// Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC.
// Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707.
// All Rights reserved. See files LICENSE and NOTICE for details.
//
// This file is part of CEED, a collection of benchmarks, miniapps, software
// libraries and APIs for efficient high-order finite element and spectral
// element discretizations for exascale applications. For more information and
// source code availability see http://github.com/ceed.
//
// The CEED research is supported by the Exascale Computing Project 17-SC-20-SC,
// a collaborative effort of two U.S. Department of Energy organizations (Office
// of Science and the National Nuclear Security Administration) responsible for
// the planning and preparation of a capable exascale ecosystem, including
// software, applications, hardware, advanced system engineering and early
// testbed platforms, in support of the nation's exascale computing imperative.

#include <string.h>
#include "ceed-ref.h"

// Contracts on the middle index
// NOTRANSPOSE: V_ajc = T_jb U_abc
// TRANSPOSE:   V_ajc = T_bj U_abc
// If Add != 0, "=" is replaced by "+="
static int CeedTensorContract_Ref(Ceed ceed, CeedInt A, CeedInt B, CeedInt C, CeedInt J,
                                  const CeedScalar *restrict t, CeedTransposeMode tmode,
                                  const CeedInt Add,
                                  const CeedScalar *restrict u, CeedScalar *restrict v) {
  CeedInt tstride0 = B, tstride1 = 1;
  if (tmode == CEED_TRANSPOSE) {
    tstride0 = 1; tstride1 = J;
  }

  if (!Add)
    for (CeedInt q=0; q<A*J*C; q++)
      v[q] = (CeedScalar) 0.0;

  for (CeedInt a=0; a<A; a++)
    for (CeedInt b=0; b<B; b++)
      for (CeedInt j=0; j<J; j++) {
        CeedScalar tq = t[j*tstride0 + b*tstride1];
        for (CeedInt c=0; c<C; c++)
          v[(a*J+j)*C+c] += tq * u[(a*B+b)*C+c];
      }
  return 0;
}

static int CeedBasisApply_Ref(CeedBasis basis, CeedInt nelem,
                              CeedTransposeMode tmode, CeedEvalMode emode,
                              const CeedScalar *u, CeedScalar *v) {
  int ierr;
  Ceed ceed;
  ierr = CeedBasisGetCeed(basis, &ceed); CeedChk(ierr);
  CeedInt dim, ncomp, ndof, nqpt;
  ierr = CeedBasisGetDimension(basis, &dim); CeedChk(ierr);
  ierr = CeedBasisGetNumComponents(basis, &ncomp); CeedChk(ierr);
  ierr = CeedBasisGetNumNodes(basis, &ndof); CeedChk(ierr);
  ierr = CeedBasisGetNumQuadraturePoints(basis, &nqpt); CeedChk(ierr);
  const CeedInt add = (tmode == CEED_TRANSPOSE);

  if (nelem != 1)
    return CeedError(ceed, 1,
                     "This backend does not support BasisApply for multiple elements");

  // Clear v if operating in transpose
  if (tmode == CEED_TRANSPOSE) {
    const CeedInt vsize = ncomp*ndof;
    for (CeedInt i = 0; i < vsize; i++)
      v[i] = (CeedScalar) 0;
  }
  // Tensor basis
  bool tensorbasis;
  ierr = CeedBasisGetTensorStatus(basis, &tensorbasis); CeedChk(ierr);
  if (tensorbasis) {
    CeedInt P1d, Q1d;
    ierr = CeedBasisGetNumNodes1D(basis, &P1d); CeedChk(ierr);
    ierr = CeedBasisGetNumQuadraturePoints1D(basis, &Q1d); CeedChk(ierr);
    switch (emode) {
    // Interpolate to/from quadrature points
    case CEED_EVAL_INTERP: {
      CeedInt P = P1d, Q = Q1d;
      if (tmode == CEED_TRANSPOSE) {
        P = Q1d; Q = P1d;
      }
      CeedInt pre = ncomp*CeedIntPow(P, dim-1), post = 1;
      CeedScalar tmp[2][ncomp*Q*CeedIntPow(P>Q?P:Q, dim-1)];
      CeedScalar *interp1d;
      ierr = CeedBasisGetInterp(basis, &interp1d); CeedChk(ierr);
      for (CeedInt d=0; d<dim; d++) {
        ierr = CeedTensorContract_Ref(ceed, pre, P, post, Q, interp1d,
                                      tmode, add&&(d==dim-1),
                                      d==0?u:tmp[d%2], d==dim-1?v:tmp[(d+1)%2]);
        CeedChk(ierr);
        pre /= P;
        post *= Q;
      }
    } break;
    // Evaluate the gradient to/from quadrature points
    case CEED_EVAL_GRAD: {
      CeedInt P = P1d, Q = Q1d;
      // In CEED_NOTRANSPOSE mode:
      // u has shape [dim, ncomp, P^dim, nelem], row-major layout
      // v has shape [dim, ncomp, Q^dim, nelem], row-major layout
      // In CEED_TRANSPOSE mode, the sizes of u and v are switched.
      if (tmode == CEED_TRANSPOSE) {
        P = Q1d, Q = P1d;
      }
      CeedScalar tmp[2][ncomp*Q*CeedIntPow(P>Q?P:Q, dim-1)];
      CeedScalar *interp1d, *grad1d;
      ierr = CeedBasisGetInterp(basis, &interp1d); CeedChk(ierr);
      ierr = CeedBasisGetGrad(basis, &grad1d); CeedChk(ierr);
      for (CeedInt p = 0; p < dim; p++) {
        CeedInt pre = ncomp*CeedIntPow(P, dim-1), post = 1;
        for (CeedInt d=0; d<dim; d++) {
          ierr = CeedTensorContract_Ref(ceed, pre, P, post, Q,
                                        (p==d)?grad1d:interp1d,
                                        tmode, add&&(d==dim-1),
                                        (d == 0
                                         ? (tmode==CEED_NOTRANSPOSE?u:u+p*ncomp*nqpt)
                                         : tmp[d%2]),
                                        (d == dim-1
                                         ? (tmode==CEED_TRANSPOSE?v:v+p*ncomp*nqpt)
                                         : tmp[(d+1)%2]));
          CeedChk(ierr);
          pre /= P;
          post *= Q;
        }
      }
    } break;
    // Retrieve interpolation weights
    case CEED_EVAL_WEIGHT: {
      if (tmode == CEED_TRANSPOSE)
        return CeedError(ceed, 1,
                         "CEED_EVAL_WEIGHT incompatible with CEED_TRANSPOSE");
      CeedInt Q = Q1d;
      CeedScalar *qweight1d;
      ierr = CeedBasisGetQWeights(basis, &qweight1d); CeedChk(ierr);
      for (CeedInt d=0; d<dim; d++) {
        CeedInt pre = CeedIntPow(Q, dim-d-1), post = CeedIntPow(Q, d);
        for (CeedInt i=0; i<pre; i++)
          for (CeedInt j=0; j<Q; j++)
            for (CeedInt k=0; k<post; k++)
              v[(i*Q + j)*post + k] = qweight1d[j]
                                      * (d == 0 ? 1 : v[(i*Q + j)*post + k]);
      }
    } break;
    // Evaluate the divergence to/from the quadrature points
    case CEED_EVAL_DIV:
      return CeedError(ceed, 1, "CEED_EVAL_DIV not supported");
    // Evaluate the curl to/from the quadrature points
    case CEED_EVAL_CURL:
      return CeedError(ceed, 1, "CEED_EVAL_CURL not supported");
    // Take no action, BasisApply should not have been called
    case CEED_EVAL_NONE:
      return CeedError(ceed, 1,
                       "CEED_EVAL_NONE does not make sense in this context");
    }
  } else {
    // Non-tensor basis
    switch (emode) {
    case CEED_EVAL_INTERP: {
      CeedInt P = ndof, Q = nqpt;
      CeedScalar *interp;
      ierr = CeedBasisGetInterp(basis, &interp); CeedChk(ierr);
      if (tmode == CEED_TRANSPOSE) {
        P = nqpt; Q = ndof;
      }
      ierr = CeedTensorContract_Ref(ceed, ncomp, P, 1, Q, interp,
                                    tmode, add, u, v);
      CeedChk(ierr);
    }
    break;
    case CEED_EVAL_GRAD: {
      CeedInt P = ndof, Q = dim*nqpt;
      CeedScalar *grad;
      ierr = CeedBasisGetGrad(basis, &grad); CeedChk(ierr);
      if (tmode == CEED_TRANSPOSE) {
        P = dim*nqpt; Q = ndof;
      }
      ierr = CeedTensorContract_Ref(ceed, ncomp, P, 1, Q, grad,
                                    tmode, add, u, v);
      CeedChk(ierr);
    }
    break;
    case CEED_EVAL_WEIGHT: {
      if (tmode == CEED_TRANSPOSE)
        return CeedError(ceed, 1,
                         "CEED_EVAL_WEIGHT incompatible with CEED_TRANSPOSE");
      CeedScalar *qweight;
      ierr = CeedBasisGetQWeights(basis, &qweight); CeedChk(ierr);
      for (CeedInt i=0; i<nqpt; i++)
        v[i] = qweight[i];
    } break;
    case CEED_EVAL_DIV:
      return CeedError(ceed, 1, "CEED_EVAL_DIV not supported");
    case CEED_EVAL_CURL:
      return CeedError(ceed, 1, "CEED_EVAL_CURL not supported");
    case CEED_EVAL_NONE:
      return CeedError(ceed, 1,
                       "CEED_EVAL_NONE does not make sense in this context");
    }
  }
  return 0;
}

static int CeedBasisDestroy_Ref(CeedBasis basis) {
  return 0;
}

int CeedBasisCreateTensorH1_Ref(CeedInt dim, CeedInt P1d,
                                CeedInt Q1d, const CeedScalar *interp1d,
                                const CeedScalar *grad1d,
                                const CeedScalar *qref1d,
                                const CeedScalar *qweight1d,
                                CeedBasis basis) {
  basis->Apply = CeedBasisApply_Ref;
  basis->Destroy = CeedBasisDestroy_Ref;
  return 0;
}

int CeedBasisCreateH1_Ref(CeedElemTopology topo, CeedInt dim,
                          CeedInt ndof, CeedInt nqpts,
                          const CeedScalar *interp,
                          const CeedScalar *grad,
                          const CeedScalar *qref,
                          const CeedScalar *qweight,
                          CeedBasis basis) {
  basis->Apply = CeedBasisApply_Ref;
  basis->Destroy = CeedBasisDestroy_Ref;
  return 0;
}