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 8 /// @file 9 /// Internal header for CUDA shared memory tensor product basis 10 #ifndef CEED_CUDA_SHARED_BASIS_TENSOR_H 11 #define CEED_CUDA_SHARED_BASIS_TENSOR_H 12 13 #include <ceed.h> 14 15 #include "cuda-shared-basis-read-write-templates.h" 16 #include "cuda-shared-basis-tensor-templates.h" 17 18 //------------------------------------------------------------------------------ 19 // Interp kernel by dim 20 //------------------------------------------------------------------------------ 21 extern "C" __global__ void Interp(const CeedInt num_elem, const CeedScalar *c_B, const CeedScalar *__restrict__ d_U, CeedScalar *__restrict__ d_V) { 22 extern __shared__ CeedScalar slice[]; 23 24 SharedData_Cuda data; 25 data.t_id_x = threadIdx.x; 26 data.t_id_y = threadIdx.y; 27 data.t_id_z = threadIdx.z; 28 data.t_id = threadIdx.x + threadIdx.y * blockDim.x + threadIdx.z * blockDim.y * blockDim.x; 29 data.slice = slice + data.t_id_z * T_1D * (BASIS_DIM > 1 ? T_1D : 1); 30 31 CeedScalar r_U[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_P_1D : 1)]; 32 CeedScalar r_V[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 33 34 for (CeedInt elem = blockIdx.x * blockDim.z + threadIdx.z; elem < num_elem; elem += gridDim.x * blockDim.z) { 35 if (BASIS_DIM == 1) { 36 ReadElementStrided1d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * num_elem, BASIS_P_1D, d_U, r_U); 37 Interp1d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, r_V); 38 WriteElementStrided1d<BASIS_NUM_COMP, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * num_elem, BASIS_Q_1D, r_V, d_V); 39 } else if (BASIS_DIM == 2) { 40 ReadElementStrided2d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * num_elem, BASIS_P_1D * BASIS_P_1D, d_U, r_U); 41 InterpTensor2d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, r_V); 42 WriteElementStrided2d<BASIS_NUM_COMP, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * BASIS_Q_1D * num_elem, BASIS_Q_1D * BASIS_Q_1D, r_V, d_V); 43 } else if (BASIS_DIM == 3) { 44 ReadElementStrided3d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * BASIS_P_1D * num_elem, 45 BASIS_P_1D * BASIS_P_1D * BASIS_P_1D, d_U, r_U); 46 InterpTensor3d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, r_V); 47 WriteElementStrided3d<BASIS_NUM_COMP, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * BASIS_Q_1D * BASIS_Q_1D * num_elem, 48 BASIS_Q_1D * BASIS_Q_1D * BASIS_Q_1D, r_V, d_V); 49 } 50 } 51 } 52 53 extern "C" __global__ void InterpTranspose(const CeedInt num_elem, const CeedScalar *c_B, const CeedScalar *__restrict__ d_U, 54 CeedScalar *__restrict__ d_V) { 55 extern __shared__ CeedScalar slice[]; 56 57 SharedData_Cuda data; 58 data.t_id_x = threadIdx.x; 59 data.t_id_y = threadIdx.y; 60 data.t_id_z = threadIdx.z; 61 data.t_id = threadIdx.x + threadIdx.y * blockDim.x + threadIdx.z * blockDim.y * blockDim.x; 62 data.slice = slice + data.t_id_z * T_1D * (BASIS_DIM > 1 ? T_1D : 1); 63 64 CeedScalar r_U[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 65 CeedScalar r_V[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_P_1D : 1)]; 66 67 for (CeedInt elem = blockIdx.x * blockDim.z + threadIdx.z; elem < num_elem; elem += gridDim.x * blockDim.z) { 68 if (BASIS_DIM == 1) { 69 ReadElementStrided1d<BASIS_NUM_COMP, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * num_elem, BASIS_Q_1D, d_U, r_U); 70 InterpTranspose1d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, r_V); 71 WriteElementStrided1d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * num_elem, BASIS_P_1D, r_V, d_V); 72 } else if (BASIS_DIM == 2) { 73 ReadElementStrided2d<BASIS_NUM_COMP, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * BASIS_Q_1D * num_elem, BASIS_Q_1D * BASIS_Q_1D, d_U, r_U); 74 InterpTransposeTensor2d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, r_V); 75 WriteElementStrided2d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * num_elem, BASIS_P_1D * BASIS_P_1D, r_V, d_V); 76 } else if (BASIS_DIM == 3) { 77 ReadElementStrided3d<BASIS_NUM_COMP, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * BASIS_Q_1D * BASIS_Q_1D * num_elem, 78 BASIS_Q_1D * BASIS_Q_1D * BASIS_Q_1D, d_U, r_U); 79 InterpTransposeTensor3d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, r_V); 80 WriteElementStrided3d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * BASIS_P_1D * num_elem, 81 BASIS_P_1D * BASIS_P_1D * BASIS_P_1D, r_V, d_V); 82 } 83 } 84 } 85 86 //------------------------------------------------------------------------------ 87 // Grad kernel by dim 88 //------------------------------------------------------------------------------ 89 extern "C" __global__ void Grad(const CeedInt num_elem, const CeedScalar *c_B, const CeedScalar *c_G, const CeedScalar *__restrict__ d_U, 90 CeedScalar *__restrict__ d_V) { 91 extern __shared__ CeedScalar slice[]; 92 93 SharedData_Cuda data; 94 data.t_id_x = threadIdx.x; 95 data.t_id_y = threadIdx.y; 96 data.t_id_z = threadIdx.z; 97 data.t_id = threadIdx.x + threadIdx.y * blockDim.x + threadIdx.z * blockDim.y * blockDim.x; 98 data.slice = slice + data.t_id_z * T_1D * (BASIS_DIM > 1 ? T_1D : 1); 99 100 CeedScalar r_U[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_P_1D : 1)]; 101 CeedScalar r_V[BASIS_NUM_COMP * BASIS_DIM * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 102 103 for (CeedInt elem = blockIdx.x * blockDim.z + threadIdx.z; elem < num_elem; elem += gridDim.x * blockDim.z) { 104 if (BASIS_DIM == 1) { 105 ReadElementStrided1d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * num_elem, BASIS_P_1D, d_U, r_U); 106 Grad1d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, c_G, r_V); 107 WriteElementStrided1d<BASIS_NUM_COMP, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * num_elem, BASIS_Q_1D, r_V, d_V); 108 } else if (BASIS_DIM == 2) { 109 ReadElementStrided2d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * num_elem, BASIS_P_1D * BASIS_P_1D, d_U, r_U); 110 GradTensor2d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, c_G, r_V); 111 WriteElementStrided2d<BASIS_NUM_COMP * BASIS_DIM, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * BASIS_Q_1D * num_elem, BASIS_Q_1D * BASIS_Q_1D, r_V, 112 d_V); 113 } else if (BASIS_DIM == 3) { 114 ReadElementStrided3d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * BASIS_P_1D * num_elem, 115 BASIS_P_1D * BASIS_P_1D * BASIS_P_1D, d_U, r_U); 116 if (BASIS_HAS_COLLOCATED_GRAD) GradTensorCollocated3d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, c_G, r_V); 117 else GradTensor3d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, c_G, r_V); 118 WriteElementStrided3d<BASIS_NUM_COMP * BASIS_DIM, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * BASIS_Q_1D * BASIS_Q_1D * num_elem, 119 BASIS_Q_1D * BASIS_Q_1D * BASIS_Q_1D, r_V, d_V); 120 } 121 } 122 } 123 124 extern "C" __global__ void GradTranspose(const CeedInt num_elem, const CeedScalar *c_B, const CeedScalar *c_G, const CeedScalar *__restrict__ d_U, 125 CeedScalar *__restrict__ d_V) { 126 extern __shared__ CeedScalar slice[]; 127 128 SharedData_Cuda data; 129 data.t_id_x = threadIdx.x; 130 data.t_id_y = threadIdx.y; 131 data.t_id_z = threadIdx.z; 132 data.t_id = threadIdx.x + threadIdx.y * blockDim.x + threadIdx.z * blockDim.y * blockDim.x; 133 data.slice = slice + data.t_id_z * T_1D * (BASIS_DIM > 1 ? T_1D : 1); 134 135 CeedScalar r_U[BASIS_NUM_COMP * BASIS_DIM * (BASIS_DIM > 2 ? BASIS_Q_1D : 1)]; 136 CeedScalar r_V[BASIS_NUM_COMP * (BASIS_DIM > 2 ? BASIS_P_1D : 1)]; 137 138 for (CeedInt elem = blockIdx.x * blockDim.z + threadIdx.z; elem < num_elem; elem += gridDim.x * blockDim.z) { 139 if (BASIS_DIM == 1) { 140 ReadElementStrided1d<BASIS_NUM_COMP, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * num_elem, BASIS_Q_1D, d_U, r_U); 141 GradTranspose1d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, c_G, r_V); 142 WriteElementStrided1d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * num_elem, BASIS_P_1D, r_V, d_V); 143 } else if (BASIS_DIM == 2) { 144 ReadElementStrided2d<BASIS_NUM_COMP * BASIS_DIM, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * BASIS_Q_1D * num_elem, BASIS_Q_1D * BASIS_Q_1D, d_U, 145 r_U); 146 GradTransposeTensor2d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, c_G, r_V); 147 WriteElementStrided2d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * num_elem, BASIS_P_1D * BASIS_P_1D, r_V, d_V); 148 } else if (BASIS_DIM == 3) { 149 ReadElementStrided3d<BASIS_NUM_COMP * BASIS_DIM, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * BASIS_Q_1D * BASIS_Q_1D * num_elem, 150 BASIS_Q_1D * BASIS_Q_1D * BASIS_Q_1D, d_U, r_U); 151 if (BASIS_HAS_COLLOCATED_GRAD) GradTransposeTensorCollocated3d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, c_G, r_V); 152 else GradTransposeTensor3d<BASIS_NUM_COMP, BASIS_P_1D, BASIS_Q_1D>(data, r_U, c_B, c_G, r_V); 153 WriteElementStrided3d<BASIS_NUM_COMP, BASIS_P_1D>(data, elem, 1, BASIS_P_1D * BASIS_P_1D * BASIS_P_1D * num_elem, 154 BASIS_P_1D * BASIS_P_1D * BASIS_P_1D, r_V, d_V); 155 } 156 } 157 } 158 159 //------------------------------------------------------------------------------ 160 // Weight kernels by dim 161 //------------------------------------------------------------------------------ 162 extern "C" __global__ void Weight(const CeedInt num_elem, const CeedScalar *__restrict__ q_weight_1d, CeedScalar *__restrict__ d_W) { 163 extern __shared__ CeedScalar slice[]; 164 165 SharedData_Cuda data; 166 data.t_id_x = threadIdx.x; 167 data.t_id_y = threadIdx.y; 168 data.t_id_z = threadIdx.z; 169 data.t_id = threadIdx.x + threadIdx.y * blockDim.x + threadIdx.z * blockDim.y * blockDim.x; 170 data.slice = slice + data.t_id_z * T_1D * (BASIS_DIM > 1 ? T_1D : 1); 171 172 CeedScalar r_W[BASIS_DIM > 2 ? BASIS_Q_1D : 1]; 173 174 for (CeedInt elem = blockIdx.x * blockDim.z + threadIdx.z; elem < num_elem; elem += gridDim.x * blockDim.z) { 175 if (BASIS_DIM == 1) { 176 Weight1d<BASIS_Q_1D>(data, q_weight_1d, r_W); 177 WriteElementStrided1d<1, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * num_elem, BASIS_Q_1D, r_W, d_W); 178 } else if (BASIS_DIM == 2) { 179 WeightTensor2d<BASIS_Q_1D>(data, q_weight_1d, r_W); 180 WriteElementStrided2d<1, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * BASIS_Q_1D * num_elem, BASIS_Q_1D * BASIS_Q_1D, r_W, d_W); 181 } else if (BASIS_DIM == 3) { 182 WeightTensor3d<BASIS_Q_1D>(data, q_weight_1d, r_W); 183 WriteElementStrided3d<1, BASIS_Q_1D>(data, elem, 1, BASIS_Q_1D * BASIS_Q_1D * BASIS_Q_1D * num_elem, BASIS_Q_1D * BASIS_Q_1D * BASIS_Q_1D, r_W, 184 d_W); 185 } 186 } 187 } 188 189 //------------------------------------------------------------------------------ 190 191 #endif // CEED_CUDA_SHARED_BASIS_TENSOR_H 192