1 // Copyright (c) 2017-2026, 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 struct BuildContext *build_data = (struct BuildContext *)ctx; 18 19 // in[0] is Jacobians with shape [dim, dim, Q] 20 // in[1] is quadrature weights, size (Q) 21 const CeedScalar *w = in[1]; 22 CeedScalar(*q_data)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[0]; 23 24 // At every quadrature point, compute w/det(J).adj(J).adj(J)^T and store 25 // the symmetric part of the result. 26 switch (build_data->dim + 10 * build_data->space_dim) { 27 case 11: { 28 const CeedScalar(*J)[1][CEED_Q_VLA] = (const CeedScalar(*)[1][CEED_Q_VLA])in[0]; 29 30 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 31 // Mass 32 q_data[0][i] = w[i] * J[0][0][i]; 33 34 // Diffusion 35 q_data[1][i] = w[i] / J[0][0][i]; 36 } // End of Quadrature Point Loop 37 } break; 38 case 22: { 39 const CeedScalar(*J)[2][CEED_Q_VLA] = (const CeedScalar(*)[2][CEED_Q_VLA])in[0]; 40 41 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 42 // J: 0 2 q_data: 0 2 adj(J): J22 -J12 43 // 1 3 2 1 -J10 J00 44 const CeedScalar J00 = J[0][0][i]; 45 const CeedScalar J10 = J[0][1][i]; 46 const CeedScalar J01 = J[1][0][i]; 47 const CeedScalar J11 = J[1][1][i]; 48 const CeedScalar qw = w[i] / (J00 * J11 - J10 * J01); 49 50 // Mass 51 q_data[0][i] = w[i] * (J00 * J11 - J10 * J01); 52 53 // Diffusion 54 q_data[1][i] = qw * (J01 * J01 + J11 * J11); 55 q_data[2][i] = qw * (J00 * J00 + J10 * J10); 56 q_data[3][i] = -qw * (J00 * J01 + J10 * J11); 57 } // End of Quadrature Point Loop 58 } break; 59 case 33: { 60 const CeedScalar(*J)[3][CEED_Q_VLA] = (const CeedScalar(*)[3][CEED_Q_VLA])in[0]; 61 62 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 63 // Compute the adjoint 64 CeedScalar A[3][3]; 65 66 for (CeedInt j = 0; j < 3; j++) { 67 for (CeedInt k = 0; k < 3; k++) { 68 // Equivalent code with J as a VLA and no mod operations: 69 // A[k][j] = J[j+1][k+1]*J[j+2][k+2] - J[j+1][k+2]*J[j+2][k+1] 70 A[k][j] = 71 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]; 72 } 73 } 74 75 // Compute quadrature weight / det(J) 76 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]); 77 78 // Mass 79 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]); 80 81 // Diffusion 82 // Stored in Voigt convention 83 // 1 6 5 84 // 6 2 4 85 // 5 4 3 86 q_data[1][i] = qw * (A[0][0] * A[0][0] + A[0][1] * A[0][1] + A[0][2] * A[0][2]); 87 q_data[2][i] = qw * (A[1][0] * A[1][0] + A[1][1] * A[1][1] + A[1][2] * A[1][2]); 88 q_data[3][i] = qw * (A[2][0] * A[2][0] + A[2][1] * A[2][1] + A[2][2] * A[2][2]); 89 q_data[4][i] = qw * (A[1][0] * A[2][0] + A[1][1] * A[2][1] + A[1][2] * A[2][2]); 90 q_data[5][i] = qw * (A[0][0] * A[2][0] + A[0][1] * A[2][1] + A[0][2] * A[2][2]); 91 q_data[6][i] = qw * (A[0][0] * A[1][0] + A[0][1] * A[1][1] + A[0][2] * A[1][2]); 92 } // End of Quadrature Point Loop 93 } break; 94 } 95 return CEED_ERROR_SUCCESS; 96 } 97 98 /// libCEED Q-function for applying a mass + diffusion operator 99 CEED_QFUNCTION(apply_mass_diff)(void *ctx, const CeedInt Q, const CeedScalar *const *in, CeedScalar *const *out) { 100 struct BuildContext *build_data = (struct BuildContext *)ctx; 101 102 // in[1], out[1] solution values with shape [1, 1, Q] 103 // in[1], out[1] solution gradients with shape [dim, 1, Q] 104 // in[2] is quadrature data with shape [num_components, Q] 105 const CeedScalar(*q_data)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[2]; 106 107 switch (build_data->dim) { 108 case 1: { 109 const CeedScalar *u = in[0], *ug = in[1]; 110 CeedScalar *v = out[0], *vg = out[1]; 111 112 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 113 // Mass 114 v[i] = q_data[0][i] * u[i]; 115 116 // Diffusion 117 vg[i] = q_data[1][i] * ug[i]; 118 } // End of Quadrature Point Loop 119 } break; 120 case 2: { 121 const CeedScalar *u = in[0]; 122 const CeedScalar(*ug)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[1]; 123 CeedScalar *v = out[0]; 124 CeedScalar(*vg)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[1]; 125 126 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 127 // Mass 128 v[i] = q_data[0][i] * u[i]; 129 130 // Diffusion 131 // Read q_data (dXdxdXdx_T symmetric matrix) 132 // Stored in Voigt convention 133 // 1 3 134 // 23 2 135 const CeedScalar dXdxdXdx_T[2][2] = { 136 {q_data[1][i], q_data[3][i]}, 137 {q_data[3][i], q_data[2][i]} 138 }; 139 140 // j = direction of vg 141 for (int j = 0; j < 2; j++) vg[j][i] = (ug[0][i] * dXdxdXdx_T[0][j] + ug[1][i] * dXdxdXdx_T[1][j]); 142 } // End of Quadrature Point Loop 143 } break; 144 case 3: { 145 const CeedScalar *u = in[0]; 146 const CeedScalar(*ug)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[1]; 147 CeedScalar *v = out[0]; 148 CeedScalar(*vg)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[1]; 149 150 CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 151 // Mass 152 v[i] = q_data[0][i] * u[i]; 153 154 // Diffusion 155 // Read q_data (dXdxdXdx_T symmetric matrix) 156 // Stored in Voigt convention 157 // 1 6 5 158 // 6 2 4 159 // 5 4 3 160 const CeedScalar dXdxdXdx_T[3][3] = { 161 {q_data[1][i], q_data[6][i], q_data[5][i]}, 162 {q_data[6][i], q_data[2][i], q_data[4][i]}, 163 {q_data[5][i], q_data[4][i], q_data[3][i]} 164 }; 165 166 // j = direction of vg 167 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]); 168 } // End of Quadrature Point Loop 169 } break; 170 } 171 return CEED_ERROR_SUCCESS; 172 } 173