);
          Kokkos::parallel_for("apply operator in-place", policy,
          KOKKOS_LAMBDA (const int &i0) {
            view(i0) = unaryOperator(view(i0));
          });
          
        }
        break;
        case 2:
        {
          const int dataRank = 2;
          auto policy = dataExtentRangePolicy<dataRank>();
          auto view = getUnderlyingView<dataRank>();
          
          Kokkos::parallel_for("apply operator in-place", policy,
          KOKKOS_LAMBDA (const int &i0, const int &i1) {
            view(i0,i1) = unaryOperator(view(i0,i1));
          });
        }
        break;
        case 3:
        {
          const int dataRank = 3;
          auto policy = dataExtentRangePolicy<dataRank>();
          auto view = getUnderlyingView<dataRank>();
          
          Kokkos::parallel_for("apply operator in-place", policy,
          KOKKOS_LAMBDA (const int &i0, const int &i1, const int &i2) {
            view(i0,i1,i2) = unaryOperator(view(i0,i1,i2));
          });
        }
        break;
        case 4:
        {
          const int dataRank = 4;
          auto policy = dataExtentRangePolicy<dataRank>();
          auto view = getUnderlyingView<dataRank>();
          
          Kokkos::parallel_for("apply operator in-place", policy,
          KOKKOS_LAMBDA (const int &i0, const int &i1, const int &i2, const int &i3) {
            view(i0,i1,i2,i3) = unaryOperator(view(i0,i1,i2,i3));
          });
        }
        break;
        case 5:
        {
          const int dataRank = 5;
          auto policy = dataExtentRangePolicy<dataRank>();
          auto view = getUnderlyingView<dataRank>();
          
          Kokkos::parallel_for("apply operator in-place", policy,
          KOKKOS_LAMBDA (const int &i0, const int &i1, const int &i2, const int &i3, const int &i4) {
            view(i0,i1,i2,i3,i4) = unaryOperator(view(i0,i1,i2,i3,i4));
          });
        }
        break;
        case 6:
        {
          const int dataRank = 6;
          auto policy = dataExtentRangePolicy<dataRank>();
          auto view = getUnderlyingView<dataRank>();
          
          Kokkos::parallel_for("apply operator in-place", policy,
          KOKKOS_LAMBDA (const int &i0, const int &i1, const int &i2, const int &i3, const int &i4, const int &i5) {
            view(i0,i1,i2,i3,i4,i5) = unaryOperator(view(i0,i1,i2,i3,i4,i5));
          });
        }
        break;
        case 7:
        {
          const int dataRank = 7;
          auto policy6 = dataExtentRangePolicy<6>();
          auto view = getUnderlyingView<dataRank>();
          
          const int dim_i6 = view.extent_int(6);
          
          Kokkos::parallel_for("apply operator in-place", policy6,
          KOKKOS_LAMBDA (const int &i0, const int &i1, const int &i2, const int &i3, const int &i4, const int &i5) {
            for (int i6=0; i6<dim_i6; i6++)
            {
              view(i0,i1,i2,i3,i4,i5,i6) = unaryOperator(view(i0,i1,i2,i3,i4,i5,i6));
            }
          });
        }
        break;
        default:
          INTREPID2_TEST_FOR_EXCEPTION(true,std::invalid_argument,"Unsupported data rank");
      }
    }

    template<class ...IntArgs>
    KOKKOS_INLINE_FUNCTION
    reference_type getWritableEntryWithPassThroughOption(const bool &passThroughBlockDiagonalArgs, const IntArgs... intArgs) const
    {
  #ifdef INTREPID2_HAVE_DEBUG
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(numArgs != rank_, std::invalid_argument, "getWritableEntry() should have the same number of arguments as the logical rank.");
  #endif
        constexpr int numArgs = sizeof...(intArgs);
        if (underlyingMatchesLogical_)
        {
          // in this case, we require that numArgs == dataRank_
          return getUnderlyingView<numArgs>()(intArgs...);
        }
        
        // extract the type of the first argument; use that for the arrays below
        using int_type = std::tuple_element_t<0, std::tuple<IntArgs...>>;
        
        const Kokkos::Array<int_type, numArgs+1> args {intArgs...,0}; // we pad with one extra entry (0) to avoid gcc compiler warnings about references beyond the bounds of the array (the [d+1]'s below)
        Kokkos::Array<int_type, 7> refEntry;
        for (int d=0; d<numArgs; d++)
        {
          switch (variationType_[d])
          {
            case CONSTANT: refEntry[d] = 0;                              break;
            case GENERAL:  refEntry[d] = args[d];                        break;
            case MODULAR:  refEntry[d] = args[d] % variationModulus_[d]; break;
            case BLOCK_PLUS_DIAGONAL:
            {
              if (passThroughBlockDiagonalArgs)
              {
                refEntry[d]   = args[d];
                refEntry[d+1] = args[d+1]; // this had better be == 0
                INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(args[d+1] != 0, std::invalid_argument, "getWritableEntry() called with passThroughBlockDiagonalArgs = true, but nonzero second matrix argument.");
              }
              else
              {
                const int numNondiagonalEntries = blockPlusDiagonalNumNondiagonalEntries(blockPlusDiagonalLastNonDiagonal_);
                
                const int_type &i = args[d];
                if (d+1 >= numArgs)
                {
                  INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(true, std::invalid_argument, "BLOCK_PLUS_DIAGONAL must be present for two dimensions; here, encountered only one");
                }
                else
                {
                  const int_type &j = args[d+1];
                  
                  if ((i > static_cast<int_type>(blockPlusDiagonalLastNonDiagonal_)) || (j > static_cast<int_type>(blockPlusDiagonalLastNonDiagonal_)))
                  {
                    if (i != j)
                    {
                      // off diagonal: zero
                      return zeroView_(0); // NOTE: this branches in an argument-dependent way; this is not great for CUDA performance.  When using BLOCK_PLUS_DIAGONAL, should generally avoid calls to this getEntry() method.  (Use methods that directly take advantage of the data packing instead.)
                    }
                    else
                    {
                      refEntry[d] = blockPlusDiagonalDiagonalEntryIndex(blockPlusDiagonalLastNonDiagonal_, numNondiagonalEntries, i);
                    }
                  }
                  else
                  {
                    refEntry[d] = blockPlusDiagonalBlockEntryIndex(blockPlusDiagonalLastNonDiagonal_, numNondiagonalEntries, i, j);
                  }
 
                  // skip next d (this is required also to be BLOCK_PLUS_DIAGONAL, and we've consumed its arg as j above)
                  refEntry[d+1] = 0;
                }
              }
              d++;
            }
          }
        }
         // refEntry should be zero-filled beyond numArgs, for cases when rank_ < dataRank_ (this only is allowed if the extra dimensions each has extent 1).
        for (int d=numArgs; d<7; d++)
        {
          refEntry[d] = 0;
        }
        
        if (dataRank_ == 1)
        {
          return data1_(refEntry[activeDims_[0]]);
        }
        else if (dataRank_ == 2)
        {
          return data2_(refEntry[activeDims_[0]],refEntry[activeDims_[1]]);
        }
        else if (dataRank_ == 3)
        {
          return data3_(refEntry[activeDims_[0]],refEntry[activeDims_[1]],refEntry[activeDims_[2]]);
        }
        else if (dataRank_ == 4)
        {
          return data4_(refEntry[activeDims_[0]],refEntry[activeDims_[1]],refEntry[activeDims_[2]],refEntry[activeDims_[3]]);
        }
        else if (dataRank_ == 5)
        {
          return data5_(refEntry[activeDims_[0]],refEntry[activeDims_[1]],refEntry[activeDims_[2]],refEntry[activeDims_[3]],
                        refEntry[activeDims_[4]]);
        }
        else if (dataRank_ == 6)
        {
          return data6_(refEntry[activeDims_[0]],refEntry[activeDims_[1]],refEntry[activeDims_[2]],refEntry[activeDims_[3]],
                        refEntry[activeDims_[4]],refEntry[activeDims_[5]]);
        }
        else // dataRank_ == 7
        {
          return data7_(refEntry[activeDims_[0]],refEntry[activeDims_[1]],refEntry[activeDims_[2]],refEntry[activeDims_[3]],
                        refEntry[activeDims_[4]],refEntry[activeDims_[5]],refEntry[activeDims_[6]]);
        }
      
    }
    
    template<class ...IntArgs>
    KOKKOS_INLINE_FUNCTION
    reference_type getWritableEntry(const IntArgs... intArgs) const
    {
      return getWritableEntryWithPassThroughOption(false, intArgs...);
    }
  public:
    template<class ToContainer, class FromContainer>
    static void copyContainer(ToContainer to, FromContainer from)
    {
//      std::cout << "Entered copyContainer().\n";
      auto policy = Kokkos::MDRangePolicy<execution_space,Kokkos::Rank<6>>({0,0,0,0,0,0},{from.extent_int(0),from.extent_int(1),from.extent_int(2), from.extent_int(3), from.extent_int(4), from.extent_int(5)});
      
      Kokkos::parallel_for("copyContainer", policy,
      KOKKOS_LAMBDA (const int &i0, const int &i1, const int &i2, const int &i3, const int &i4, const int &i5) {
        for (int i6=0; i6<from.extent_int(6); i6++)
        {
          to.access(i0,i1,i2,i3,i4,i5,i6) = from.access(i0,i1,i2,i3,i4,i5,i6);
        }
      });
    }
    
    void allocateAndCopyFromDynRankView(ScalarView<DataScalar,DeviceType> data)
    {
//      std::cout << "Entered allocateAndCopyFromDynRankView().\n";
      switch (dataRank_)
      {
        case 1: data1_ = Kokkos::View<DataScalar*,       DeviceType>("Intrepid2 Data", data.extent_int(0)); break;
        case 2: data2_ = Kokkos::View<DataScalar**,      DeviceType>("Intrepid2 Data", data.extent_int(0), data.extent_int(1)); break;
        case 3: data3_ = Kokkos::View<DataScalar***,     DeviceType>("Intrepid2 Data", data.extent_int(0), data.extent_int(1), data.extent_int(2)); break;
        case 4: data4_ = Kokkos::View<DataScalar****,    DeviceType>("Intrepid2 Data", data.extent_int(0), data.extent_int(1), data.extent_int(2), data.extent_int(3)); break;
        case 5: data5_ = Kokkos::View<DataScalar*****,   DeviceType>("Intrepid2 Data", data.extent_int(0), data.extent_int(1), data.extent_int(2), data.extent_int(3), data.extent_int(4)); break;
        case 6: data6_ = Kokkos::View<DataScalar******,  DeviceType>("Intrepid2 Data", data.extent_int(0), data.extent_int(1), data.extent_int(2), data.extent_int(3), data.extent_int(4), data.extent_int(5)); break;
        case 7: data7_ = Kokkos::View<DataScalar*******, DeviceType>("Intrepid2 Data", data.extent_int(0), data.extent_int(1), data.extent_int(2), data.extent_int(3), data.extent_int(4), data.extent_int(5), data.extent_int(6)); break;
        default: INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Invalid data rank");
      }
      
      switch (dataRank_)
      {
        case 1: copyContainer(data1_,data); break;
        case 2: copyContainer(data2_,data); break;
        case 3: copyContainer(data3_,data); break;
        case 4: copyContainer(data4_,data); break;
        case 5: copyContainer(data5_,data); break;
        case 6: copyContainer(data6_,data); break;
        case 7: copyContainer(data7_,data); break;
        default: INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Invalid data rank");
      }
    }
    
    Data(std::vector<DimensionInfo> dimInfoVector)
    :
    // initialize member variables as if default constructor; if dimInfoVector is empty, we want default constructor behavior.
    dataRank_(0), extents_({0,0,0,0,0,0,0}), variationType_({CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT}), blockPlusDiagonalLastNonDiagonal_(-1), rank_(dimInfoVector.size())
    {
      // If dimInfoVector is empty, the member initialization above is correct; otherwise, we set as below.
      // Either way, once members are initialized, we must call setActiveDims().
      if (dimInfoVector.size() != 0)
      {
        std::vector<int> dataExtents;
 
        bool blockPlusDiagonalEncountered = false;
        for (int d=0; d<rank_; d++)
        {
          const DimensionInfo & dimInfo = dimInfoVector[d];
          extents_[d] = dimInfo.logicalExtent;
          variationType_[d] = dimInfo.variationType;
          const bool isBlockPlusDiagonal = (variationType_[d] == BLOCK_PLUS_DIAGONAL);
          const bool isSecondBlockPlusDiagonal = isBlockPlusDiagonal && blockPlusDiagonalEncountered;
          if (isBlockPlusDiagonal)
          {
            blockPlusDiagonalEncountered = true;
            blockPlusDiagonalLastNonDiagonal_ = dimInfo.blockPlusDiagonalLastNonDiagonal;
          }
          if ((variationType_[d] != CONSTANT) && (!isSecondBlockPlusDiagonal))
          {
            dataExtents.push_back(dimInfo.dataExtent);
          }
        }
        if (dataExtents.size() == 0)
        {
          // constant data
          dataExtents.push_back(1);
        }
        dataRank_ = dataExtents.size();
        switch (dataRank_)
        {
          case 1: data1_ = Kokkos::View<DataScalar*,       DeviceType>("Intrepid2 Data", dataExtents[0]); break;
          case 2: data2_ = Kokkos::View<DataScalar**,      DeviceType>("Intrepid2 Data", dataExtents[0], dataExtents[1]); break;
          case 3: data3_ = Kokkos::View<DataScalar***,     DeviceType>("Intrepid2 Data", dataExtents[0], dataExtents[1], dataExtents[2]); break;
          case 4: data4_ = Kokkos::View<DataScalar****,    DeviceType>("Intrepid2 Data", dataExtents[0], dataExtents[1], dataExtents[2], dataExtents[3]); break;
          case 5: data5_ = Kokkos::View<DataScalar*****,   DeviceType>("Intrepid2 Data", dataExtents[0], dataExtents[1], dataExtents[2], dataExtents[3], dataExtents[4]); break;
          case 6: data6_ = Kokkos::View<DataScalar******,  DeviceType>("Intrepid2 Data", dataExtents[0], dataExtents[1], dataExtents[2], dataExtents[3], dataExtents[4], dataExtents[5]); break;
          case 7: data7_ = Kokkos::View<DataScalar*******, DeviceType>("Intrepid2 Data", dataExtents[0], dataExtents[1], dataExtents[2], dataExtents[3], dataExtents[4], dataExtents[5], dataExtents[6]); break;
          default: INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Invalid data rank");
        }
      }
      setActiveDims();
    }
    
    Data(const ScalarView<DataScalar,DeviceType> &data, int rank, Kokkos::Array<int,7> extents, Kokkos::Array<DataVariationType,7> variationType, const int blockPlusDiagonalLastNonDiagonal = -1)
    :
    dataRank_(data.rank()), extents_(extents), variationType_(variationType), blockPlusDiagonalLastNonDiagonal_(blockPlusDiagonalLastNonDiagonal), rank_(rank)
    {
      allocateAndCopyFromDynRankView(data);
      setActiveDims();
    }
    
    template<typename OtherDeviceType, class = typename std::enable_if< std::is_same<typename DeviceType::memory_space, typename OtherDeviceType::memory_space>::value>::type,
                                       class = typename std::enable_if<!std::is_same<DeviceType,OtherDeviceType>::value>::type>
    Data(const Data<DataScalar,OtherDeviceType> &data)
    :
    dataRank_(data.getUnderlyingViewRank()), extents_(data.getExtents()), variationType_(data.getVariationTypes()), blockPlusDiagonalLastNonDiagonal_(data.blockPlusDiagonalLastNonDiagonal()), rank_(data.rank())
    {
//      std::cout << "Entered copy-like Data constructor.\n";
      if (dataRank_ != 0) // dataRank_ == 0 indicates an invalid Data object (a placeholder, can indicate zero value)
      {
        const auto view = data.getUnderlyingView();
        switch (dataRank_)
        {
          case 1: data1_ = data.getUnderlyingView1(); break;
          case 2: data2_ = data.getUnderlyingView2(); break;
          case 3: data3_ = data.getUnderlyingView3(); break;
          case 4: data4_ = data.getUnderlyingView4(); break;
          case 5: data5_ = data.getUnderlyingView5(); break;
          case 6: data6_ = data.getUnderlyingView6(); break;
          case 7: data7_ = data.getUnderlyingView7(); break;
          default: INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Invalid data rank");
        }
      }
      setActiveDims();
    }
    
    template<typename OtherDeviceType, class = typename std::enable_if<!std::is_same<typename DeviceType::memory_space, typename OtherDeviceType::memory_space>::value>::type>
    Data(const Data<DataScalar,OtherDeviceType> &data)
    :
    dataRank_(data.getUnderlyingViewRank()), extents_(data.getExtents()), variationType_(data.getVariationTypes()), blockPlusDiagonalLastNonDiagonal_(data.blockPlusDiagonalLastNonDiagonal()), rank_(data.rank())
    {
//      std::cout << "Entered copy-like Data constructor.\n";
      if (dataRank_ != 0) // dataRank_ == 0 indicates an invalid Data object (a placeholder, can indicate zero value)
      {
        const auto view = data.getUnderlyingView();
        switch (dataRank_)
        {
          case 1: data1_ = Kokkos::View<DataScalar*,       DeviceType>("Intrepid2 Data", view.extent_int(0)); break;
          case 2: data2_ = Kokkos::View<DataScalar**,      DeviceType>("Intrepid2 Data", view.extent_int(0), view.extent_int(1)); break;
          case 3: data3_ = Kokkos::View<DataScalar***,     DeviceType>("Intrepid2 Data", view.extent_int(0), view.extent_int(1), view.extent_int(2)); break;
          case 4: data4_ = Kokkos::View<DataScalar****,    DeviceType>("Intrepid2 Data", view.extent_int(0), view.extent_int(1), view.extent_int(2), view.extent_int(3)); break;
          case 5: data5_ = Kokkos::View<DataScalar*****,   DeviceType>("Intrepid2 Data", view.extent_int(0), view.extent_int(1), view.extent_int(2), view.extent_int(3), view.extent_int(4)); break;
          case 6: data6_ = Kokkos::View<DataScalar******,  DeviceType>("Intrepid2 Data", view.extent_int(0), view.extent_int(1), view.extent_int(2), view.extent_int(3), view.extent_int(4), view.extent_int(5)); break;
          case 7: data7_ = Kokkos::View<DataScalar*******, DeviceType>("Intrepid2 Data", view.extent_int(0), view.extent_int(1), view.extent_int(2), view.extent_int(3), view.extent_int(4), view.extent_int(5), view.extent_int(6)); break;
          default: INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Invalid data rank");
        }
        
        // copy
        // (Note: Kokkos::deep_copy() will not generally work if the layouts are different; that's why we do a manual copy here once we have the data on the host):
        // first, mirror and copy dataView; then copy to the appropriate data_ member
        using MemorySpace = typename DeviceType::memory_space;
        switch (dataRank_)
        {
          case 1: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView1()); copyContainer(data1_, dataViewMirror);} break;
          case 2: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView2()); copyContainer(data2_, dataViewMirror);} break;
          case 3: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView3()); copyContainer(data3_, dataViewMirror);} break;
          case 4: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView4()); copyContainer(data4_, dataViewMirror);} break;
          case 5: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView5()); copyContainer(data5_, dataViewMirror);} break;
          case 6: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView6()); copyContainer(data6_, dataViewMirror);} break;
          case 7: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView7()); copyContainer(data7_, dataViewMirror);} break;
          default: INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Invalid data rank");
        }
      }
      setActiveDims();
    }
    
//    Data(const Data<DataScalar,ExecSpaceType> &data)
//    :
//    dataRank_(data.getUnderlyingViewRank()), extents_(data.getExtents()), variationType_(data.getVariationTypes()), blockPlusDiagonalLastNonDiagonal_(data.blockPlusDiagonalLastNonDiagonal()), rank_(data.rank())
//    {
//      std::cout << "Entered Data copy constructor.\n";
//      if (dataRank_ != 0) // dataRank_ == 0 indicates an invalid Data object (a placeholder, can indicate zero value)
//      {
//        const auto view = data.getUnderlyingView();
//        switch (dataRank_)
//        {
//          case 1: data1_ = Kokkos::View<DataScalar*,       DeviceType>("Intrepid2 Data - explicit copy constructor(for debugging)", view.extent_int(0)); break;
//          case 2: data2_ = Kokkos::View<DataScalar**,      DeviceType>("Intrepid2 Data - explicit copy constructor(for debugging)", view.extent_int(0), view.extent_int(1)); break;
//          case 3: data3_ = Kokkos::View<DataScalar***,     DeviceType>("Intrepid2 Data - explicit copy constructor(for debugging)", view.extent_int(0), view.extent_int(1), view.extent_int(2)); break;
//          case 4: data4_ = Kokkos::View<DataScalar****,    DeviceType>("Intrepid2 Data - explicit copy constructor(for debugging)", view.extent_int(0), view.extent_int(1), view.extent_int(2), view.extent_int(3)); break;
//          case 5: data5_ = Kokkos::View<DataScalar*****,   DeviceType>("Intrepid2 Data - explicit copy constructor(for debugging)", view.extent_int(0), view.extent_int(1), view.extent_int(2), view.extent_int(3), view.extent_int(4)); break;
//          case 6: data6_ = Kokkos::View<DataScalar******,  DeviceType>("Intrepid2 Data - explicit copy constructor(for debugging)", view.extent_int(0), view.extent_int(1), view.extent_int(2), view.extent_int(3), view.extent_int(4), view.extent_int(5)); break;
//          case 7: data7_ = Kokkos::View<DataScalar*******, DeviceType>("Intrepid2 Data - explicit copy constructor(for debugging)", view.extent_int(0), view.extent_int(1), view.extent_int(2), view.extent_int(3), view.extent_int(4), view.extent_int(5), view.extent_int(6)); break;
//          default: INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Invalid data rank");
//        }
//
//        // copy
//        // (Note: Kokkos::deep_copy() will not generally work if the layouts are different; that's why we do a manual copy here once we have the data on the host):
//        // first, mirror and copy dataView; then copy to the appropriate data_ member
//        using MemorySpace = typename DeviceType::memory_space;
//        switch (dataRank_)
//        {
//          case 1: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView1()); copyContainer(data1_, dataViewMirror);} break;
//          case 2: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView2()); copyContainer(data2_, dataViewMirror);} break;
//          case 3: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView3()); copyContainer(data3_, dataViewMirror);} break;
//          case 4: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView4()); copyContainer(data4_, dataViewMirror);} break;
//          case 5: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView5()); copyContainer(data5_, dataViewMirror);} break;
//          case 6: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView6()); copyContainer(data6_, dataViewMirror);} break;
//          case 7: {auto dataViewMirror = Kokkos::create_mirror_view_and_copy(MemorySpace(), data.getUnderlyingView7()); copyContainer(data7_, dataViewMirror);} break;
//          default: INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Invalid data rank");
//        }
//      }
//
//      setActiveDims();
//    }
    
    Data(ScalarView<DataScalar,DeviceType> data)
    :
    Data(data,
         data.rank(),
         Kokkos::Array<int,7> {data.extent_int(0),data.extent_int(1),data.extent_int(2),data.extent_int(3),data.extent_int(4),data.extent_int(5),data.extent_int(6)},
         Kokkos::Array<DataVariationType,7> {GENERAL,GENERAL,GENERAL,GENERAL,GENERAL,GENERAL,GENERAL}, -1)
    {}
    
    template<size_t rank, class ...DynRankViewProperties>
    Data(const Kokkos::DynRankView<DataScalar,DeviceType, DynRankViewProperties...> &data, Kokkos::Array<int,rank> extents, Kokkos::Array<DataVariationType,rank> variationType, const int blockPlusDiagonalLastNonDiagonal = -1)
    :
    dataRank_(data.rank()), extents_({1,1,1,1,1,1,1}), variationType_({CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT}), blockPlusDiagonalLastNonDiagonal_(blockPlusDiagonalLastNonDiagonal), rank_(rank)
    {
//      std::cout << "Entered a DynRankView Data() constructor.\n";
      allocateAndCopyFromDynRankView(data);
      for (unsigned d=0; d<rank; d++)
      {
        extents_[d]       = extents[d];
        variationType_[d] = variationType[d];
      }
      setActiveDims();
    }
        
    template<size_t rank, class ...ViewProperties>
    Data(Kokkos::View<DataScalar*,DeviceType, ViewProperties...> data, Kokkos::Array<int,rank> extents, Kokkos::Array<DataVariationType,rank> variationType, const int blockPlusDiagonalLastNonDiagonal = -1)
    :
    dataRank_(data.rank), extents_({1,1,1,1,1,1,1}), variationType_({CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT}), blockPlusDiagonalLastNonDiagonal_(blockPlusDiagonalLastNonDiagonal), rank_(rank)
    {
      data1_ = data;
      for (unsigned d=0; d<rank; d++)
      {
        extents_[d]       = extents[d];
        variationType_[d] = variationType[d];
      }
      setActiveDims();
    }
    
    template<size_t rank, class ...ViewProperties>
    Data(Kokkos::View<DataScalar**,DeviceType, ViewProperties...> data, Kokkos::Array<int,rank> extents, Kokkos::Array<DataVariationType,rank> variationType, const int blockPlusDiagonalLastNonDiagonal = -1)
    :
    dataRank_(data.rank), extents_({1,1,1,1,1,1,1}), variationType_({CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT}), blockPlusDiagonalLastNonDiagonal_(blockPlusDiagonalLastNonDiagonal), rank_(rank)
    {
      data2_ = data;
      for (unsigned d=0; d<rank; d++)
      {
        extents_[d]       = extents[d];
        variationType_[d] = variationType[d];
      }
      setActiveDims();
    }
    
    template<size_t rank, class ...ViewProperties>
    Data(Kokkos::View<DataScalar***,DeviceType, ViewProperties...> data, Kokkos::Array<int,rank> extents, Kokkos::Array<DataVariationType,rank> variationType, const int blockPlusDiagonalLastNonDiagonal = -1)
    :
    dataRank_(data.rank), extents_({1,1,1,1,1,1,1}), variationType_({CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT}), blockPlusDiagonalLastNonDiagonal_(blockPlusDiagonalLastNonDiagonal), rank_(rank)
    {
      data3_ = data;
      for (unsigned d=0; d<rank; d++)
      {
        extents_[d]       = extents[d];
        variationType_[d] = variationType[d];
      }
      setActiveDims();
    }
    
    template<size_t rank, class ...ViewProperties>
    Data(Kokkos::View<DataScalar****,DeviceType, ViewProperties...> data, Kokkos::Array<int,rank> extents, Kokkos::Array<DataVariationType,rank> variationType, const int blockPlusDiagonalLastNonDiagonal = -1)
    :
    dataRank_(data.rank), extents_({1,1,1,1,1,1,1}), variationType_({CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT}), blockPlusDiagonalLastNonDiagonal_(blockPlusDiagonalLastNonDiagonal), rank_(rank)
    {
      data4_ = data;
      for (unsigned d=0; d<rank; d++)
      {
        extents_[d]       = extents[d];
        variationType_[d] = variationType[d];
      }
      setActiveDims();
    }
    
    template<size_t rank, class ...ViewProperties>
    Data(Kokkos::View<DataScalar*****,DeviceType, ViewProperties...> data, Kokkos::Array<int,rank> extents, Kokkos::Array<DataVariationType,rank> variationType, const int blockPlusDiagonalLastNonDiagonal = -1)
    :
    dataRank_(data.rank), extents_({1,1,1,1,1,1,1}), variationType_({CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT}), blockPlusDiagonalLastNonDiagonal_(blockPlusDiagonalLastNonDiagonal), rank_(rank)
    {
      data5_ = data;
      for (unsigned d=0; d<rank; d++)
      {
        extents_[d]       = extents[d];
        variationType_[d] = variationType[d];
      }
      setActiveDims();
    }
    
    template<size_t rank, class ...ViewProperties>
    Data(Kokkos::View<DataScalar******,DeviceType, ViewProperties...> data, Kokkos::Array<int,rank> extents, Kokkos::Array<DataVariationType,rank> variationType, const int blockPlusDiagonalLastNonDiagonal = -1)
    :
    dataRank_(data.rank), extents_({1,1,1,1,1,1,1}), variationType_({CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT}), blockPlusDiagonalLastNonDiagonal_(blockPlusDiagonalLastNonDiagonal), rank_(rank)
    {
      data6_ = data;
      for (unsigned d=0; d<rank; d++)
      {
        extents_[d]       = extents[d];
        variationType_[d] = variationType[d];
      }
      setActiveDims();
    }
    
    template<size_t rank, class ...ViewProperties>
    Data(Kokkos::View<DataScalar*******,DeviceType, ViewProperties...> data, Kokkos::Array<int,rank> extents, Kokkos::Array<DataVariationType,rank> variationType, const int blockPlusDiagonalLastNonDiagonal = -1)
    :
    dataRank_(data.rank), extents_({1,1,1,1,1,1,1}), variationType_({CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT}), blockPlusDiagonalLastNonDiagonal_(blockPlusDiagonalLastNonDiagonal), rank_(rank)
    {
      data7_ = data;
      for (unsigned d=0; d<rank; d++)
      {
        extents_[d]       = extents[d];
        variationType_[d] = variationType[d];
      }
      setActiveDims();
    }
    
    template<class ViewScalar, class ...ViewProperties>
    Data(const unsigned rank, Kokkos::View<ViewScalar,DeviceType, ViewProperties...> data, Kokkos::Array<int,7> extents, Kokkos::Array<DataVariationType,7> variationType, const int blockPlusDiagonalLastNonDiagonal = -1)
    :
    dataRank_(data.rank), extents_({1,1,1,1,1,1,1}), variationType_({CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT}), blockPlusDiagonalLastNonDiagonal_(blockPlusDiagonalLastNonDiagonal), rank_(rank)
    {
      setUnderlyingView<data.rank>(data);
      for (unsigned d=0; d<rank; d++)
      {
        extents_[d]       = extents[d];
        variationType_[d] = variationType[d];
      }
      setActiveDims();
    }
    
    template<size_t rank>
    Data(DataScalar constantValue, Kokkos::Array<int,rank> extents)
    :
    dataRank_(1), extents_({1,1,1,1,1,1,1}), variationType_({CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT}), blockPlusDiagonalLastNonDiagonal_(-1), rank_(rank)
    {
      data1_ = Kokkos::View<DataScalar*,DeviceType>("Constant Data",1);
      Kokkos::deep_copy(data1_, constantValue);
      for (unsigned d=0; d<rank; d++)
      {
        extents_[d] = extents[d];
      }
      setActiveDims();
    }
            
    Data()
    :
    dataRank_(0), extents_({0,0,0,0,0,0,0}), variationType_({CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT,CONSTANT}), blockPlusDiagonalLastNonDiagonal_(-1), rank_(0)
    {
      setActiveDims();
    }
    
    KOKKOS_INLINE_FUNCTION
    const int & blockPlusDiagonalLastNonDiagonal() const
    {
      return blockPlusDiagonalLastNonDiagonal_;
    }
    
    KOKKOS_INLINE_FUNCTION
    Kokkos::Array<int,7> getExtents() const
    {
      return extents_;
    }
    
    KOKKOS_INLINE_FUNCTION
    DimensionInfo getDimensionInfo(const int &dim) const
    {
      DimensionInfo dimInfo;
      
      dimInfo.logicalExtent = extent_int(dim);
      dimInfo.variationType = variationType_[dim];
      dimInfo.dataExtent    = getDataExtent(dim);
      dimInfo.variationModulus = variationModulus_[dim];
      
      if (dimInfo.variationType == BLOCK_PLUS_DIAGONAL)
      {
        dimInfo.blockPlusDiagonalLastNonDiagonal = blockPlusDiagonalLastNonDiagonal_;
      }
      return dimInfo;
    }
    
    KOKKOS_INLINE_FUNCTION
    DimensionInfo combinedDataDimensionInfo(const Data &otherData, const int &dim) const
    {
      const DimensionInfo myDimInfo    = getDimensionInfo(dim);
      const DimensionInfo otherDimInfo = otherData.getDimensionInfo(dim);
      
      return combinedDimensionInfo(myDimInfo, otherDimInfo);
    }
    
    template<int rank>
    KOKKOS_INLINE_FUNCTION
    enable_if_t<rank==1, const Kokkos::View<typename RankExpander<DataScalar, rank>::value_type, DeviceType> &>
    getUnderlyingView() const
    {
      #ifdef HAVE_INTREPID2_DEBUG
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(dataRank_ != rank, std::invalid_argument, "getUnderlyingView() called for rank that does not match dataRank_");
      #endif
      return data1_;
    }
    
    template<int rank>
    KOKKOS_INLINE_FUNCTION
    enable_if_t<rank==2, const Kokkos::View<typename RankExpander<DataScalar, rank>::value_type, DeviceType> &>
    getUnderlyingView() const
    {
      #ifdef HAVE_INTREPID2_DEBUG
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(dataRank_ != rank, std::invalid_argument, "getUnderlyingView() called for rank that does not match dataRank_");
      #endif
      return data2_;
    }
    
    template<int rank>
    KOKKOS_INLINE_FUNCTION
    enable_if_t<rank==3, const Kokkos::View<typename RankExpander<DataScalar, rank>::value_type, DeviceType> &>
    getUnderlyingView() const
    {
      #ifdef HAVE_INTREPID2_DEBUG
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(dataRank_ != rank, std::invalid_argument, "getUnderlyingView() called for rank that does not match dataRank_");
      #endif
      return data3_;
    }
    
    template<int rank>
    KOKKOS_INLINE_FUNCTION
    enable_if_t<rank==4, const Kokkos::View<typename RankExpander<DataScalar, rank>::value_type, DeviceType> &>
    getUnderlyingView() const
    {
      #ifdef HAVE_INTREPID2_DEBUG
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(dataRank_ != rank, std::invalid_argument, "getUnderlyingView() called for rank that does not match dataRank_");
      #endif
      return data4_;
    }
    
    template<int rank>
    KOKKOS_INLINE_FUNCTION
    enable_if_t<rank==5, const Kokkos::View<typename RankExpander<DataScalar, rank>::value_type, DeviceType> &>
    getUnderlyingView() const
    {
      #ifdef HAVE_INTREPID2_DEBUG
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(dataRank_ != rank, std::invalid_argument, "getUnderlyingView() called for rank that does not match dataRank_");
      #endif
      return data5_;
    }
    
    template<int rank>
    KOKKOS_INLINE_FUNCTION
    enable_if_t<rank==6, const Kokkos::View<typename RankExpander<DataScalar, rank>::value_type, DeviceType> &>
    getUnderlyingView() const
    {
      #ifdef HAVE_INTREPID2_DEBUG
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(dataRank_ != rank, std::invalid_argument, "getUnderlyingView() called for rank that does not match dataRank_");
      #endif
      return data6_;
    }
    
    template<int rank>
    KOKKOS_INLINE_FUNCTION
    enable_if_t<rank==7, const Kokkos::View<typename RankExpander<DataScalar, rank>::value_type, DeviceType> &>
    getUnderlyingView() const
    {
      #ifdef HAVE_INTREPID2_DEBUG
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(dataRank_ != rank, std::invalid_argument, "getUnderlyingView() called for rank that does not match dataRank_");
      #endif
      return data7_;
    }
    
    KOKKOS_INLINE_FUNCTION
    const Kokkos::View<DataScalar*, DeviceType> & getUnderlyingView1() const
    {
      return getUnderlyingView<1>();
    }
    
    KOKKOS_INLINE_FUNCTION
    const Kokkos::View<DataScalar**, DeviceType> & getUnderlyingView2() const
    {
      return getUnderlyingView<2>();
    }
    
    KOKKOS_INLINE_FUNCTION
    const Kokkos::View<DataScalar***, DeviceType> & getUnderlyingView3() const
    {
      return getUnderlyingView<3>();
    }
    
    KOKKOS_INLINE_FUNCTION
    const Kokkos::View<DataScalar****, DeviceType> & getUnderlyingView4() const
    {
      return getUnderlyingView<4>();
    }
    
    KOKKOS_INLINE_FUNCTION
    const Kokkos::View<DataScalar*****, DeviceType> & getUnderlyingView5() const
    {
      return getUnderlyingView<5>();
    }
    
    KOKKOS_INLINE_FUNCTION
    const Kokkos::View<DataScalar******, DeviceType> & getUnderlyingView6() const
    {
      return getUnderlyingView<6>();
    }
    
    KOKKOS_INLINE_FUNCTION
    const Kokkos::View<DataScalar*******, DeviceType> & getUnderlyingView7() const
    {
      return getUnderlyingView<7>();
    }
    
    KOKKOS_INLINE_FUNCTION
    void setUnderlyingView1(const Kokkos::View<DataScalar*, DeviceType> & view)
    {
      data1_ = view;
    }
    
    KOKKOS_INLINE_FUNCTION
    void setUnderlyingView2(const Kokkos::View<DataScalar**, DeviceType> & view)
    {
      data2_ = view;
    }
    
    KOKKOS_INLINE_FUNCTION
    void setUnderlyingView3(const Kokkos::View<DataScalar***, DeviceType> & view)
    {
      data3_ = view;
    }
    
    KOKKOS_INLINE_FUNCTION
    void setUnderlyingView4(const Kokkos::View<DataScalar****, DeviceType> & view)
    {
      data4_ = view;
    }
    
    KOKKOS_INLINE_FUNCTION
    void setUnderlyingView5(const Kokkos::View<DataScalar*****, DeviceType> & view)
    {
      data5_ = view;
    }
    
    KOKKOS_INLINE_FUNCTION
    void setUnderlyingView6(const Kokkos::View<DataScalar******, DeviceType> & view)
    {
      data6_ = view;
    }
    
    KOKKOS_INLINE_FUNCTION
    void setUnderlyingView7(const Kokkos::View<DataScalar*******, DeviceType> & view)
    {
      data7_ = view;
    }
    
    template<int underlyingRank, class ViewScalar>
    KOKKOS_INLINE_FUNCTION
    void setUnderlyingView(const Kokkos::View<ViewScalar, DeviceType> & view)
    {
      if constexpr (underlyingRank == 1)
      {
        setUnderlyingView1(view);
      }
      else if constexpr (underlyingRank == 2)
      {
        setUnderlyingView2(view);
      }
      else if constexpr (underlyingRank == 3)
      {
        setUnderlyingView3(view);
      }
      else if constexpr (underlyingRank == 4)
      {
        setUnderlyingView4(view);
      }
      else if constexpr (underlyingRank == 5)
      {
        setUnderlyingView5(view);
      }
      else if constexpr (underlyingRank == 6)
      {
        setUnderlyingView6(view);
      }
      else if constexpr (underlyingRank == 7)
      {
        setUnderlyingView7(view);
      }
      else
      {
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(true, std::invalid_argument, "implementation for specialization missing");
      }
    }
    
    ScalarView<DataScalar,DeviceType> getUnderlyingView() const
    {
      switch (dataRank_)
      {
        case 1: return data1_;
        case 2: return data2_;
        case 3: return data3_;
        case 4: return data4_;
        case 5: return data5_;
        case 6: return data6_;
        case 7: return data7_;
        default: INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Invalid data rank");
      }
    }
    
    KOKKOS_INLINE_FUNCTION
    ordinal_type getUnderlyingViewRank() const
    {
      return dataRank_;
    }
    
    KOKKOS_INLINE_FUNCTION
    ordinal_type getUnderlyingViewSize() const
    {
      ordinal_type size = 1;
      for (ordinal_type r=0; r<dataRank_; r++)
      {
        size *= getUnderlyingViewExtent(r);
      }
      return size;
    }
    
    ScalarView<DataScalar,DeviceType> allocateDynRankViewMatchingUnderlying() const
    {
      switch (dataRank_)
      {
        case 1: return getMatchingViewWithLabel(data1_, "Intrepid2 Data", data1_.extent_int(0));
        case 2: return getMatchingViewWithLabel(data2_, "Intrepid2 Data", data2_.extent_int(0), data2_.extent_int(1));
        case 3: return getMatchingViewWithLabel(data3_, "Intrepid2 Data", data3_.extent_int(0), data3_.extent_int(1), data3_.extent_int(2));
        case 4: return getMatchingViewWithLabel(data4_, "Intrepid2 Data", data4_.extent_int(0), data4_.extent_int(1), data4_.extent_int(2), data4_.extent_int(3));
        case 5: return getMatchingViewWithLabel(data5_, "Intrepid2 Data", data5_.extent_int(0), data5_.extent_int(1), data5_.extent_int(2), data5_.extent_int(3), data5_.extent_int(4));
        case 6: return getMatchingViewWithLabel(data6_, "Intrepid2 Data", data6_.extent_int(0), data6_.extent_int(1), data6_.extent_int(2), data6_.extent_int(3), data6_.extent_int(4), data6_.extent_int(5));
        case 7: return getMatchingViewWithLabel(data7_, "Intrepid2 Data", data7_.extent_int(0), data7_.extent_int(1), data7_.extent_int(2), data7_.extent_int(3), data7_.extent_int(4), data7_.extent_int(5), data7_.extent_int(6));
        default: INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Invalid data rank");
      }
    }
    
    template<class ... DimArgs>
    ScalarView<DataScalar,DeviceType> allocateDynRankViewMatchingUnderlying(DimArgs... dims) const
    {
      switch (dataRank_)
      {
        case 1: return getMatchingViewWithLabel(data1_, "Intrepid2 Data", dims...);
        case 2: return getMatchingViewWithLabel(data2_, "Intrepid2 Data", dims...);
        case 3: return getMatchingViewWithLabel(data3_, "Intrepid2 Data", dims...);
        case 4: return getMatchingViewWithLabel(data4_, "Intrepid2 Data", dims...);
        case 5: return getMatchingViewWithLabel(data5_, "Intrepid2 Data", dims...);
        case 6: return getMatchingViewWithLabel(data6_, "Intrepid2 Data", dims...);
        case 7: return getMatchingViewWithLabel(data7_, "Intrepid2 Data", dims...);
        default: INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Invalid data rank");
      }
    }
  
    void clear() const
    {
      switch (dataRank_)
      {
        case 1: Kokkos::deep_copy(data1_, 0.0); break;
        case 2: Kokkos::deep_copy(data2_, 0.0); break;
        case 3: Kokkos::deep_copy(data3_, 0.0); break;
        case 4: Kokkos::deep_copy(data4_, 0.0); break;
        case 5: Kokkos::deep_copy(data5_, 0.0); break;
        case 6: Kokkos::deep_copy(data6_, 0.0); break;
        case 7: Kokkos::deep_copy(data7_, 0.0); break;
        default: INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Invalid data rank");
      }
    }
    
    void copyDataFromDynRankViewMatchingUnderlying(const ScalarView<DataScalar,DeviceType> &dynRankView) const
    {
//      std::cout << "Entered copyDataFromDynRankViewMatchingUnderlying().\n";
      switch (dataRank_)
      {
        case 1: copyContainer(data1_,dynRankView); break;
        case 2: copyContainer(data2_,dynRankView); break;
        case 3: copyContainer(data3_,dynRankView); break;
        case 4: copyContainer(data4_,dynRankView); break;
        case 5: copyContainer(data5_,dynRankView); break;
        case 6: copyContainer(data6_,dynRankView); break;
        case 7: copyContainer(data7_,dynRankView); break;
        default: INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Invalid data rank");
      }
    }
    
    KOKKOS_INLINE_FUNCTION int getDataExtent(const ordinal_type &d) const
    {
      for (int i=0; i<numActiveDims_; i++)
      {
        if (activeDims_[i] == d)
        {
          return getUnderlyingViewExtent(i);
        }
        else if (activeDims_[i] > d)
        {
          return 1; // data does not vary in the specified dimension
        }
      }
      return 1; // data does not vary in the specified dimension
    }
    
    KOKKOS_INLINE_FUNCTION
    int getVariationModulus(const int &d) const
    {
      return variationModulus_[d];
    }
    
    KOKKOS_INLINE_FUNCTION
    const Kokkos::Array<DataVariationType,7> & getVariationTypes() const
    {
      return variationType_;
    }
    
    template<class ...IntArgs>
    KOKKOS_INLINE_FUNCTION
    return_type getEntryWithPassThroughOption(const bool &passThroughBlockDiagonalArgs, const IntArgs&... intArgs) const
    {
      return getWritableEntryWithPassThroughOption(passThroughBlockDiagonalArgs, intArgs...);
    }
    
    template<class ...IntArgs>
    KOKKOS_INLINE_FUNCTION
    return_type getEntry(const IntArgs&... intArgs) const
    {
      return getEntryWithPassThroughOption(false, intArgs...);
    }
    
    template <bool...> struct bool_pack;
 
    template <bool... v>
    using all_true = std::is_same<bool_pack<true, v...>, bool_pack<v..., true>>;
    
    template <class ...IntArgs>
    using valid_args = all_true<std::is_integral<IntArgs>{}...>;
    
    static_assert(valid_args<int,long,unsigned>::value, "valid args works");

    template <class ...IntArgs>
    KOKKOS_INLINE_FUNCTION
#ifndef __INTEL_COMPILER
    // icc has a bug that prevents compilation with this enable_if_t
    // (possibly the same as https://community.intel.com/t5/Intel-C-Compiler/Intel-Compiler-bug-while-deducing-template-arguments-inside/m-p/1164358)
    // so with icc we'll just skip the argument type/count check
    enable_if_t<valid_args<IntArgs...>::value && (sizeof...(IntArgs) <= 7),return_type>
#else
    return_type
#endif
    operator()(const IntArgs&... intArgs) const {
      return getEntry(intArgs...);
    }

    KOKKOS_INLINE_FUNCTION
    int extent_int(const int& r) const
    {
      return extents_[r];
    }
    
    template <typename iType>
    KOKKOS_INLINE_FUNCTION constexpr
    typename std::enable_if<std::is_integral<iType>::value, size_t>::type
    extent(const iType& r) const {
      return extents_[r];
    }
    
    KOKKOS_INLINE_FUNCTION bool isDiagonal() const
    {
      if (blockPlusDiagonalLastNonDiagonal_ >= 1) return false;
      int numBlockPlusDiagonalTypes = 0;
      for (unsigned r = 0; r<variationType_.size(); r++)
      {
        const auto &entryType = variationType_[r];
        if (entryType == BLOCK_PLUS_DIAGONAL) numBlockPlusDiagonalTypes++;
      }
      // 2 BLOCK_PLUS_DIAGONAL entries, combined with blockPlusDiagonalLastNonDiagonal being -1 or 0 indicates diagonal
      if      (numBlockPlusDiagonalTypes == 2) return true;
      else if (numBlockPlusDiagonalTypes == 0) return false; // no BLOCK_PLUS_DIAGONAL --> not a diagonal matrix
      else INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(true, std::invalid_argument, "Unexpected number of ranks marked as BLOCK_PLUS_DIAGONAL (should be 0 or 2)");
      return false; // statement should be unreachable; included because compilers don't necessarily recognize that fact...
    }
    
    Data<DataScalar,DeviceType> allocateConstantData( const DataScalar &value )
    {
      return Data<DataScalar,DeviceType>(value, this->getExtents());
    }
    
    static Data<DataScalar,DeviceType> allocateInPlaceCombinationResult( const Data<DataScalar,DeviceType> &A, const Data<DataScalar,DeviceType> &B )
    {
      INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(A.rank() != B.rank(), std::invalid_argument, "A and B must have the same logical shape");
      const int rank = A.rank();
      std::vector<DimensionInfo> dimInfo(rank);
      for (int d=0; d<rank; d++)
      {
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(A.extent_int(d) != B.extent_int(d), std::invalid_argument, "A and B must have the same logical shape");
        dimInfo[d] = A.combinedDataDimensionInfo(B, d);
      }
      Data<DataScalar,DeviceType> result(dimInfo);
      return result;
    }
    
    static Data<DataScalar,DeviceType> allocateMatMatResult( const bool transposeA, const Data<DataScalar,DeviceType> &A_MatData, const bool transposeB, const Data<DataScalar,DeviceType> &B_MatData )
    {
      // we treat last two logical dimensions of matData as the matrix; last dimension of vecData as the vector
      INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(A_MatData.rank() != B_MatData.rank(), std::invalid_argument, "AmatData and BmatData have incompatible ranks");
      
      const int D1_DIM = A_MatData.rank() - 2;
      const int D2_DIM = A_MatData.rank() - 1;
      
      const int A_rows = A_MatData.extent_int(D1_DIM);
      const int A_cols = A_MatData.extent_int(D2_DIM);
      const int B_rows = B_MatData.extent_int(D1_DIM);
      const int B_cols = B_MatData.extent_int(D2_DIM);
      
      const int leftRows  = transposeA ? A_cols : A_rows;
      const int leftCols  = transposeA ? A_rows : A_cols;
      const int rightRows = transposeB ? B_cols : B_rows;
      const int rightCols = transposeB ? B_rows : B_cols;
      
      INTREPID2_TEST_FOR_EXCEPTION(leftCols != rightRows, std::invalid_argument, "incompatible matrix dimensions");
      
      Kokkos::Array<int,7> resultExtents;                      // logical extents
      Kokkos::Array<DataVariationType,7> resultVariationTypes; // for each dimension, whether the data varies in that dimension
      
      resultExtents[D1_DIM] = leftRows;
      resultExtents[D2_DIM] = rightCols;
      int resultBlockPlusDiagonalLastNonDiagonal = -1;
      if ( (A_MatData.getVariationTypes()[D1_DIM] == BLOCK_PLUS_DIAGONAL) && (B_MatData.getVariationTypes()[D1_DIM] == BLOCK_PLUS_DIAGONAL) )
      {
        // diagonal times diagonal is diagonal; the result will have the maximum of A and B's non-diagonal block size
        resultVariationTypes[D1_DIM] = BLOCK_PLUS_DIAGONAL;
        resultVariationTypes[D2_DIM] = BLOCK_PLUS_DIAGONAL;
        resultBlockPlusDiagonalLastNonDiagonal = std::max(A_MatData.blockPlusDiagonalLastNonDiagonal(), B_MatData.blockPlusDiagonalLastNonDiagonal());
      }
      
      const int resultRank = A_MatData.rank();
      
      auto A_VariationTypes = A_MatData.getVariationTypes();
      auto B_VariationTypes = B_MatData.getVariationTypes();
      
      Kokkos::Array<int,7> resultActiveDims;
      Kokkos::Array<int,7> resultDataDims;
      int resultNumActiveDims = 0; // how many of the 7 entries are actually filled in
      // the following loop is over the dimensions *prior* to matrix dimensions
      for (int i=0; i<resultRank-2; i++)
      {
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(A_MatData.extent_int(i) != B_MatData.extent_int(i), std::invalid_argument, "A and B extents must match in each non-matrix dimension");
        
        resultExtents[i] = A_MatData.extent_int(i);
        
        const DataVariationType &A_VariationType = A_VariationTypes[i];
        const DataVariationType &B_VariationType = B_VariationTypes[i];
        
        // BLOCK_PLUS_DIAGONAL should only occur in matData, and only in the matrix (final) dimensions
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(A_VariationType == BLOCK_PLUS_DIAGONAL, std::invalid_argument, "unsupported variationType");
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(B_VariationType == BLOCK_PLUS_DIAGONAL, std::invalid_argument, "unsupported variationType");
        
        int dataSize = 0;
        DataVariationType resultVariationType;
        if ((A_VariationType == GENERAL) || (B_VariationType == GENERAL))
        {
          resultVariationType = GENERAL;
          dataSize = resultExtents[i];
        }
        else if ((B_VariationType == CONSTANT) && (A_VariationType == CONSTANT))
        {
          resultVariationType = CONSTANT;
          dataSize = 1;
        }
        else if ((B_VariationType == MODULAR) && (A_VariationType == CONSTANT))
        {
          resultVariationType = MODULAR;
          dataSize = B_MatData.getVariationModulus(i);
        }
        else if ((B_VariationType == CONSTANT) && (A_VariationType == MODULAR))
        {
          resultVariationType = MODULAR;
          dataSize = A_MatData.getVariationModulus(i);
        }
        else
        {
          // both are modular.  We allow this if they agree on the modulus
          auto A_Modulus = A_MatData.getVariationModulus(i);
          auto B_Modulus = B_MatData.getVariationModulus(i);
          INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(A_Modulus != B_Modulus, std::invalid_argument, "If both matrices have variation type MODULAR, they must agree on the modulus");
          resultVariationType = MODULAR;
          dataSize = A_Modulus;
        }
        resultVariationTypes[i] = resultVariationType;
        
        if (resultVariationType != CONSTANT)
        {
          resultActiveDims[resultNumActiveDims] = i;
          resultDataDims[resultNumActiveDims]   = dataSize;
          resultNumActiveDims++;
        }
      }
      
      // set things for final dimensions:
      resultExtents[D1_DIM] = leftRows;
      resultExtents[D2_DIM] = rightCols;
      
      if ( (A_MatData.getVariationTypes()[D1_DIM] == BLOCK_PLUS_DIAGONAL) && (B_MatData.getVariationTypes()[D1_DIM] == BLOCK_PLUS_DIAGONAL) )
      {
        // diagonal times diagonal is diagonal; the result will have the maximum of A and B's non-diagonal block size
        resultVariationTypes[D1_DIM] = BLOCK_PLUS_DIAGONAL;
        resultVariationTypes[D2_DIM] = BLOCK_PLUS_DIAGONAL;
        resultBlockPlusDiagonalLastNonDiagonal = std::max(A_MatData.blockPlusDiagonalLastNonDiagonal(), B_MatData.blockPlusDiagonalLastNonDiagonal());
        
        resultActiveDims[resultNumActiveDims]   = resultRank - 2;
        
        const int numDiagonalEntries    = leftRows - (resultBlockPlusDiagonalLastNonDiagonal + 1);
        const int numNondiagonalEntries = (resultBlockPlusDiagonalLastNonDiagonal + 1) * (resultBlockPlusDiagonalLastNonDiagonal + 1);
        
        resultDataDims[resultNumActiveDims] = numDiagonalEntries + numNondiagonalEntries;
        resultNumActiveDims++;
      }
      else
      {
        // pretty much the only variation types that make sense for matrix dims are GENERAL and BLOCK_PLUS_DIAGONAL
        resultVariationTypes[D1_DIM] = GENERAL;
        resultVariationTypes[D2_DIM] = GENERAL;
        
        resultActiveDims[resultNumActiveDims]   = resultRank - 2;
        resultActiveDims[resultNumActiveDims+1] = resultRank - 1;
        
        resultDataDims[resultNumActiveDims]     = leftRows;
        resultDataDims[resultNumActiveDims+1]   = rightCols;
        resultNumActiveDims += 2;
      }
      
      for (int i=resultRank; i<7; i++)
      {
        resultVariationTypes[i] = CONSTANT;
        resultExtents[i]        = 1;
      }
      
      ScalarView<DataScalar,DeviceType> data;
      if (resultNumActiveDims == 1)
      {
        auto viewToMatch = A_MatData.getUnderlyingView1(); // new view will match this one in layout and fad dimension, if any
        data = getMatchingViewWithLabel(viewToMatch, "Data mat-mat result", resultDataDims[0]);
      }
      else if (resultNumActiveDims == 2)
      {
        auto viewToMatch = A_MatData.getUnderlyingView2(); // new view will match this one in layout and fad dimension, if any
        data = getMatchingViewWithLabel(viewToMatch, "Data mat-mat result", resultDataDims[0], resultDataDims[1]);
      }
      else if (resultNumActiveDims == 3)
      {
        auto viewToMatch = A_MatData.getUnderlyingView3(); // new view will match this one in layout and fad dimension, if any
        data = getMatchingViewWithLabel(viewToMatch, "Data mat-mat result", resultDataDims[0], resultDataDims[1], resultDataDims[2]);
      }
      else if (resultNumActiveDims == 4)
      {
        auto viewToMatch = A_MatData.getUnderlyingView4(); // new view will match this one in layout and fad dimension, if any
        data = getMatchingViewWithLabel(viewToMatch, "Data mat-mat result", resultDataDims[0], resultDataDims[1], resultDataDims[2],
                                        resultDataDims[3]);
      }
      else if (resultNumActiveDims == 5)
      {
        auto viewToMatch = A_MatData.getUnderlyingView5(); // new view will match this one in layout and fad dimension, if any
        data = getMatchingViewWithLabel(viewToMatch, "Data mat-mat result", resultDataDims[0], resultDataDims[1], resultDataDims[2],
                                        resultDataDims[3], resultDataDims[4]);
      }
      else if (resultNumActiveDims == 6)
      {
        auto viewToMatch = A_MatData.getUnderlyingView6(); // new view will match this one in layout and fad dimension, if any
        data = getMatchingViewWithLabel(viewToMatch, "Data mat-mat result", resultDataDims[0], resultDataDims[1], resultDataDims[2],
                                        resultDataDims[3], resultDataDims[4], resultDataDims[5]);
      }
      else // resultNumActiveDims == 7
      {
        auto viewToMatch = A_MatData.getUnderlyingView7(); // new view will match this one in layout and fad dimension, if any
        data = getMatchingViewWithLabel(viewToMatch, "Data mat-mat result", resultDataDims[0], resultDataDims[1], resultDataDims[2],
                                        resultDataDims[3], resultDataDims[4], resultDataDims[5], resultDataDims[6]);
      }
      
      return Data<DataScalar,DeviceType>(data,resultRank,resultExtents,resultVariationTypes,resultBlockPlusDiagonalLastNonDiagonal);
    }
    
    static Data<DataScalar,DeviceType> allocateMatVecResult( const Data<DataScalar,DeviceType> &matData, const Data<DataScalar,DeviceType> &vecData )
    {
      // we treat last two logical dimensions of matData as the matrix; last dimension of vecData as the vector
      INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(matData.rank() != vecData.rank() + 1, std::invalid_argument, "matData and vecData have incompatible ranks");
      const int vecDim  = vecData.extent_int(vecData.rank() - 1);
      const int matRows = matData.extent_int(matData.rank() - 2);
      const int matCols = matData.extent_int(matData.rank() - 1);
      
      const int resultRank = vecData.rank();
      
      INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(matCols != vecDim, std::invalid_argument, "matData column count != vecData dimension");
      
      Kokkos::Array<int,7> resultExtents;                      // logical extents
      Kokkos::Array<DataVariationType,7> resultVariationTypes; // for each dimension, whether the data varies in that dimension
      auto vecVariationTypes = vecData.getVariationTypes();
      auto matVariationTypes = matData.getVariationTypes();
      
      Kokkos::Array<int,7> resultActiveDims;
      Kokkos::Array<int,7> resultDataDims;
      int resultNumActiveDims = 0; // how many of the 7 entries are actually filled in
      // the following loop is over the dimensions *prior* to matrix/vector dimensions
      for (int i=0; i<resultRank-1; i++)
      {
        resultExtents[i] = vecData.extent_int(i);
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(vecData.extent_int(i) != matData.extent_int(i), std::invalid_argument, "matData and vecData extents must match in each non-matrix/vector dimension");
        
        const DataVariationType &vecVariationType = vecVariationTypes[i];
        const DataVariationType &matVariationType = matVariationTypes[i];
        
        // BLOCK_PLUS_DIAGONAL should only occur in matData, and only in the matrix (final) dimensions
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(vecVariationType == BLOCK_PLUS_DIAGONAL, std::invalid_argument, "unsupported variationType");
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(matVariationType == BLOCK_PLUS_DIAGONAL, std::invalid_argument, "unsupported variationType");
        
        int dataSize = 0;
        DataVariationType resultVariationType;
        if ((vecVariationType == GENERAL) || (matVariationType == GENERAL))
        {
          resultVariationType = GENERAL;
          dataSize = resultExtents[i];
        }
        else if ((matVariationType == CONSTANT) && (vecVariationType == CONSTANT))
        {
          resultVariationType = CONSTANT;
          dataSize = 1;
        }
        else if ((matVariationType == MODULAR) && (vecVariationType == CONSTANT))
        {
          resultVariationType = MODULAR;
          dataSize = matData.getVariationModulus(i);
        }
        else if ((matVariationType == CONSTANT) && (vecVariationType == MODULAR))
        {
          resultVariationType = MODULAR;
          dataSize = matData.getVariationModulus(i);
        }
        else
        {
          // both are modular.  We allow this if they agree on the modulus
          auto matModulus = matData.getVariationModulus(i);
          auto vecModulus = vecData.getVariationModulus(i);
          INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(matModulus != vecModulus, std::invalid_argument, "If matrix and vector both have variation type MODULAR, they must agree on the modulus");
          resultVariationType = MODULAR;
          dataSize = matModulus;
        }
        resultVariationTypes[i] = resultVariationType;
        
        if (resultVariationType != CONSTANT)
        {
          resultActiveDims[resultNumActiveDims] = i;
          resultDataDims[resultNumActiveDims]   = dataSize;
          resultNumActiveDims++;
        }
      }
      // for the final dimension, the variation type is always GENERAL
      // (Some combinations, e.g. CONSTANT/CONSTANT *would* generate a CONSTANT result, but constant matrices don't make a lot of sense beyond 1x1 matrices…)
      resultVariationTypes[resultNumActiveDims] = GENERAL;
      resultActiveDims[resultNumActiveDims]     = resultRank - 1;
      resultDataDims[resultNumActiveDims]       = matRows;
      resultExtents[resultRank-1]               = matRows;
      resultNumActiveDims++;
      
      for (int i=resultRank; i<7; i++)
      {
        resultVariationTypes[i] = CONSTANT;
        resultExtents[i]        = 1;
      }
      
      ScalarView<DataScalar,DeviceType> data;
      if (resultNumActiveDims == 1)
      {
        data = matData.allocateDynRankViewMatchingUnderlying(resultDataDims[0]);
      }
      else if (resultNumActiveDims == 2)
      {
        data = matData.allocateDynRankViewMatchingUnderlying(resultDataDims[0], resultDataDims[1]);
      }
      else if (resultNumActiveDims == 3)
      {
        data = matData.allocateDynRankViewMatchingUnderlying(resultDataDims[0], resultDataDims[1], resultDataDims[2]);
      }
      else if (resultNumActiveDims == 4)
      {
        data = matData.allocateDynRankViewMatchingUnderlying(resultDataDims[0], resultDataDims[1], resultDataDims[2],
                                                             resultDataDims[3]);
      }
      else if (resultNumActiveDims == 5)
      {
        data = matData.allocateDynRankViewMatchingUnderlying(resultDataDims[0], resultDataDims[1], resultDataDims[2],
                                                             resultDataDims[3], resultDataDims[4]);
      }
      else if (resultNumActiveDims == 6)
      {
        data = matData.allocateDynRankViewMatchingUnderlying(resultDataDims[0], resultDataDims[1], resultDataDims[2],
                                                             resultDataDims[3], resultDataDims[4], resultDataDims[5]);
      }
      else // resultNumActiveDims == 7
      {
        data = matData.allocateDynRankViewMatchingUnderlying(resultDataDims[0], resultDataDims[1], resultDataDims[2],
                                                             resultDataDims[3], resultDataDims[4], resultDataDims[5], resultDataDims[6]);
      }
      
      return Data<DataScalar,DeviceType>(data,resultRank,resultExtents,resultVariationTypes);
    }
    
    template<int rank>
    enable_if_t<(rank!=1) && (rank!=7), Kokkos::MDRangePolicy<typename DeviceType::execution_space,Kokkos::Rank<rank>> >
    dataExtentRangePolicy()
    {
      using ExecutionSpace = typename DeviceType::execution_space;
      Kokkos::Array<int,rank> startingOrdinals;
      Kokkos::Array<int,rank> extents;
      
      for (int d=0; d<rank; d++)
      {
        startingOrdinals[d] = 0;
        extents[d] = getDataExtent(d);
      }
      auto policy = Kokkos::MDRangePolicy<ExecutionSpace,Kokkos::Rank<rank>>(startingOrdinals,extents);
      return policy;
    }
    
    template<int rank>
    enable_if_t<rank==7, Kokkos::MDRangePolicy<typename DeviceType::execution_space,Kokkos::Rank<6>> >
    dataExtentRangePolicy()
    {
      using ExecutionSpace = typename DeviceType::execution_space;
      Kokkos::Array<int,6> startingOrdinals;
      Kokkos::Array<int,6> extents;
      
      for (int d=0; d<6; d++)
      {
        startingOrdinals[d] = 0;
        extents[d] = getDataExtent(d);
      }
      auto policy = Kokkos::MDRangePolicy<ExecutionSpace,Kokkos::Rank<6>>(startingOrdinals,extents);
      return policy;
    }
    
    template<int rank>
    inline
    enable_if_t<rank==1, Kokkos::RangePolicy<typename DeviceType::execution_space> >
    dataExtentRangePolicy()
    {
      using ExecutionSpace = typename DeviceType::execution_space;
      Kokkos::RangePolicy<ExecutionSpace> policy(ExecutionSpace(),0,getDataExtent(0));
      return policy;
    }
    
    Data shallowCopy(const int rank, const Kokkos::Array<int,7> &extents, const Kokkos::Array<DataVariationType,7> &variationTypes) const
    {
      switch (dataRank_)
      {
        case 1: return Data(rank, data1_, extents, variationTypes);
        case 2: return Data(rank, data2_, extents, variationTypes);
        case 3: return Data(rank, data3_, extents, variationTypes);
        case 4: return Data(rank, data4_, extents, variationTypes);
        case 5: return Data(rank, data5_, extents, variationTypes);
        case 6: return Data(rank, data6_, extents, variationTypes);
        case 7: return Data(rank, data7_, extents, variationTypes);
        default:
          INTREPID2_TEST_FOR_EXCEPTION(true, std::invalid_argument, "Unhandled dataRank_");
      }
    }
    
    template<class BinaryOperator>
    void storeInPlaceCombination(const Data<DataScalar,DeviceType> &A, const Data<DataScalar,DeviceType> &B, BinaryOperator binaryOperator);
    
    void storeInPlaceSum(const Data<DataScalar,DeviceType> &A, const Data<DataScalar,DeviceType> &B)
    {
      ScalarSumFunctor<DataScalar> sum;
      storeInPlaceCombination(A, B, sum);
    }
    
    void storeInPlaceProduct(const Data<DataScalar,DeviceType> &A, const Data<DataScalar,DeviceType> &B)
    {
      ScalarProductFunctor<DataScalar> product;
      storeInPlaceCombination(A, B, product);
    }
    
    void storeInPlaceDifference(const Data<DataScalar,DeviceType> &A, const Data<DataScalar,DeviceType> &B)
    {
      ScalarDifferenceFunctor<DataScalar> difference;
      storeInPlaceCombination(A, B, difference);
    }
    
    void storeInPlaceQuotient(const Data<DataScalar,DeviceType> &A, const Data<DataScalar,DeviceType> &B)
    {
      ScalarQuotientFunctor<DataScalar> quotient;
      storeInPlaceCombination(A, B, quotient);
    }
    
    void storeMatVec( const Data<DataScalar,DeviceType> &matData, const Data<DataScalar,DeviceType> &vecData )
    {
      // TODO: add a compile-time (SFINAE-type) guard against DataScalar types that do not support arithmetic operations.  (We support Orientation as a DataScalar type; it might suffice just to compare DataScalar to Orientation, and eliminate this method for that case.)
      // TODO: check for invalidly shaped containers.
      
      const int matRows = matData.extent_int(matData.rank() - 2);
      const int matCols = matData.extent_int(matData.rank() - 1);
      
      // shallow copy of this to avoid implicit references to this in call to getWritableEntry() below
      Data<DataScalar,DeviceType> thisData = *this;
      
      using ExecutionSpace = typename DeviceType::execution_space;
      // note the use of getDataExtent() below: we only range over the possibly-distinct entries
      if (rank_ == 3)
      {
        // typical case for e.g. gradient data: (C,P,D)
        auto policy = Kokkos::MDRangePolicy<ExecutionSpace,Kokkos::Rank<3>>({0,0,0},{getDataExtent(0),getDataExtent(1),matRows});
        Kokkos::parallel_for("compute mat-vec", policy,
        KOKKOS_LAMBDA (const int &cellOrdinal, const int &pointOrdinal, const int &i) {
          auto & val_i = thisData.getWritableEntry(cellOrdinal, pointOrdinal, i);
          val_i = 0;
          for (int j=0; j<matCols; j++)
          {
            val_i += matData(cellOrdinal,pointOrdinal,i,j) * vecData(cellOrdinal,pointOrdinal,j);
          }
        });
      }
      else if (rank_ == 2)
      {
        //
        auto policy = Kokkos::MDRangePolicy<ExecutionSpace,Kokkos::Rank<2>>({0,0},{getDataExtent(0),matRows});
        Kokkos::parallel_for("compute mat-vec", policy,
        KOKKOS_LAMBDA (const int &vectorOrdinal, const int &i) {
          auto & val_i = thisData.getWritableEntry(vectorOrdinal, i);
          val_i = 0;
          for (int j=0; j<matCols; j++)
          {
            val_i += matData(vectorOrdinal,i,j) * vecData(vectorOrdinal,j);
          }
        });
      }
      else if (rank_ == 1)
      {
        // single-vector case
        Kokkos::RangePolicy<ExecutionSpace> policy(0,matRows);
        Kokkos::parallel_for("compute mat-vec", policy,
        KOKKOS_LAMBDA (const int &i) {
          auto & val_i = thisData.getWritableEntry(i);
          val_i = 0;
          for (int j=0; j<matCols; j++)
          {
            val_i += matData(i,j) * vecData(j);
          }
        });
      }
      else
      {
        // TODO: handle other cases
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(true, std::logic_error, "rank not yet supported");
      }
    }
    
    void storeMatMat( const bool transposeA, const Data<DataScalar,DeviceType> &A_MatData, const bool transposeB, const Data<DataScalar,DeviceType> &B_MatData )
    {
      // TODO: add a compile-time (SFINAE-type) guard against DataScalar types that do not support arithmetic operations.  (We support Orientation as a DataScalar type; it might suffice just to compare DataScalar to Orientation, and eliminate this method for that case.)
      // TODO: check for invalidly shaped containers.
      
      // we treat last two logical dimensions of matData as the matrix; last dimension of vecData as the vector
      INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(A_MatData.rank() != B_MatData.rank(), std::invalid_argument, "AmatData and BmatData have incompatible ranks");
      
      const int D1_DIM = A_MatData.rank() - 2;
      const int D2_DIM = A_MatData.rank() - 1;
      
      const int A_rows = A_MatData.extent_int(D1_DIM);
      const int A_cols = A_MatData.extent_int(D2_DIM);
      const int B_rows = B_MatData.extent_int(D1_DIM);
      const int B_cols = B_MatData.extent_int(D2_DIM);
     
      const int leftRows  = transposeA ? A_cols : A_rows;
      const int leftCols  = transposeA ? A_rows : A_cols;
      const int rightCols = transposeB ? B_rows : B_cols;
      
#ifdef INTREPID2_HAVE_DEBUG
      const int rightRows = transposeB ? B_cols : B_rows;
      INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(leftCols != rightRows, std::invalid_argument, "inner dimensions do not match");
#endif
      
      // shallow copy of this to avoid implicit references to this in call to getWritableEntry() below
      Data<DataScalar,DeviceType> thisData = *this;
      
      using ExecutionSpace = typename DeviceType::execution_space;
      
      const int diagonalStart = (variationType_[D1_DIM] == BLOCK_PLUS_DIAGONAL) ? blockPlusDiagonalLastNonDiagonal_ + 1 : leftRows;
      // note the use of getDataExtent() below: we only range over the possibly-distinct entries
      if (rank_ == 3)
      {
        // (C,D,D), say
        auto policy = Kokkos::RangePolicy<ExecutionSpace>(0,getDataExtent(0));
        Kokkos::parallel_for("compute mat-mat", policy,
        KOKKOS_LAMBDA (const int &matrixOrdinal) {
          for (int i=0; i<diagonalStart; i++)
          {
            for (int j=0; j<rightCols; j++)
            {
              auto & val_ij = thisData.getWritableEntry(matrixOrdinal, i, j);
              val_ij = 0;
              for (int k=0; k<leftCols; k++)
              {
                const auto & left  = transposeA ? A_MatData(matrixOrdinal,k,i) : A_MatData(matrixOrdinal,i,k);
                const auto & right = transposeB ? B_MatData(matrixOrdinal,j,k) : B_MatData(matrixOrdinal,k,j);
                val_ij += left * right;
              }
            }
          }
          for (int i=diagonalStart; i<leftRows; i++)
          {
            auto & val_ii = thisData.getWritableEntry(matrixOrdinal, i, i);
            const auto & left  = A_MatData(matrixOrdinal,i,i);
            const auto & right = B_MatData(matrixOrdinal,i,i);
            val_ii = left * right;
          }
        });
      }
      else if (rank_ == 4)
      {
        // (C,P,D,D), perhaps
        auto policy = Kokkos::MDRangePolicy<ExecutionSpace, Kokkos::Rank<2> >({0,0},{getDataExtent(0),getDataExtent(1)});
        if (underlyingMatchesLogical_) // receiving data object is completely expanded
        {
          Kokkos::parallel_for("compute mat-mat", policy,
          KOKKOS_LAMBDA (const int &cellOrdinal, const int &pointOrdinal) {
            for (int i=0; i<leftCols; i++)
            {
              for (int j=0; j<rightCols; j++)
              {
                auto & val_ij = thisData.getUnderlyingView4()(cellOrdinal,pointOrdinal, i, j);
                val_ij = 0;
                for (int k=0; k<leftCols; k++)
                {
                  const auto & left  = transposeA ? A_MatData(cellOrdinal,pointOrdinal,k,i) : A_MatData(cellOrdinal,pointOrdinal,i,k);
                  const auto & right = transposeB ? B_MatData(cellOrdinal,pointOrdinal,j,k) : B_MatData(cellOrdinal,pointOrdinal,k,j);
                  val_ij += left * right;
                }
              }
            }
          });
        }
        else
        {
          Kokkos::parallel_for("compute mat-mat", policy,
          KOKKOS_LAMBDA (const int &cellOrdinal, const int &pointOrdinal) {
            for (int i=0; i<diagonalStart; i++)
            {
              for (int j=0; j<rightCols; j++)
              {
                auto & val_ij = thisData.getWritableEntry(cellOrdinal,pointOrdinal, i, j);
                val_ij = 0;
                for (int k=0; k<leftCols; k++)
                {
                  const auto & left  = transposeA ? A_MatData(cellOrdinal,pointOrdinal,k,i) : A_MatData(cellOrdinal,pointOrdinal,i,k);
                  const auto & right = transposeB ? B_MatData(cellOrdinal,pointOrdinal,j,k) : B_MatData(cellOrdinal,pointOrdinal,k,j);
                  val_ij += left * right;
                }
              }
            }
            for (int i=diagonalStart; i<leftRows; i++)
            {
              auto & val_ii = thisData.getWritableEntry(cellOrdinal,pointOrdinal, i, i);
              const auto & left  = A_MatData(cellOrdinal,pointOrdinal,i,i);
              const auto & right = B_MatData(cellOrdinal,pointOrdinal,i,i);
              val_ii = left * right;
            }
          });
        }
      }
      else
      {
        // TODO: handle other cases
        INTREPID2_TEST_FOR_EXCEPTION_DEVICE_SAFE(true, std::logic_error, "rank not yet supported");
      }
    }
    
    KOKKOS_INLINE_FUNCTION constexpr bool isValid() const
    {
      return extents_[0] > 0;
    }
    
    KOKKOS_INLINE_FUNCTION
    unsigned rank() const
    {
      return rank_;
    }
    
    void setExtent(const ordinal_type &d, const ordinal_type &newExtent)
    {
      INTREPID2_TEST_FOR_EXCEPTION(variationType_[d] == BLOCK_PLUS_DIAGONAL, std::invalid_argument, "setExtent is not supported for BLOCK_PLUS_DIAGONAL dimensions");
      
      if (variationType_[d] == MODULAR)
      {
        bool dividesEvenly = ((newExtent / variationModulus_[d]) * variationModulus_[d] == newExtent);
        INTREPID2_TEST_FOR_EXCEPTION(!dividesEvenly, std::invalid_argument, "when setExtent is called on dimenisions with MODULAR variation, the modulus must divide the new extent evenly");
      }
      
      if ((newExtent != extents_[d]) && (variationType_[d] == GENERAL))
      {
        // then we need to resize; let's determine the full set of new extents
        std::vector<ordinal_type> newExtents(dataRank_,-1);
        for (int r=0; r<dataRank_; r++)
        {
          if (activeDims_[r] == d)
          {
            // this is the changed dimension
            newExtents[r] = newExtent;
          }
          else
          {
            // unchanged; copy from existing
            newExtents[r] = getUnderlyingViewExtent(r);
          }
        }
        
        switch (dataRank_)
        {
          case 1: Kokkos::resize(data1_,newExtents[0]);
            break;
          case 2: Kokkos::resize(data2_,newExtents[0],newExtents[1]);
            break;
          case 3: Kokkos::resize(data3_,newExtents[0],newExtents[1],newExtents[2]);
            break;
          case 4: Kokkos::resize(data4_,newExtents[0],newExtents[1],newExtents[2],newExtents[3]);
            break;
          case 5: Kokkos::resize(data5_,newExtents[0],newExtents[1],newExtents[2],newExtents[3],newExtents[4]);
            break;
          case 6: Kokkos::resize(data6_,newExtents[0],newExtents[1],newExtents[2],newExtents[3],newExtents[4],newExtents[5]);
            break;
          case 7: Kokkos::resize(data7_,newExtents[0],newExtents[1],newExtents[2],newExtents[3],newExtents[4],newExtents[5],newExtents[6]);
            break;
          default: INTREPID2_TEST_FOR_EXCEPTION(true, std::logic_error, "Unexpected dataRank_ value");
        }
      }
      
      extents_[d] = newExtent;
    }
    
    KOKKOS_INLINE_FUNCTION
    bool underlyingMatchesLogical() const
    {
      return underlyingMatchesLogical_;
    }
  };
}
 
// we do ETI for doubles and default ExecutionSpace's device_type
extern template class Intrepid2::Data<double,Kokkos::DefaultExecutionSpace::device_type>;
 
#include "Intrepid2_DataDef.hpp"
 
#endif /* Intrepid2_Data_h */