MFEM  v4.5.2
Finite element discretization library
tmop_pa_h3s_c0.cpp
Go to the documentation of this file.
1 // Copyright (c) 2010-2023, Lawrence Livermore National Security, LLC. Produced
2 // at the Lawrence Livermore National Laboratory. All Rights reserved. See files
3 // LICENSE and NOTICE for details. LLNL-CODE-806117.
4 //
5 // This file is part of the MFEM library. For more information and source code
6 // availability visit https://mfem.org.
7 //
8 // MFEM is free software; you can redistribute it and/or modify it under the
9 // terms of the BSD-3 license. We welcome feedback and contributions, see file
10 // CONTRIBUTING.md for details.
11 
12 #include "../tmop.hpp"
13 #include "tmop_pa.hpp"
14 #include "../linearform.hpp"
15 #include "../../general/forall.hpp"
16 #include "../../linalg/kernels.hpp"
17 
18 namespace mfem
19 {
20 
21 MFEM_REGISTER_TMOP_KERNELS(void, SetupGradPA_Kernel_C0_3D,
22  const double lim_normal,
23  const Vector &lim_dist,
24  const Vector &c0_,
25  const int NE,
26  const DenseTensor &j_,
27  const Array<double> &w_,
28  const Array<double> &b_,
29  const Array<double> &bld_,
30  const Vector &x0_,
31  const Vector &x1_,
32  Vector &h0_,
33  const bool exp_lim,
34  const int d1d,
35  const int q1d)
36 {
37  constexpr int DIM = 3;
38  const int D1D = T_D1D ? T_D1D : d1d;
39  const int Q1D = T_Q1D ? T_Q1D : q1d;
40 
41  const bool const_c0 = c0_.Size() == 1;
42  const auto C0 = const_c0 ?
43  Reshape(c0_.Read(), 1, 1, 1, 1) :
44  Reshape(c0_.Read(), Q1D, Q1D, Q1D, NE);
45  const auto LD = Reshape(lim_dist.Read(), D1D, D1D, D1D, NE);
46  const auto J = Reshape(j_.Read(), DIM, DIM, Q1D, Q1D, Q1D, NE);
47  const auto b = Reshape(b_.Read(), Q1D, D1D);
48  const auto bld = Reshape(bld_.Read(), Q1D, D1D);
49  const auto W = Reshape(w_.Read(), Q1D, Q1D, Q1D);
50  const auto X0 = Reshape(x0_.Read(), D1D, D1D, D1D, DIM, NE);
51  const auto X1 = Reshape(x1_.Read(), D1D, D1D, D1D, DIM, NE);
52 
53  auto H0 = Reshape(h0_.Write(), DIM, DIM, Q1D, Q1D, Q1D, NE);
54 
55  MFEM_FORALL_3D(e, NE, Q1D, Q1D, Q1D,
56  {
57  constexpr int DIM = 3;
58  const int D1D = T_D1D ? T_D1D : d1d;
59  const int Q1D = T_Q1D ? T_Q1D : q1d;
60  constexpr int MQ1 = T_Q1D ? T_Q1D : T_MAX;
61  constexpr int MD1 = T_D1D ? T_D1D : T_MAX;
62  constexpr int MDQ = (MQ1 > MD1) ? MQ1 : MD1;
63 
64  MFEM_SHARED double B[MQ1*MD1];
65  MFEM_SHARED double sBLD[MQ1*MD1];
66  kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,bld,sBLD);
67  ConstDeviceMatrix BLD(sBLD, D1D, Q1D);
68 
69  MFEM_SHARED double sm0[MDQ*MDQ*MDQ];
70  MFEM_SHARED double sm1[MDQ*MDQ*MDQ];
71  DeviceCube DDD(sm0, MD1,MD1,MD1);
72  DeviceCube DDQ(sm1, MD1,MD1,MQ1);
73  DeviceCube DQQ(sm0, MD1,MQ1,MQ1);
74  DeviceCube QQQ(sm1, MQ1,MQ1,MQ1);
75 
76  MFEM_SHARED double DDD0[3][MD1*MD1*MD1];
77  MFEM_SHARED double DDQ0[3][MD1*MD1*MQ1];
78  MFEM_SHARED double DQQ0[3][MD1*MQ1*MQ1];
79  MFEM_SHARED double QQQ0[3][MQ1*MQ1*MQ1];
80 
81  MFEM_SHARED double DDD1[3][MD1*MD1*MD1];
82  MFEM_SHARED double DDQ1[3][MD1*MD1*MQ1];
83  MFEM_SHARED double DQQ1[3][MD1*MQ1*MQ1];
84  MFEM_SHARED double QQQ1[3][MQ1*MQ1*MQ1];
85 
86  kernels::internal::LoadX(e,D1D,LD,DDD);
87  kernels::internal::LoadX<MD1>(e,D1D,X0,DDD0);
88  kernels::internal::LoadX<MD1>(e,D1D,X1,DDD1);
89 
90  kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,b,B);
91 
92  kernels::internal::EvalX(D1D,Q1D,BLD,DDD,DDQ);
93  kernels::internal::EvalY(D1D,Q1D,BLD,DDQ,DQQ);
94  kernels::internal::EvalZ(D1D,Q1D,BLD,DQQ,QQQ);
95 
96  kernels::internal::EvalX<MD1,MQ1>(D1D,Q1D,B,DDD0,DDQ0);
97  kernels::internal::EvalY<MD1,MQ1>(D1D,Q1D,B,DDQ0,DQQ0);
98  kernels::internal::EvalZ<MD1,MQ1>(D1D,Q1D,B,DQQ0,QQQ0);
99 
100  kernels::internal::EvalX<MD1,MQ1>(D1D,Q1D,B,DDD1,DDQ1);
101  kernels::internal::EvalY<MD1,MQ1>(D1D,Q1D,B,DDQ1,DQQ1);
102  kernels::internal::EvalZ<MD1,MQ1>(D1D,Q1D,B,DQQ1,QQQ1);
103 
104  MFEM_FOREACH_THREAD(qz,z,Q1D)
105  {
106  MFEM_FOREACH_THREAD(qy,y,Q1D)
107  {
108  MFEM_FOREACH_THREAD(qx,x,Q1D)
109  {
110  const double *Jtr = &J(0,0,qx,qy,qz,e);
111  const double detJtr = kernels::Det<3>(Jtr);
112  const double weight = W(qx,qy,qz) * detJtr;
113  const double coeff0 = const_c0 ? C0(0,0,0,0) : C0(qx,qy,qz,e);
114  const double weight_m = weight * lim_normal * coeff0;
115 
116  double D, p0[3], p1[3];
117  kernels::internal::PullEval(qx,qy,qz,QQQ,D);
118  kernels::internal::PullEval<MQ1>(Q1D,qx,qy,qz,QQQ0,p0);
119  kernels::internal::PullEval<MQ1>(Q1D,qx,qy,qz,QQQ1,p1);
120 
121  const double dist = D; // GetValues, default comp set to 0
122 
123  // lim_func->Eval_d2(p1, p0, d_vals(q), grad_grad);
124 
125  double grad_grad[9];
126 
127  if (!exp_lim)
128  {
129  // d2.Diag(1.0 / (dist * dist), x.Size());
130  const double c = 1.0 / (dist * dist);
131  kernels::Diag<3>(c, grad_grad);
132  }
133  else
134  {
135  double tmp[3];
136  kernels::Subtract<3>(1.0, p1, p0, tmp);
137  double dsq = kernels::DistanceSquared<3>(p1,p0);
138  double dist_squared = dist*dist;
139  double dist_squared_squared = dist_squared*dist_squared;
140  double f = exp(10.0*((dsq / dist_squared)-1.0));
141  grad_grad[0] = ((400.0*tmp[0]*tmp[0]*f)/dist_squared_squared)+
142  (20.0*f/dist_squared);
143  grad_grad[1] = (400.0*tmp[0]*tmp[1]*f)/dist_squared_squared;
144  grad_grad[2] = (400.0*tmp[0]*tmp[2]*f)/dist_squared_squared;
145  grad_grad[3] = grad_grad[1];
146  grad_grad[4] = ((400.0*tmp[1]*tmp[1]*f)/dist_squared_squared)+
147  (20.0*f/dist_squared);
148  grad_grad[5] = (400.0*tmp[1]*tmp[2]*f)/dist_squared_squared;
149  grad_grad[6] = grad_grad[2];
150  grad_grad[7] = grad_grad[5];
151  grad_grad[8] = ((400.0*tmp[2]*tmp[2]*f)/dist_squared_squared)+
152  (20.0*f/dist_squared);
153  }
154  ConstDeviceMatrix gg(grad_grad,DIM,DIM);
155 
156  for (int i = 0; i < DIM; i++)
157  {
158  for (int j = 0; j < DIM; j++)
159  {
160  H0(i,j,qx,qy,qz,e) = weight_m * gg(i,j);
161  }
162  }
163  }
164  }
165  }
166  });
167 }
168 
170 {
171  const int N = PA.ne;
172  const int D1D = PA.maps_lim->ndof;
173  const int Q1D = PA.maps_lim->nqpt;
174  const int id = (D1D << 4 ) | Q1D;
175  const double ln = lim_normal;
176  const Vector &LD = PA.LD;
177  const DenseTensor &J = PA.Jtr;
178  const Array<double> &W = PA.ir->GetWeights();
179  const Array<double> &B = PA.maps->B;
180  const Array<double> &BLD = PA.maps_lim->B;
181  const Vector &C0 = PA.C0;
182  const Vector &X0 = PA.X0;
183  Vector &H0 = PA.H0;
184 
185  auto el = dynamic_cast<TMOP_ExponentialLimiter *>(lim_func);
186  const bool exp_lim = (el) ? true : false;
187 
188  MFEM_LAUNCH_TMOP_KERNEL(SetupGradPA_Kernel_C0_3D,id,ln,LD,C0,N,J,W,B,BLD,X0,X,
189  H0,exp_lim);
190 }
191 
192 } // namespace mfem
const T * Read(bool on_dev=true) const
Shortcut for mfem::Read(a.GetMemory(), a.Size(), on_dev).
Definition: array.hpp:307
struct mfem::TMOP_Integrator::@23 PA
int Size() const
Returns the size of the vector.
Definition: vector.hpp:199
virtual const double * Read(bool on_dev=true) const
Shortcut for mfem::Read(vec.GetMemory(), vec.Size(), on_dev).
Definition: vector.hpp:448
constexpr int DIM
const double * Read(bool on_dev=true) const
Shortcut for mfem::Read( GetMemory(), TotalSize(), on_dev).
Definition: densemat.hpp:1112
double f(const Vector &xvec)
Definition: lor_mms.hpp:32
virtual double * Write(bool on_dev=true)
Shortcut for mfem::Write(vec.GetMemory(), vec.Size(), on_dev).
Definition: vector.hpp:456
double b
Definition: lissajous.cpp:42
MFEM_REGISTER_TMOP_KERNELS(void, DatcSize, const int NE, const int ncomp, const int sizeidx, const DenseMatrix &w_, const Array< double > &b_, const Vector &x_, DenseTensor &j_, const int d1d, const int q1d)
Definition: tmop_pa_da3.cpp:20
void AssembleGradPA_C0_3D(const Vector &) const
A basic generic Tensor class, appropriate for use on the GPU.
Definition: dtensor.hpp:81
Exponential limiter function in TMOP_Integrator.
Definition: tmop.hpp:1124
Vector data type.
Definition: vector.hpp:60
MFEM_HOST_DEVICE DeviceTensor< sizeof...(Dims), T > Reshape(T *ptr, Dims... dims)
Wrap a pointer as a DeviceTensor with automatically deduced template parameters.
Definition: dtensor.hpp:131
Rank 3 tensor (array of matrices)
Definition: densemat.hpp:978
TMOP_LimiterFunction * lim_func
Definition: tmop.hpp:1665