tmop_pa_h3m.cpp

Go to the documentation of this file.

// Copyright (c) 2010-2025, Lawrence Livermore National Security, LLC. Produced
// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
// LICENSE and NOTICE for details. LLNL-CODE-806117.
//
// This file is part of the MFEM library. For more information and source code
// availability visit https://mfem.org.
//
// MFEM is free software; you can redistribute it and/or modify it under the
// terms of the BSD-3 license. We welcome feedback and contributions, see file
// CONTRIBUTING.md for details.
 
#include "../tmop.hpp"
#include "tmop_pa.hpp"
#include "../../general/forall.hpp"
#include "../../linalg/kernels.hpp"
 
namespace mfem
{
 
MFEM_REGISTER_TMOP_KERNELS(void, AddMultGradPA_Kernel_3D,
                           const int NE,
                           const Array<real_t> &b_,
                           const Array<real_t> &g_,
                           const DenseTensor &j_,
                           const Vector &h_,
                           const Vector &x_,
                           Vector &y_,
                           const int d1d,
                           const int q1d)
{
   constexpr int DIM = 3;
   const int D1D = T_D1D ? T_D1D : d1d;
   const int Q1D = T_Q1D ? T_Q1D : q1d;
 
   const auto b = Reshape(b_.Read(), Q1D, D1D);
   const auto g = Reshape(g_.Read(), Q1D, D1D);
   const auto J = Reshape(j_.Read(), DIM, DIM, Q1D, Q1D, Q1D, NE);
   const auto X = Reshape(x_.Read(), D1D, D1D, D1D, DIM, NE);
   const auto H = Reshape(h_.Read(), DIM, DIM, DIM, DIM, Q1D, Q1D, Q1D, NE);
   auto Y = Reshape(y_.ReadWrite(), D1D, D1D, D1D, DIM, NE);
 
   mfem::forall_3D(NE, Q1D, Q1D, Q1D, [=] MFEM_HOST_DEVICE (int e)
   {
      constexpr int DIM = 3;
      const int D1D = T_D1D ? T_D1D : d1d;
      const int Q1D = T_Q1D ? T_Q1D : q1d;
      constexpr int MQ1 = T_Q1D ? T_Q1D : T_MAX;
      constexpr int MD1 = T_D1D ? T_D1D : T_MAX;
 
      MFEM_SHARED real_t BG[2][MQ1*MD1];
      MFEM_SHARED real_t DDD[3][MD1*MD1*MD1];
      MFEM_SHARED real_t DDQ[9][MD1*MD1*MQ1];
      MFEM_SHARED real_t DQQ[9][MD1*MQ1*MQ1];
      MFEM_SHARED real_t QQQ[9][MQ1*MQ1*MQ1];
 
      kernels::internal::LoadX<MD1>(e,D1D,X,DDD);
      kernels::internal::LoadBG<MD1,MQ1>(D1D,Q1D,b,g,BG);
 
      kernels::internal::GradX<MD1,MQ1>(D1D,Q1D,BG,DDD,DDQ);
      kernels::internal::GradY<MD1,MQ1>(D1D,Q1D,BG,DDQ,DQQ);
      kernels::internal::GradZ<MD1,MQ1>(D1D,Q1D,BG,DQQ,QQQ);
 
      MFEM_FOREACH_THREAD(qz,z,Q1D)
      {
         MFEM_FOREACH_THREAD(qy,y,Q1D)
         {
            MFEM_FOREACH_THREAD(qx,x,Q1D)
            {
               const real_t *Jtr = &J(0,0,qx,qy,qz,e);
 
               // Jrt = Jtr^{-1}
               real_t Jrt[9];
               kernels::CalcInverse<3>(Jtr, Jrt);
 
               // Jpr = X^T.DSh
               real_t Jpr[9];
               kernels::internal::PullGrad<MQ1>(Q1D, qx,qy,qz, QQQ, Jpr);
 
               // Jpt = X^T.DS = (X^T.DSh).Jrt = Jpr.Jrt
               real_t Jpt[9];
               kernels::Mult(3,3,3, Jpr, Jrt, Jpt);
 
               // B = Jpt : H
               real_t B[9];
               DeviceMatrix M(B,3,3);
               ConstDeviceMatrix J(Jpt,3,3);
               for (int i = 0; i < DIM; i++)
               {
                  for (int j = 0; j < DIM; j++)
                  {
                     M(i,j) = 0.0;
                     for (int r = 0; r < DIM; r++)
                     {
                        for (int c = 0; c < DIM; c++)
                        {
                           M(i,j) += H(r,c,i,j,qx,qy,qz,e) * J(r,c);
                        }
                     }
                  }
               }
 
               // Y +=  DS . M^t += DSh . (Jrt . M^t)
               real_t A[9];
               kernels::MultABt(3,3,3, Jrt, B, A);
               kernels::internal::PushGrad<MQ1>(Q1D, qx,qy,qz, A, QQQ);
            }
         }
      }
      MFEM_SYNC_THREAD;
      kernels::internal::LoadBGt<MD1,MQ1>(D1D,Q1D,b,g,BG);
      kernels::internal::GradZt<MD1,MQ1>(D1D,Q1D,BG,QQQ,DQQ);
      kernels::internal::GradYt<MD1,MQ1>(D1D,Q1D,BG,DQQ,DDQ);
      kernels::internal::GradXt<MD1,MQ1>(D1D,Q1D,BG,DDQ,Y,e);
   });
}
 
void TMOP_Integrator::AddMultGradPA_3D(const Vector &R, Vector &C) const
{
   const int N = PA.ne;
   const int D1D = PA.maps->ndof;
   const int Q1D = PA.maps->nqpt;
   const int id = (D1D << 4 ) | Q1D;
   const DenseTensor &J = PA.Jtr;
   const Array<real_t> &B = PA.maps->B;
   const Array<real_t> &G = PA.maps->G;
   const Vector &H = PA.H;
 
   MFEM_LAUNCH_TMOP_KERNEL(AddMultGradPA_Kernel_3D,id,N,B,G,J,H,R,C);
}
 
} // namespace mfem