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 CEED_HIP_REF_H 9 #define CEED_HIP_REF_H 10 11 #include <ceed.h> 12 #include <ceed/backend.h> 13 #include <ceed/jit-source/hip/hip-types.h> 14 #include <hip/hip_runtime.h> 15 #if (HIP_VERSION >= 50200000) 16 #include <hipblas/hipblas.h> // IWYU pragma: export 17 #else 18 #include <hipblas.h> // IWYU pragma: export 19 #endif 20 21 typedef struct { 22 CeedScalar *h_array; 23 CeedScalar *h_array_borrowed; 24 CeedScalar *h_array_owned; 25 CeedScalar *d_array; 26 CeedScalar *d_array_borrowed; 27 CeedScalar *d_array_owned; 28 } CeedVector_Hip; 29 30 typedef struct { 31 hipModule_t module; 32 hipFunction_t StridedNoTranspose; 33 hipFunction_t StridedTranspose; 34 hipFunction_t OffsetNoTranspose; 35 hipFunction_t OffsetTranspose; 36 hipFunction_t OffsetTransposeDet; 37 hipFunction_t OrientedNoTranspose; 38 hipFunction_t OrientedTranspose; 39 hipFunction_t OrientedTransposeDet; 40 hipFunction_t CurlOrientedNoTranspose; 41 hipFunction_t CurlOrientedTranspose; 42 hipFunction_t CurlOrientedTransposeDet; 43 hipFunction_t CurlOrientedUnsignedNoTranspose; 44 hipFunction_t CurlOrientedUnsignedTranspose; 45 hipFunction_t CurlOrientedUnsignedTransposeDet; 46 CeedInt num_nodes; 47 CeedInt *h_ind; 48 CeedInt *h_ind_allocated; 49 CeedInt *d_ind; 50 CeedInt *d_ind_allocated; 51 CeedInt *d_t_offsets; 52 CeedInt *d_t_indices; 53 CeedInt *d_l_vec_indices; 54 bool *h_orients; 55 bool *h_orients_allocated; 56 bool *d_orients; 57 bool *d_orients_allocated; 58 CeedInt8 *h_curl_orients; 59 CeedInt8 *h_curl_orients_allocated; 60 CeedInt8 *d_curl_orients; 61 CeedInt8 *d_curl_orients_allocated; 62 } CeedElemRestriction_Hip; 63 64 typedef struct { 65 hipModule_t module; 66 hipFunction_t Interp; 67 hipFunction_t Grad; 68 hipFunction_t Weight; 69 CeedScalar *d_interp_1d; 70 CeedScalar *d_grad_1d; 71 CeedScalar *d_q_weight_1d; 72 } CeedBasis_Hip; 73 74 typedef struct { 75 hipModule_t module; 76 hipFunction_t Interp; 77 hipFunction_t InterpTranspose; 78 hipFunction_t Deriv; 79 hipFunction_t DerivTranspose; 80 hipFunction_t Weight; 81 CeedScalar *d_interp; 82 CeedScalar *d_grad; 83 CeedScalar *d_div; 84 CeedScalar *d_curl; 85 CeedScalar *d_q_weight; 86 } CeedBasisNonTensor_Hip; 87 88 typedef struct { 89 hipModule_t module; 90 char *qfunction_name; 91 char *qfunction_source; 92 hipFunction_t QFunction; 93 Fields_Hip fields; 94 void *d_c; 95 } CeedQFunction_Hip; 96 97 typedef struct { 98 void *h_data; 99 void *h_data_borrowed; 100 void *h_data_owned; 101 void *d_data; 102 void *d_data_borrowed; 103 void *d_data_owned; 104 } CeedQFunctionContext_Hip; 105 106 typedef struct { 107 hipModule_t module; 108 hipFunction_t LinearDiagonal; 109 hipFunction_t LinearPointBlock; 110 CeedElemRestriction diag_rstr, point_block_diag_rstr; 111 CeedVector elem_diag, point_block_elem_diag; 112 CeedEvalMode *d_eval_modes_in, *d_eval_modes_out; 113 CeedScalar *d_identity, *d_interp_in, *d_grad_in, *d_div_in, *d_curl_in; 114 CeedScalar *d_interp_out, *d_grad_out, *d_div_out, *d_curl_out; 115 } CeedOperatorDiag_Hip; 116 117 typedef struct { 118 hipModule_t module; 119 hipFunction_t LinearAssemble; 120 CeedInt block_size_x, block_size_y, elems_per_block; 121 CeedScalar *d_B_in, *d_B_out; 122 } CeedOperatorAssemble_Hip; 123 124 typedef struct { 125 CeedVector *e_vecs; // E-vectors, inputs followed by outputs 126 CeedVector *q_vecs_in; // Input Q-vectors needed to apply operator 127 CeedVector *q_vecs_out; // Output Q-vectors needed to apply operator 128 CeedInt num_inputs, num_outputs; 129 CeedInt num_active_in, num_active_out; 130 CeedVector *qf_active_in; 131 CeedOperatorDiag_Hip *diag; 132 CeedOperatorAssemble_Hip *asmb; 133 } CeedOperator_Hip; 134 135 CEED_INTERN int CeedGetHipblasHandle_Hip(Ceed ceed, hipblasHandle_t *handle); 136 137 CEED_INTERN int CeedVectorCreate_Hip(CeedSize n, CeedVector vec); 138 139 CEED_INTERN int CeedElemRestrictionCreate_Hip(CeedMemType mem_type, CeedCopyMode copy_mode, const CeedInt *indices, const bool *orients, 140 const CeedInt8 *curl_orients, CeedElemRestriction r); 141 142 CEED_INTERN int CeedBasisCreateTensorH1_Hip(CeedInt dim, CeedInt P_1d, CeedInt Q_1d, const CeedScalar *interp_1d, const CeedScalar *grad_1d, 143 const CeedScalar *q_ref_1d, const CeedScalar *q_weight_1d, CeedBasis basis); 144 CEED_INTERN int CeedBasisCreateH1_Hip(CeedElemTopology topo, CeedInt dim, CeedInt num_nodes, CeedInt num_qpts, const CeedScalar *interp, 145 const CeedScalar *grad, const CeedScalar *q_ref, const CeedScalar *q_weight, CeedBasis basis); 146 CEED_INTERN int CeedBasisCreateHdiv_Hip(CeedElemTopology topo, CeedInt dim, CeedInt num_nodes, CeedInt num_qpts, const CeedScalar *interp, 147 const CeedScalar *div, const CeedScalar *q_ref, const CeedScalar *q_weight, CeedBasis basis); 148 CEED_INTERN int CeedBasisCreateHcurl_Hip(CeedElemTopology topo, CeedInt dim, CeedInt num_nodes, CeedInt num_qpts, const CeedScalar *interp, 149 const CeedScalar *curl, const CeedScalar *q_ref, const CeedScalar *q_weight, CeedBasis basis); 150 151 CEED_INTERN int CeedQFunctionCreate_Hip(CeedQFunction qf); 152 153 CEED_INTERN int CeedQFunctionContextCreate_Hip(CeedQFunctionContext ctx); 154 155 CEED_INTERN int CeedOperatorCreate_Hip(CeedOperator op); 156 157 #endif // CEED_HIP_REF_H 158