1 // Copyright (c) 2017-2025, Lawrence Livermore National Security, LLC and other CEED contributors. 2 // All Rights Reserved. See the top-level LICENSE and NOTICE files for details. 3 // 4 // SPDX-License-Identifier: BSD-2-Clause 5 // 6 // This file is part of CEED: http://github.com/ceed 7 8 #include <ceed/types.h> 9 10 /// A structure used to pass additional data to f_build_mass_diff 11 struct BuildContext { 12 CeedInt dim, space_dim; 13 }; 14 15 /// libCEED Q-function for building quadrature data for a mass + diffusion operator 16 CEED_QFUNCTION(build_mass_diff)(void *ctx, const CeedInt Q, const CeedScalar *const *in, CeedScalar *const *out) { 17 // in[0] is Jacobians with shape [dim, dim, Q] 18 // in[1] is quadrature weights, size (Q) 19 const CeedScalar *w = in[1]; 20 CeedScalar(*q_data)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[0]; 21 struct BuildContext *build_data = (struct BuildContext *)ctx; 22 23 // At every quadrature point, compute w/det(J).adj(J).adj(J)^T and store 24 // the symmetric part of the result. 25 switch (build_data->dim + 10 * build_data->space_dim) { 26 case 11: { 27 const CeedScalar(*J)[1][CEED_Q_VLA] = (const CeedScalar(*)[1][CEED_Q_VLA])in[0]; 28 29 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 30 // Mass 31 q_data[0][i] = w[i] * J[0][0][i]; 32 33 // Diffusion 34 q_data[1][i] = w[i] / J[0][0][i]; 35 } // End of Quadrature Point Loop 36 } break; 37 case 22: { 38 const CeedScalar(*J)[2][CEED_Q_VLA] = (const CeedScalar(*)[2][CEED_Q_VLA])in[0]; 39 40 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 41 // J: 0 2 q_data: 0 2 adj(J): J22 -J12 42 // 1 3 2 1 -J10 J00 43 const CeedScalar J00 = J[0][0][i]; 44 const CeedScalar J10 = J[0][1][i]; 45 const CeedScalar J01 = J[1][0][i]; 46 const CeedScalar J11 = J[1][1][i]; 47 const CeedScalar qw = w[i] / (J00 * J11 - J10 * J01); 48 49 // Mass 50 q_data[0][i] = w[i] * (J00 * J11 - J10 * J01); 51 52 // Diffusion 53 q_data[1][i] = qw * (J01 * J01 + J11 * J11); 54 q_data[2][i] = qw * (J00 * J00 + J10 * J10); 55 q_data[3][i] = -qw * (J00 * J01 + J10 * J11); 56 } // End of Quadrature Point Loop 57 } break; 58 case 33: { 59 const CeedScalar(*J)[3][CEED_Q_VLA] = (const CeedScalar(*)[3][CEED_Q_VLA])in[0]; 60 61 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 62 // Compute the adjoint 63 CeedScalar A[3][3]; 64 65 for (CeedInt j = 0; j < 3; j++) { 66 for (CeedInt k = 0; k < 3; k++) { 67 // Equivalent code with J as a VLA and no mod operations: 68 // A[k][j] = J[j+1][k+1]*J[j+2][k+2] - J[j+1][k+2]*J[j+2][k+1] 69 A[k][j] = 70 J[(k + 1) % 3][(j + 1) % 3][i] * J[(k + 2) % 3][(j + 2) % 3][i] - J[(k + 2) % 3][(j + 1) % 3][i] * J[(k + 1) % 3][(j + 2) % 3][i]; 71 } 72 } 73 74 // Compute quadrature weight / det(J) 75 const CeedScalar qw = w[i] / (J[0][0][i] * A[0][0] + J[0][1][i] * A[0][1] + J[0][2][i] * A[0][2]); 76 77 // Mass 78 q_data[0][i] = w[i] * (J[0][0][i] * A[0][0] + J[0][1][i] * A[0][1] + J[0][2][i] * A[0][2]); 79 80 // Diffusion 81 // Stored in Voigt convention 82 // 1 6 5 83 // 6 2 4 84 // 5 4 3 85 q_data[1][i] = qw * (A[0][0] * A[0][0] + A[0][1] * A[0][1] + A[0][2] * A[0][2]); 86 q_data[2][i] = qw * (A[1][0] * A[1][0] + A[1][1] * A[1][1] + A[1][2] * A[1][2]); 87 q_data[3][i] = qw * (A[2][0] * A[2][0] + A[2][1] * A[2][1] + A[2][2] * A[2][2]); 88 q_data[4][i] = qw * (A[1][0] * A[2][0] + A[1][1] * A[2][1] + A[1][2] * A[2][2]); 89 q_data[5][i] = qw * (A[0][0] * A[2][0] + A[0][1] * A[2][1] + A[0][2] * A[2][2]); 90 q_data[6][i] = qw * (A[0][0] * A[1][0] + A[0][1] * A[1][1] + A[0][2] * A[1][2]); 91 } // End of Quadrature Point Loop 92 } break; 93 } 94 return CEED_ERROR_SUCCESS; 95 } 96 97 /// libCEED Q-function for applying a mass + diffusion operator 98 CEED_QFUNCTION(apply_mass_diff)(void *ctx, const CeedInt Q, const CeedScalar *const *in, CeedScalar *const *out) { 99 struct BuildContext *build_data = (struct BuildContext *)ctx; 100 // in[1], out[1] solution values with shape [1, 1, Q] 101 // in[1], out[1] solution gradients with shape [dim, 1, Q] 102 // in[2] is quadrature data with shape [num_components, Q] 103 const CeedScalar(*q_data)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[2]; 104 105 switch (build_data->dim) { 106 case 1: { 107 const CeedScalar *u = in[0], *ug = in[1]; 108 CeedScalar *v = out[0], *vg = out[1]; 109 110 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 111 // Mass 112 v[i] = q_data[0][i] * u[i]; 113 114 // Diffusion 115 vg[i] = q_data[1][i] * ug[i]; 116 } // End of Quadrature Point Loop 117 } break; 118 case 2: { 119 const CeedScalar *u = in[0]; 120 const CeedScalar(*ug)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[1]; 121 CeedScalar *v = out[0]; 122 CeedScalar(*vg)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[1]; 123 124 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 125 // Mass 126 v[i] = q_data[0][i] * u[i]; 127 128 // Diffusion 129 // Read q_data (dXdxdXdx_T symmetric matrix) 130 // Stored in Voigt convention 131 // 1 3 132 // 23 2 133 const CeedScalar dXdxdXdx_T[2][2] = { 134 {q_data[1][i], q_data[3][i]}, 135 {q_data[3][i], q_data[2][i]} 136 }; 137 138 // j = direction of vg 139 for (int j = 0; j < 2; j++) vg[j][i] = (ug[0][i] * dXdxdXdx_T[0][j] + ug[1][i] * dXdxdXdx_T[1][j]); 140 } // End of Quadrature Point Loop 141 } break; 142 case 3: { 143 const CeedScalar *u = in[0]; 144 const CeedScalar(*ug)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[1]; 145 CeedScalar *v = out[0]; 146 CeedScalar(*vg)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[1]; 147 148 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 149 // Mass 150 v[i] = q_data[0][i] * u[i]; 151 152 // Diffusion 153 // Read q_data (dXdxdXdx_T symmetric matrix) 154 // Stored in Voigt convention 155 // 1 6 5 156 // 6 2 4 157 // 5 4 3 158 const CeedScalar dXdxdXdx_T[3][3] = { 159 {q_data[1][i], q_data[6][i], q_data[5][i]}, 160 {q_data[6][i], q_data[2][i], q_data[4][i]}, 161 {q_data[5][i], q_data[4][i], q_data[3][i]} 162 }; 163 164 // j = direction of vg 165 for (int j = 0; j < 3; j++) vg[j][i] = (ug[0][i] * dXdxdXdx_T[0][j] + ug[1][i] * dXdxdXdx_T[1][j] + ug[2][i] * dXdxdXdx_T[2][j]); 166 } // End of Quadrature Point Loop 167 } break; 168 } 169 return CEED_ERROR_SUCCESS; 170 } 171