1*7b3ff069SJeremy L Thompson // Copyright (c) 2017-2025, Lawrence Livermore National Security, LLC and other CEED contributors. 2*7b3ff069SJeremy L Thompson // All Rights Reserved. See the top-level LICENSE and NOTICE files for details. 3*7b3ff069SJeremy L Thompson // 4*7b3ff069SJeremy L Thompson // SPDX-License-Identifier: BSD-2-Clause 5*7b3ff069SJeremy L Thompson // 6*7b3ff069SJeremy L Thompson // This file is part of CEED: http://github.com/ceed 7*7b3ff069SJeremy L Thompson #pragma once 8*7b3ff069SJeremy L Thompson 9*7b3ff069SJeremy L Thompson #include <ceed/types.h> 10*7b3ff069SJeremy L Thompson #include "ex-common.h" 11*7b3ff069SJeremy L Thompson 12*7b3ff069SJeremy L Thompson /// libCEED Q-function for building quadrature data for a diffusion operator 13*7b3ff069SJeremy L Thompson CEED_QFUNCTION(build_diff)(void *ctx, const CeedInt Q, const CeedScalar *const *in, CeedScalar *const *out) { 14*7b3ff069SJeremy L Thompson // in[0] is Jacobians with shape [dim, dim, Q] 15*7b3ff069SJeremy L Thompson // in[1] is quadrature weights, size (Q) 16*7b3ff069SJeremy L Thompson const CeedScalar *w = in[1]; 17*7b3ff069SJeremy L Thompson CeedScalar(*q_data)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[0]; 18*7b3ff069SJeremy L Thompson struct BuildContext *build_data = (struct BuildContext *)ctx; 19*7b3ff069SJeremy L Thompson 20*7b3ff069SJeremy L Thompson // At every quadrature point, compute w/det(J).adj(J).adj(J)^T and store 21*7b3ff069SJeremy L Thompson // the symmetric part of the result. 22*7b3ff069SJeremy L Thompson switch (build_data->dim + 10 * build_data->space_dim) { 23*7b3ff069SJeremy L Thompson case 11: { 24*7b3ff069SJeremy L Thompson const CeedScalar(*J)[1][CEED_Q_VLA] = (const CeedScalar(*)[1][CEED_Q_VLA])in[0]; 25*7b3ff069SJeremy L Thompson 26*7b3ff069SJeremy L Thompson CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { q_data[0][i] = w[i] / J[0][0][i]; } // End of Quadrature Point Loop 27*7b3ff069SJeremy L Thompson } break; 28*7b3ff069SJeremy L Thompson case 22: { 29*7b3ff069SJeremy L Thompson const CeedScalar(*J)[2][CEED_Q_VLA] = (const CeedScalar(*)[2][CEED_Q_VLA])in[0]; 30*7b3ff069SJeremy L Thompson 31*7b3ff069SJeremy L Thompson CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 32*7b3ff069SJeremy L Thompson // J: 0 2 q_data: 0 2 adj(J): J11 -J01 33*7b3ff069SJeremy L Thompson // 1 3 2 1 -J10 J00 34*7b3ff069SJeremy L Thompson const CeedScalar J00 = J[0][0][i]; 35*7b3ff069SJeremy L Thompson const CeedScalar J10 = J[0][1][i]; 36*7b3ff069SJeremy L Thompson const CeedScalar J01 = J[1][0][i]; 37*7b3ff069SJeremy L Thompson const CeedScalar J11 = J[1][1][i]; 38*7b3ff069SJeremy L Thompson const CeedScalar qw = w[i] / (J00 * J11 - J10 * J01); 39*7b3ff069SJeremy L Thompson 40*7b3ff069SJeremy L Thompson q_data[0][i] = qw * (J01 * J01 + J11 * J11); 41*7b3ff069SJeremy L Thompson q_data[1][i] = qw * (J00 * J00 + J10 * J10); 42*7b3ff069SJeremy L Thompson q_data[2][i] = -qw * (J00 * J01 + J10 * J11); 43*7b3ff069SJeremy L Thompson } // End of Quadrature Point Loop 44*7b3ff069SJeremy L Thompson } break; 45*7b3ff069SJeremy L Thompson case 33: { 46*7b3ff069SJeremy L Thompson const CeedScalar(*J)[3][CEED_Q_VLA] = (const CeedScalar(*)[3][CEED_Q_VLA])in[0]; 47*7b3ff069SJeremy L Thompson 48*7b3ff069SJeremy L Thompson CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 49*7b3ff069SJeremy L Thompson // Compute the adjoint 50*7b3ff069SJeremy L Thompson CeedScalar A[3][3]; 51*7b3ff069SJeremy L Thompson 52*7b3ff069SJeremy L Thompson for (CeedInt j = 0; j < 3; j++) { 53*7b3ff069SJeremy L Thompson for (CeedInt k = 0; k < 3; k++) { 54*7b3ff069SJeremy L Thompson // Equivalent code with J as a VLA and no mod operations: 55*7b3ff069SJeremy L Thompson // A[k][j] = J[j+1][k+1]*J[j+2][k+2] - J[j+1][k+2]*J[j+2][k+1] 56*7b3ff069SJeremy L Thompson A[k][j] = 57*7b3ff069SJeremy L Thompson 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]; 58*7b3ff069SJeremy L Thompson } 59*7b3ff069SJeremy L Thompson } 60*7b3ff069SJeremy L Thompson 61*7b3ff069SJeremy L Thompson // Compute quadrature weight / det(J) 62*7b3ff069SJeremy L Thompson 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]); 63*7b3ff069SJeremy L Thompson 64*7b3ff069SJeremy L Thompson // Compute geometric factors 65*7b3ff069SJeremy L Thompson // Stored in Voigt convention 66*7b3ff069SJeremy L Thompson // 0 5 4 67*7b3ff069SJeremy L Thompson // 5 1 3 68*7b3ff069SJeremy L Thompson // 4 3 2 69*7b3ff069SJeremy L Thompson q_data[0][i] = qw * (A[0][0] * A[0][0] + A[0][1] * A[0][1] + A[0][2] * A[0][2]); 70*7b3ff069SJeremy L Thompson q_data[1][i] = qw * (A[1][0] * A[1][0] + A[1][1] * A[1][1] + A[1][2] * A[1][2]); 71*7b3ff069SJeremy L Thompson q_data[2][i] = qw * (A[2][0] * A[2][0] + A[2][1] * A[2][1] + A[2][2] * A[2][2]); 72*7b3ff069SJeremy L Thompson q_data[3][i] = qw * (A[1][0] * A[2][0] + A[1][1] * A[2][1] + A[1][2] * A[2][2]); 73*7b3ff069SJeremy L Thompson q_data[4][i] = qw * (A[0][0] * A[2][0] + A[0][1] * A[2][1] + A[0][2] * A[2][2]); 74*7b3ff069SJeremy L Thompson q_data[5][i] = qw * (A[0][0] * A[1][0] + A[0][1] * A[1][1] + A[0][2] * A[1][2]); 75*7b3ff069SJeremy L Thompson } // End of Quadrature Point Loop 76*7b3ff069SJeremy L Thompson } break; 77*7b3ff069SJeremy L Thompson } 78*7b3ff069SJeremy L Thompson return CEED_ERROR_SUCCESS; 79*7b3ff069SJeremy L Thompson } 80*7b3ff069SJeremy L Thompson 81*7b3ff069SJeremy L Thompson /// libCEED Q-function for applying a diff operator 82*7b3ff069SJeremy L Thompson CEED_QFUNCTION(apply_diff)(void *ctx, const CeedInt Q, const CeedScalar *const *in, CeedScalar *const *out) { 83*7b3ff069SJeremy L Thompson struct BuildContext *build_data = (struct BuildContext *)ctx; 84*7b3ff069SJeremy L Thompson // in[0], out[0] solution gradients with shape [dim, 1, Q] 85*7b3ff069SJeremy L Thompson // in[1] is quadrature data with shape [num_components, Q] 86*7b3ff069SJeremy L Thompson const CeedScalar(*q_data)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[1]; 87*7b3ff069SJeremy L Thompson 88*7b3ff069SJeremy L Thompson switch (build_data->dim) { 89*7b3ff069SJeremy L Thompson case 1: { 90*7b3ff069SJeremy L Thompson const CeedScalar *ug = in[0]; 91*7b3ff069SJeremy L Thompson CeedScalar *vg = out[0]; 92*7b3ff069SJeremy L Thompson 93*7b3ff069SJeremy L Thompson CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { vg[i] = ug[i] * q_data[0][i]; } // End of Quadrature Point Loop 94*7b3ff069SJeremy L Thompson } break; 95*7b3ff069SJeremy L Thompson case 2: { 96*7b3ff069SJeremy L Thompson const CeedScalar(*ug)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[0]; 97*7b3ff069SJeremy L Thompson CeedScalar(*vg)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[0]; 98*7b3ff069SJeremy L Thompson 99*7b3ff069SJeremy L Thompson CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 100*7b3ff069SJeremy L Thompson // Read q_data (dXdxdXdx_T symmetric matrix) 101*7b3ff069SJeremy L Thompson // Stored in Voigt convention 102*7b3ff069SJeremy L Thompson // 0 2 103*7b3ff069SJeremy L Thompson // 2 1 104*7b3ff069SJeremy L Thompson const CeedScalar dXdxdXdx_T[2][2] = { 105*7b3ff069SJeremy L Thompson {q_data[0][i], q_data[2][i]}, 106*7b3ff069SJeremy L Thompson {q_data[2][i], q_data[1][i]} 107*7b3ff069SJeremy L Thompson }; 108*7b3ff069SJeremy L Thompson 109*7b3ff069SJeremy L Thompson // j = direction of vg 110*7b3ff069SJeremy L Thompson for (int j = 0; j < 2; j++) vg[j][i] = (ug[0][i] * dXdxdXdx_T[0][j] + ug[1][i] * dXdxdXdx_T[1][j]); 111*7b3ff069SJeremy L Thompson } // End of Quadrature Point Loop 112*7b3ff069SJeremy L Thompson } break; 113*7b3ff069SJeremy L Thompson case 3: { 114*7b3ff069SJeremy L Thompson const CeedScalar(*ug)[CEED_Q_VLA] = (const CeedScalar(*)[CEED_Q_VLA])in[0]; 115*7b3ff069SJeremy L Thompson CeedScalar(*vg)[CEED_Q_VLA] = (CeedScalar(*)[CEED_Q_VLA])out[0]; 116*7b3ff069SJeremy L Thompson 117*7b3ff069SJeremy L Thompson CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++) { 118*7b3ff069SJeremy L Thompson // Read q_data (dXdxdXdx_T symmetric matrix) 119*7b3ff069SJeremy L Thompson // Stored in Voigt convention 120*7b3ff069SJeremy L Thompson // 0 5 4 121*7b3ff069SJeremy L Thompson // 5 1 3 122*7b3ff069SJeremy L Thompson // 4 3 2 123*7b3ff069SJeremy L Thompson const CeedScalar dXdxdXdx_T[3][3] = { 124*7b3ff069SJeremy L Thompson {q_data[0][i], q_data[5][i], q_data[4][i]}, 125*7b3ff069SJeremy L Thompson {q_data[5][i], q_data[1][i], q_data[3][i]}, 126*7b3ff069SJeremy L Thompson {q_data[4][i], q_data[3][i], q_data[2][i]} 127*7b3ff069SJeremy L Thompson }; 128*7b3ff069SJeremy L Thompson 129*7b3ff069SJeremy L Thompson // j = direction of vg 130*7b3ff069SJeremy L Thompson 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]); 131*7b3ff069SJeremy L Thompson } // End of Quadrature Point Loop 132*7b3ff069SJeremy L Thompson } break; 133*7b3ff069SJeremy L Thompson } 134*7b3ff069SJeremy L Thompson return CEED_ERROR_SUCCESS; 135*7b3ff069SJeremy L Thompson } 136