1 // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 2 // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 3 // All Rights reserved. See files LICENSE and NOTICE for details. 4 // 5 // This file is part of CEED, a collection of benchmarks, miniapps, software 6 // libraries and APIs for efficient high-order finite element and spectral 7 // element discretizations for exascale applications. For more information and 8 // source code availability see http://github.com/ceed. 9 // 10 // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 11 // a collaborative effort of two U.S. Department of Energy organizations (Office 12 // of Science and the National Nuclear Security Administration) responsible for 13 // the planning and preparation of a capable exascale ecosystem, including 14 // software, applications, hardware, advanced system engineering and early 15 // testbed platforms, in support of the nation's exascale computing imperative. 16 17 #ifndef _ceed_cuda_h 18 #define _ceed_cuda_h 19 20 #include <ceed/ceed.h> 21 #include <ceed/backend.h> 22 #include <cuda.h> 23 #include "../cuda/ceed-cuda-common.h" 24 25 typedef struct { 26 CeedScalar *h_array; 27 CeedScalar *h_array_borrowed; 28 CeedScalar *h_array_owned; 29 CeedScalar *d_array; 30 CeedScalar *d_array_borrowed; 31 CeedScalar *d_array_owned; 32 } CeedVector_Cuda; 33 34 typedef struct { 35 CUmodule module; 36 CUfunction noTrStrided; 37 CUfunction noTrOffset; 38 CUfunction trStrided; 39 CUfunction trOffset; 40 CeedInt nnodes; 41 CeedInt *h_ind; 42 CeedInt *h_ind_allocated; 43 CeedInt *d_ind; 44 CeedInt *d_ind_allocated; 45 CeedInt *d_toffsets; 46 CeedInt *d_tindices; 47 CeedInt *d_lvec_indices; 48 } CeedElemRestriction_Cuda; 49 50 // We use a struct to avoid having to memCpy the array of pointers 51 // __global__ copies by value the struct. 52 typedef struct { 53 const CeedScalar *inputs[CEED_FIELD_MAX]; 54 CeedScalar *outputs[CEED_FIELD_MAX]; 55 } Fields_Cuda; 56 57 typedef struct { 58 CUmodule module; 59 char *qFunctionName; 60 char *qFunctionSource; 61 CUfunction qFunction; 62 Fields_Cuda fields; 63 void *d_c; 64 } CeedQFunction_Cuda; 65 66 typedef struct { 67 void *h_data; 68 void *h_data_borrowed; 69 void *h_data_owned; 70 void *d_data; 71 void *d_data_borrowed; 72 void *d_data_owned; 73 } CeedQFunctionContext_Cuda; 74 75 typedef struct { 76 CUmodule module; 77 CUfunction interp; 78 CUfunction grad; 79 CUfunction weight; 80 CeedScalar *d_interp1d; 81 CeedScalar *d_grad1d; 82 CeedScalar *d_qweight1d; 83 } CeedBasis_Cuda; 84 85 typedef struct { 86 CUmodule module; 87 CUfunction interp; 88 CUfunction grad; 89 CUfunction weight; 90 CeedScalar *d_interp; 91 CeedScalar *d_grad; 92 CeedScalar *d_qweight; 93 } CeedBasisNonTensor_Cuda; 94 95 typedef struct { 96 CUmodule module; 97 CUfunction linearDiagonal; 98 CUfunction linearPointBlock; 99 CeedBasis basisin, basisout; 100 CeedElemRestriction diagrstr, pbdiagrstr; 101 CeedVector elemdiag, pbelemdiag; 102 CeedInt numemodein, numemodeout, nnodes; 103 CeedEvalMode *h_emodein, *h_emodeout; 104 CeedEvalMode *d_emodein, *d_emodeout; 105 CeedScalar *d_identity, *d_interpin, *d_interpout, *d_gradin, *d_gradout; 106 } CeedOperatorDiag_Cuda; 107 108 typedef struct { 109 CeedVector *evecs; // E-vectors, inputs followed by outputs 110 CeedVector *qvecsin; // Input Q-vectors needed to apply operator 111 CeedVector *qvecsout; // Output Q-vectors needed to apply operator 112 CeedInt numein; 113 CeedInt numeout; 114 CeedInt qfnumactivein, qfnumactiveout; 115 CeedVector *qfactivein; 116 CeedOperatorDiag_Cuda *diag; 117 } CeedOperator_Cuda; 118 119 CEED_INTERN int CeedCudaGetCublasHandle(Ceed ceed, cublasHandle_t *handle); 120 121 CEED_INTERN int CeedVectorCreate_Cuda(CeedInt n, CeedVector vec); 122 123 CEED_INTERN int CeedElemRestrictionCreate_Cuda(CeedMemType mtype, 124 CeedCopyMode cmode, const CeedInt *indices, CeedElemRestriction r); 125 126 CEED_INTERN int CeedElemRestrictionCreateBlocked_Cuda(const CeedMemType mtype, 127 const CeedCopyMode cmode, const CeedInt *indices, 128 const CeedElemRestriction res); 129 130 CEED_INTERN int CeedBasisApplyElems_Cuda(CeedBasis basis, const CeedInt nelem, 131 CeedTransposeMode tmode, CeedEvalMode emode, const CeedVector u, CeedVector v); 132 133 CEED_INTERN int CeedQFunctionApplyElems_Cuda(CeedQFunction qf, const CeedInt Q, 134 const CeedVector *const u, const CeedVector *v); 135 136 CEED_INTERN int CeedBasisCreateTensorH1_Cuda(CeedInt dim, CeedInt P1d, 137 CeedInt Q1d, 138 const CeedScalar *interp1d, 139 const CeedScalar *grad1d, 140 const CeedScalar *qref1d, 141 const CeedScalar *qweight1d, 142 CeedBasis basis); 143 144 CEED_INTERN int CeedBasisCreateH1_Cuda(CeedElemTopology, CeedInt, CeedInt, 145 CeedInt, const CeedScalar *, 146 const CeedScalar *, const CeedScalar *, 147 const CeedScalar *, CeedBasis); 148 149 CEED_INTERN int CeedQFunctionCreate_Cuda(CeedQFunction qf); 150 151 CEED_INTERN int CeedQFunctionContextCreate_Cuda(CeedQFunctionContext ctx); 152 153 CEED_INTERN int CeedOperatorCreate_Cuda(CeedOperator op); 154 155 CEED_INTERN int CeedCompositeOperatorCreate_Cuda(CeedOperator op); 156 #endif 157