1 // Copyright (c) 2010-2021, 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 "../../general/forall.hpp"
15 #include "../../linalg/kernels.hpp"
16
17 namespace mfem
18 {
19
MFEM_REGISTER_TMOP_KERNELS(void,AddMultGradPA_Kernel_3D,const int NE,const Array<double> & b_,const Array<double> & g_,const DenseTensor & j_,const Vector & h_,const Vector & x_,Vector & y_,const int d1d,const int q1d)20 MFEM_REGISTER_TMOP_KERNELS(void, AddMultGradPA_Kernel_3D,
21 const int NE,
22 const Array<double> &b_,
23 const Array<double> &g_,
24 const DenseTensor &j_,
25 const Vector &h_,
26 const Vector &x_,
27 Vector &y_,
28 const int d1d,
29 const int q1d)
30 {
31 constexpr int DIM = 3;
32 const int D1D = T_D1D ? T_D1D : d1d;
33 const int Q1D = T_Q1D ? T_Q1D : q1d;
34
35 const auto b = Reshape(b_.Read(), Q1D, D1D);
36 const auto g = Reshape(g_.Read(), Q1D, D1D);
37 const auto J = Reshape(j_.Read(), DIM, DIM, Q1D, Q1D, Q1D, NE);
38 const auto X = Reshape(x_.Read(), D1D, D1D, D1D, DIM, NE);
39 const auto H = Reshape(h_.Read(), DIM, DIM, DIM, DIM, Q1D, Q1D, Q1D, NE);
40 auto Y = Reshape(y_.ReadWrite(), D1D, D1D, D1D, DIM, NE);
41
42 MFEM_FORALL_3D(e, NE, Q1D, Q1D, Q1D,
43 {
44 constexpr int DIM = 3;
45 const int D1D = T_D1D ? T_D1D : d1d;
46 const int Q1D = T_Q1D ? T_Q1D : q1d;
47 constexpr int MQ1 = T_Q1D ? T_Q1D : T_MAX;
48 constexpr int MD1 = T_D1D ? T_D1D : T_MAX;
49
50 MFEM_SHARED double BG[2][MQ1*MD1];
51 MFEM_SHARED double DDD[3][MD1*MD1*MD1];
52 MFEM_SHARED double DDQ[9][MD1*MD1*MQ1];
53 MFEM_SHARED double DQQ[9][MD1*MQ1*MQ1];
54 MFEM_SHARED double QQQ[9][MQ1*MQ1*MQ1];
55
56 kernels::internal::LoadX<MD1>(e,D1D,X,DDD);
57 kernels::internal::LoadBG<MD1,MQ1>(D1D,Q1D,b,g,BG);
58
59 kernels::internal::GradX<MD1,MQ1>(D1D,Q1D,BG,DDD,DDQ);
60 kernels::internal::GradY<MD1,MQ1>(D1D,Q1D,BG,DDQ,DQQ);
61 kernels::internal::GradZ<MD1,MQ1>(D1D,Q1D,BG,DQQ,QQQ);
62
63 MFEM_FOREACH_THREAD(qz,z,Q1D)
64 {
65 MFEM_FOREACH_THREAD(qy,y,Q1D)
66 {
67 MFEM_FOREACH_THREAD(qx,x,Q1D)
68 {
69 const double *Jtr = &J(0,0,qx,qy,qz,e);
70
71 // Jrt = Jtr^{-1}
72 double Jrt[9];
73 kernels::CalcInverse<3>(Jtr, Jrt);
74
75 // Jpr = X^T.DSh
76 double Jpr[9];
77 kernels::internal::PullGrad<MQ1>(Q1D, qx,qy,qz, QQQ, Jpr);
78
79 // Jpt = X^T.DS = (X^T.DSh).Jrt = Jpr.Jrt
80 double Jpt[9];
81 kernels::Mult(3,3,3, Jpr, Jrt, Jpt);
82
83 // B = Jpt : H
84 double B[9];
85 DeviceMatrix M(B,3,3);
86 ConstDeviceMatrix J(Jpt,3,3);
87 for (int i = 0; i < DIM; i++)
88 {
89 for (int j = 0; j < DIM; j++)
90 {
91 M(i,j) = 0.0;
92 for (int r = 0; r < DIM; r++)
93 {
94 for (int c = 0; c < DIM; c++)
95 {
96 M(i,j) += H(r,c,i,j,qx,qy,qz,e) * J(r,c);
97 }
98 }
99 }
100 }
101
102 // Y += DS . M^t += DSh . (Jrt . M^t)
103 double A[9];
104 kernels::MultABt(3,3,3, Jrt, B, A);
105 kernels::internal::PushGrad<MQ1>(Q1D, qx,qy,qz, A, QQQ);
106 }
107 }
108 }
109 MFEM_SYNC_THREAD;
110 kernels::internal::LoadBGt<MD1,MQ1>(D1D,Q1D,b,g,BG);
111 kernels::internal::GradZt<MD1,MQ1>(D1D,Q1D,BG,QQQ,DQQ);
112 kernels::internal::GradYt<MD1,MQ1>(D1D,Q1D,BG,DQQ,DDQ);
113 kernels::internal::GradXt<MD1,MQ1>(D1D,Q1D,BG,DDQ,Y,e);
114 });
115 }
116
AddMultGradPA_3D(const Vector & R,Vector & C) const117 void TMOP_Integrator::AddMultGradPA_3D(const Vector &R, Vector &C) const
118 {
119 const int N = PA.ne;
120 const int D1D = PA.maps->ndof;
121 const int Q1D = PA.maps->nqpt;
122 const int id = (D1D << 4 ) | Q1D;
123 const DenseTensor &J = PA.Jtr;
124 const Array<double> &B = PA.maps->B;
125 const Array<double> &G = PA.maps->G;
126 const Vector &H = PA.H;
127
128 MFEM_LAUNCH_TMOP_KERNEL(AddMultGradPA_Kernel_3D,id,N,B,G,J,H,R,C);
129 }
130
131 } // namespace mfem
132