1 // Copyright (c) 2017-2022, 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 #ifndef bp3_h 9 #define bp3_h 10 11 #include <ceed.h> 12 13 /// A structure used to pass additional data to f_build_diff and f_apply_diff 14 struct BuildContext { CeedInt dim, space_dim; }; 15 16 /// libCEED Q-function for building quadrature data for a diffusion operator 17 CEED_QFUNCTION(f_build_diff)(void *ctx, const CeedInt Q, 18 const CeedScalar *const *in, CeedScalar *const *out) { 19 BuildContext *bc = (BuildContext *)ctx; 20 // in[0] is Jacobians with shape [dim, nc=dim, Q] 21 // in[1] is quadrature weights, size (Q) 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 const CeedScalar *J = in[0], *w = in[1]; 26 CeedScalar *qdata = out[0]; 27 28 switch (bc->dim + 10*bc->space_dim) { 29 case 11: 30 // Quadrature Point Loop 31 CeedPragmaSIMD 32 for (CeedInt i=0; i<Q; i++) { 33 qdata[i] = w[i] / J[i]; 34 } 35 break; 36 case 22: 37 // Quadrature Point Loop 38 CeedPragmaSIMD 39 for (CeedInt i=0; i<Q; i++) { 40 // J: 0 2 qdata: 0 2 adj(J): J22 -J12 41 // 1 3 2 1 -J21 J11 42 const CeedScalar J11 = J[i+Q*0]; 43 const CeedScalar J21 = J[i+Q*1]; 44 const CeedScalar J12 = J[i+Q*2]; 45 const CeedScalar J22 = J[i+Q*3]; 46 const CeedScalar qw = w[i] / (J11*J22 - J21*J12); 47 qdata[i+Q*0] = qw * (J12*J12 + J22*J22); 48 qdata[i+Q*1] = qw * (J11*J11 + J21*J21); 49 qdata[i+Q*2] = - qw * (J11*J12 + J21*J22); 50 } 51 break; 52 case 33: 53 // Quadrature Point Loop 54 CeedPragmaSIMD 55 for (CeedInt i=0; i<Q; i++) { 56 // J: 0 3 6 qdata: 0 5 4 57 // 1 4 7 5 1 3 58 // 2 5 8 4 3 2 59 const CeedScalar J11 = J[i+Q*0]; 60 const CeedScalar J21 = J[i+Q*1]; 61 const CeedScalar J31 = J[i+Q*2]; 62 const CeedScalar J12 = J[i+Q*3]; 63 const CeedScalar J22 = J[i+Q*4]; 64 const CeedScalar J32 = J[i+Q*5]; 65 const CeedScalar J13 = J[i+Q*6]; 66 const CeedScalar J23 = J[i+Q*7]; 67 const CeedScalar J33 = J[i+Q*8]; 68 const CeedScalar A11 = J22*J33 - J23*J32; 69 const CeedScalar A12 = J13*J32 - J12*J33; 70 const CeedScalar A13 = J12*J23 - J13*J22; 71 const CeedScalar A21 = J23*J31 - J21*J33; 72 const CeedScalar A22 = J11*J33 - J13*J31; 73 const CeedScalar A23 = J13*J21 - J11*J23; 74 const CeedScalar A31 = J21*J32 - J22*J31; 75 const CeedScalar A32 = J12*J31 - J11*J32; 76 const CeedScalar A33 = J11*J22 - J12*J21; 77 const CeedScalar qw = w[i] / (J11*A11 + J21*A12 + J31*A13); 78 qdata[i+Q*0] = qw * (A11*A11 + A12*A12 + A13*A13); 79 qdata[i+Q*1] = qw * (A21*A21 + A22*A22 + A23*A23); 80 qdata[i+Q*2] = qw * (A31*A31 + A32*A32 + A33*A33); 81 qdata[i+Q*3] = qw * (A21*A31 + A22*A32 + A23*A33); 82 qdata[i+Q*4] = qw * (A11*A31 + A12*A32 + A13*A33); 83 qdata[i+Q*5] = qw * (A11*A21 + A12*A22 + A13*A23); 84 } 85 break; 86 } 87 return 0; 88 } 89 90 /// libCEED Q-function for applying a diff operator 91 CEED_QFUNCTION(f_apply_diff)(void *ctx, const CeedInt Q, 92 const CeedScalar *const *in, CeedScalar *const *out) { 93 BuildContext *bc = (BuildContext *)ctx; 94 // in[0], out[0] have shape [dim, nc=1, Q] 95 const CeedScalar *ug = in[0], *qdata = in[1]; 96 CeedScalar *vg = out[0]; 97 98 switch (bc->dim) { 99 case 1: 100 // Quadrature Point Loop 101 CeedPragmaSIMD 102 for (CeedInt i=0; i<Q; i++) { 103 vg[i] = ug[i] * qdata[i]; 104 } 105 break; 106 case 2: 107 // Quadrature Point Loop 108 CeedPragmaSIMD 109 for (CeedInt i=0; i<Q; i++) { 110 const CeedScalar ug0 = ug[i+Q*0]; 111 const CeedScalar ug1 = ug[i+Q*1]; 112 vg[i+Q*0] = qdata[i+Q*0]*ug0 + qdata[i+Q*2]*ug1; 113 vg[i+Q*1] = qdata[i+Q*2]*ug0 + qdata[i+Q*1]*ug1; 114 } 115 break; 116 case 3: 117 // Quadrature Point Loop 118 CeedPragmaSIMD 119 for (CeedInt i=0; i<Q; i++) { 120 const CeedScalar ug0 = ug[i+Q*0]; 121 const CeedScalar ug1 = ug[i+Q*1]; 122 const CeedScalar ug2 = ug[i+Q*2]; 123 vg[i+Q*0] = qdata[i+Q*0]*ug0 + qdata[i+Q*5]*ug1 + qdata[i+Q*4]*ug2; 124 vg[i+Q*1] = qdata[i+Q*5]*ug0 + qdata[i+Q*1]*ug1 + qdata[i+Q*3]*ug2; 125 vg[i+Q*2] = qdata[i+Q*4]*ug0 + qdata[i+Q*3]*ug1 + qdata[i+Q*2]*ug2; 126 } 127 break; 128 } 129 return 0; 130 } 131 132 #endif // bp3_h 133