1 // Copyright (c) 2017-2024, 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 #pragma once 8 9 #include <ceed.h> 10 #include <ceed/backend.h> 11 #include <ceed/jit-source/hip/hip-types.h> 12 #include <hip/hip_runtime.h> 13 #if (HIP_VERSION >= 50200000) 14 #include <hipblas/hipblas.h> // IWYU pragma: export 15 #else 16 #include <hipblas.h> // IWYU pragma: export 17 #endif 18 19 typedef struct { 20 CeedScalar *h_array; 21 CeedScalar *h_array_borrowed; 22 CeedScalar *h_array_owned; 23 CeedScalar *d_array; 24 CeedScalar *d_array_borrowed; 25 CeedScalar *d_array_owned; 26 } CeedVector_Hip; 27 28 typedef struct { 29 hipModule_t module; 30 hipFunction_t ApplyNoTranspose, ApplyTranspose; 31 hipFunction_t ApplyUnsignedNoTranspose, ApplyUnsignedTranspose; 32 hipFunction_t ApplyUnorientedNoTranspose, ApplyUnorientedTranspose; 33 CeedInt num_nodes; 34 const CeedInt *h_offsets; 35 const CeedInt *h_offsets_borrowed; 36 const CeedInt *h_offsets_owned; 37 const CeedInt *d_offsets; 38 const CeedInt *d_offsets_borrowed; 39 const CeedInt *d_offsets_owned; 40 const CeedInt *d_t_offsets; 41 const CeedInt *d_t_indices; 42 const CeedInt *d_l_vec_indices; 43 const bool *h_orients; 44 const bool *h_orients_borrowed; 45 const bool *h_orients_owned; 46 const bool *d_orients; 47 const bool *d_orients_borrowed; 48 const bool *d_orients_owned; 49 const CeedInt8 *h_curl_orients; 50 const CeedInt8 *h_curl_orients_borrowed; 51 const CeedInt8 *h_curl_orients_owned; 52 const CeedInt8 *d_curl_orients; 53 const CeedInt8 *d_curl_orients_borrowed; 54 const CeedInt8 *d_curl_orients_owned; 55 const CeedInt *h_offsets_at_points; 56 const CeedInt *h_offsets_at_points_borrowed; 57 const CeedInt *h_offsets_at_points_owned; 58 const CeedInt *d_offsets_at_points; 59 const CeedInt *d_offsets_at_points_borrowed; 60 const CeedInt *d_offsets_at_points_owned; 61 const CeedInt *h_points_per_elem; 62 const CeedInt *h_points_per_elem_borrowed; 63 const CeedInt *h_points_per_elem_owned; 64 const CeedInt *d_points_per_elem; 65 const CeedInt *d_points_per_elem_borrowed; 66 const CeedInt *d_points_per_elem_owned; 67 } CeedElemRestriction_Hip; 68 69 typedef struct { 70 hipModule_t module; 71 hipFunction_t Interp; 72 hipFunction_t Grad; 73 hipFunction_t Weight; 74 hipModule_t moduleAtPoints; 75 CeedInt num_points; 76 hipFunction_t InterpAtPoints; 77 hipFunction_t GradAtPoints; 78 CeedScalar *d_interp_1d; 79 CeedScalar *d_grad_1d; 80 CeedScalar *d_q_weight_1d; 81 CeedScalar *d_chebyshev_interp_1d; 82 } CeedBasis_Hip; 83 84 typedef struct { 85 hipModule_t module; 86 hipFunction_t Interp; 87 hipFunction_t InterpTranspose; 88 hipFunction_t Deriv; 89 hipFunction_t DerivTranspose; 90 hipFunction_t Weight; 91 CeedScalar *d_interp; 92 CeedScalar *d_grad; 93 CeedScalar *d_div; 94 CeedScalar *d_curl; 95 CeedScalar *d_q_weight; 96 } CeedBasisNonTensor_Hip; 97 98 typedef struct { 99 hipModule_t module; 100 const char *qfunction_name; 101 const char *qfunction_source; 102 hipFunction_t QFunction; 103 Fields_Hip fields; 104 void *d_c; 105 } CeedQFunction_Hip; 106 107 typedef struct { 108 void *h_data; 109 void *h_data_borrowed; 110 void *h_data_owned; 111 void *d_data; 112 void *d_data_borrowed; 113 void *d_data_owned; 114 } CeedQFunctionContext_Hip; 115 116 typedef struct { 117 hipModule_t module, module_point_block; 118 hipFunction_t LinearDiagonal; 119 hipFunction_t LinearPointBlock; 120 CeedElemRestriction diag_rstr, point_block_diag_rstr; 121 CeedVector elem_diag, point_block_elem_diag; 122 CeedEvalMode *d_eval_modes_in, *d_eval_modes_out; 123 CeedScalar *d_identity, *d_interp_in, *d_grad_in, *d_div_in, *d_curl_in; 124 CeedScalar *d_interp_out, *d_grad_out, *d_div_out, *d_curl_out; 125 } CeedOperatorDiag_Hip; 126 127 typedef struct { 128 hipModule_t module; 129 hipFunction_t LinearAssemble; 130 CeedInt block_size_x, block_size_y, elems_per_block; 131 CeedScalar *d_B_in, *d_B_out; 132 } CeedOperatorAssemble_Hip; 133 134 typedef struct { 135 CeedVector *e_vecs; // E-vectors, inputs followed by outputs 136 uint64_t *input_states; // State tracking for passive inputs 137 CeedVector *q_vecs_in; // Input Q-vectors needed to apply operator 138 CeedVector *q_vecs_out; // Output Q-vectors needed to apply operator 139 CeedInt num_inputs, num_outputs; 140 CeedInt num_active_in, num_active_out; 141 CeedVector *qf_active_in; 142 CeedOperatorDiag_Hip *diag; 143 CeedOperatorAssemble_Hip *asmb; 144 } CeedOperator_Hip; 145 146 CEED_INTERN int CeedGetHipblasHandle_Hip(Ceed ceed, hipblasHandle_t *handle); 147 148 CEED_INTERN int CeedVectorCreate_Hip(CeedSize n, CeedVector vec); 149 150 CEED_INTERN int CeedElemRestrictionCreate_Hip(CeedMemType mem_type, CeedCopyMode copy_mode, const CeedInt *offsets, const bool *orients, 151 const CeedInt8 *curl_orients, CeedElemRestriction rstr); 152 153 CEED_INTERN int CeedBasisCreateTensorH1_Hip(CeedInt dim, CeedInt P_1d, CeedInt Q_1d, const CeedScalar *interp_1d, const CeedScalar *grad_1d, 154 const CeedScalar *q_ref_1d, const CeedScalar *q_weight_1d, CeedBasis basis); 155 CEED_INTERN int CeedBasisCreateH1_Hip(CeedElemTopology topo, CeedInt dim, CeedInt num_nodes, CeedInt num_qpts, const CeedScalar *interp, 156 const CeedScalar *grad, const CeedScalar *q_ref, const CeedScalar *q_weight, CeedBasis basis); 157 CEED_INTERN int CeedBasisCreateHdiv_Hip(CeedElemTopology topo, CeedInt dim, CeedInt num_nodes, CeedInt num_qpts, const CeedScalar *interp, 158 const CeedScalar *div, const CeedScalar *q_ref, const CeedScalar *q_weight, CeedBasis basis); 159 CEED_INTERN int CeedBasisCreateHcurl_Hip(CeedElemTopology topo, CeedInt dim, CeedInt num_nodes, CeedInt num_qpts, const CeedScalar *interp, 160 const CeedScalar *curl, const CeedScalar *q_ref, const CeedScalar *q_weight, CeedBasis basis); 161 162 CEED_INTERN int CeedQFunctionCreate_Hip(CeedQFunction qf); 163 164 CEED_INTERN int CeedQFunctionContextCreate_Hip(CeedQFunctionContext ctx); 165 166 CEED_INTERN int CeedOperatorCreate_Hip(CeedOperator op); 167