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 #include <ceed.h> 9 #include <ceed/backend.h> 10 #include <ceed/jit-tools.h> 11 #include <hip/hip_runtime.h> 12 13 #include "../hip/ceed-hip-common.h" 14 #include "../hip/ceed-hip-compile.h" 15 #include "ceed-hip-ref.h" 16 17 //------------------------------------------------------------------------------ 18 // Basis apply - tensor 19 //------------------------------------------------------------------------------ 20 int CeedBasisApply_Hip(CeedBasis basis, const CeedInt num_elem, CeedTransposeMode t_mode, CeedEvalMode eval_mode, CeedVector u, CeedVector v) { 21 Ceed ceed; 22 CeedInt Q_1d, dim; 23 const CeedInt is_transpose = t_mode == CEED_TRANSPOSE; 24 const int max_block_size = 64; 25 const CeedScalar *d_u; 26 CeedScalar *d_v; 27 CeedBasis_Hip *data; 28 29 CeedCallBackend(CeedBasisGetCeed(basis, &ceed)); 30 CeedCallBackend(CeedBasisGetData(basis, &data)); 31 32 // Get read/write access to u, v 33 if (u != CEED_VECTOR_NONE) CeedCallBackend(CeedVectorGetArrayRead(u, CEED_MEM_DEVICE, &d_u)); 34 else CeedCheck(eval_mode == CEED_EVAL_WEIGHT, ceed, CEED_ERROR_BACKEND, "An input vector is required for this CeedEvalMode"); 35 CeedCallBackend(CeedVectorGetArrayWrite(v, CEED_MEM_DEVICE, &d_v)); 36 37 // Clear v for transpose operation 38 if (is_transpose) { 39 CeedSize length; 40 41 CeedCallBackend(CeedVectorGetLength(v, &length)); 42 CeedCallHip(ceed, hipMemset(d_v, 0, length * sizeof(CeedScalar))); 43 } 44 CeedCallBackend(CeedBasisGetNumQuadraturePoints1D(basis, &Q_1d)); 45 CeedCallBackend(CeedBasisGetDimension(basis, &dim)); 46 47 // Basis action 48 switch (eval_mode) { 49 case CEED_EVAL_INTERP: { 50 void *interp_args[] = {(void *)&num_elem, (void *)&is_transpose, &data->d_interp_1d, &d_u, &d_v}; 51 const CeedInt block_size = CeedIntMin(CeedIntPow(Q_1d, dim), max_block_size); 52 53 CeedCallBackend(CeedRunKernel_Hip(ceed, data->Interp, num_elem, block_size, interp_args)); 54 } break; 55 case CEED_EVAL_GRAD: { 56 void *grad_args[] = {(void *)&num_elem, (void *)&is_transpose, &data->d_interp_1d, &data->d_grad_1d, &d_u, &d_v}; 57 const CeedInt block_size = max_block_size; 58 59 CeedCallBackend(CeedRunKernel_Hip(ceed, data->Grad, num_elem, block_size, grad_args)); 60 } break; 61 case CEED_EVAL_WEIGHT: { 62 void *weight_args[] = {(void *)&num_elem, (void *)&data->d_q_weight_1d, &d_v}; 63 const int block_size_x = Q_1d; 64 const int block_size_y = dim >= 2 ? Q_1d : 1; 65 66 CeedCallBackend(CeedRunKernelDim_Hip(ceed, data->Weight, num_elem, block_size_x, block_size_y, 1, weight_args)); 67 } break; 68 case CEED_EVAL_NONE: /* handled separately below */ 69 break; 70 // LCOV_EXCL_START 71 case CEED_EVAL_DIV: 72 case CEED_EVAL_CURL: 73 return CeedError(ceed, CEED_ERROR_BACKEND, "%s not supported", CeedEvalModes[eval_mode]); 74 // LCOV_EXCL_STOP 75 } 76 77 // Restore vectors, cover CEED_EVAL_NONE 78 CeedCallBackend(CeedVectorRestoreArray(v, &d_v)); 79 if (eval_mode == CEED_EVAL_NONE) CeedCallBackend(CeedVectorSetArray(v, CEED_MEM_DEVICE, CEED_COPY_VALUES, (CeedScalar *)d_u)); 80 if (eval_mode != CEED_EVAL_WEIGHT) CeedCallBackend(CeedVectorRestoreArrayRead(u, &d_u)); 81 return CEED_ERROR_SUCCESS; 82 } 83 84 //------------------------------------------------------------------------------ 85 // Basis apply - non-tensor 86 //------------------------------------------------------------------------------ 87 int CeedBasisApplyNonTensor_Hip(CeedBasis basis, const CeedInt num_elem, CeedTransposeMode t_mode, CeedEvalMode eval_mode, CeedVector u, 88 CeedVector v) { 89 Ceed ceed; 90 CeedInt num_nodes, num_qpts; 91 const CeedInt is_transpose = t_mode == CEED_TRANSPOSE; 92 const int elems_per_block = 1; 93 const int grid = CeedDivUpInt(num_elem, elems_per_block); 94 const CeedScalar *d_u; 95 CeedScalar *d_v; 96 CeedBasisNonTensor_Hip *data; 97 98 CeedCallBackend(CeedBasisGetCeed(basis, &ceed)); 99 CeedCallBackend(CeedBasisGetData(basis, &data)); 100 CeedCallBackend(CeedBasisGetNumQuadraturePoints(basis, &num_qpts)); 101 CeedCallBackend(CeedBasisGetNumNodes(basis, &num_nodes)); 102 103 // Get read/write access to u, v 104 if (u != CEED_VECTOR_NONE) CeedCallBackend(CeedVectorGetArrayRead(u, CEED_MEM_DEVICE, &d_u)); 105 else CeedCheck(eval_mode == CEED_EVAL_WEIGHT, ceed, CEED_ERROR_BACKEND, "An input vector is required for this CeedEvalMode"); 106 CeedCallBackend(CeedVectorGetArrayWrite(v, CEED_MEM_DEVICE, &d_v)); 107 108 // Clear v for transpose operation 109 if (is_transpose) { 110 CeedSize length; 111 112 CeedCallBackend(CeedVectorGetLength(v, &length)); 113 CeedCallHip(ceed, hipMemset(d_v, 0, length * sizeof(CeedScalar))); 114 } 115 116 // Apply basis operation 117 switch (eval_mode) { 118 case CEED_EVAL_INTERP: { 119 void *interp_args[] = {(void *)&num_elem, &data->d_interp, &d_u, &d_v}; 120 const int block_size_x = is_transpose ? num_nodes : num_qpts; 121 122 if (is_transpose) { 123 CeedCallBackend(CeedRunKernelDim_Hip(ceed, data->InterpTranspose, grid, block_size_x, 1, elems_per_block, interp_args)); 124 } else { 125 CeedCallBackend(CeedRunKernelDim_Hip(ceed, data->Interp, grid, block_size_x, 1, elems_per_block, interp_args)); 126 } 127 } break; 128 case CEED_EVAL_GRAD: { 129 void *grad_args[] = {(void *)&num_elem, &data->d_grad, &d_u, &d_v}; 130 const int block_size_x = is_transpose ? num_nodes : num_qpts; 131 132 if (is_transpose) { 133 CeedCallBackend(CeedRunKernelDim_Hip(ceed, data->DerivTranspose, grid, block_size_x, 1, elems_per_block, grad_args)); 134 } else { 135 CeedCallBackend(CeedRunKernelDim_Hip(ceed, data->Deriv, grid, block_size_x, 1, elems_per_block, grad_args)); 136 } 137 } break; 138 case CEED_EVAL_DIV: { 139 void *div_args[] = {(void *)&num_elem, &data->d_div, &d_u, &d_v}; 140 const int block_size_x = is_transpose ? num_nodes : num_qpts; 141 142 if (is_transpose) { 143 CeedCallBackend(CeedRunKernelDim_Hip(ceed, data->DerivTranspose, grid, block_size_x, 1, elems_per_block, div_args)); 144 } else { 145 CeedCallBackend(CeedRunKernelDim_Hip(ceed, data->Deriv, grid, block_size_x, 1, elems_per_block, div_args)); 146 } 147 } break; 148 case CEED_EVAL_CURL: { 149 void *curl_args[] = {(void *)&num_elem, &data->d_curl, &d_u, &d_v}; 150 const int block_size_x = is_transpose ? num_nodes : num_qpts; 151 152 if (is_transpose) { 153 CeedCallBackend(CeedRunKernelDim_Hip(ceed, data->DerivTranspose, grid, block_size_x, 1, elems_per_block, curl_args)); 154 } else { 155 CeedCallBackend(CeedRunKernelDim_Hip(ceed, data->Deriv, grid, block_size_x, 1, elems_per_block, curl_args)); 156 } 157 } break; 158 case CEED_EVAL_WEIGHT: { 159 void *weight_args[] = {(void *)&num_elem, (void *)&data->d_q_weight, &d_v}; 160 161 CeedCallBackend(CeedRunKernelDim_Hip(ceed, data->Weight, grid, num_qpts, 1, elems_per_block, weight_args)); 162 } break; 163 case CEED_EVAL_NONE: /* handled separately below */ 164 break; 165 } 166 167 // Restore vectors, cover CEED_EVAL_NONE 168 CeedCallBackend(CeedVectorRestoreArray(v, &d_v)); 169 if (eval_mode == CEED_EVAL_NONE) CeedCallBackend(CeedVectorSetArray(v, CEED_MEM_DEVICE, CEED_COPY_VALUES, (CeedScalar *)d_u)); 170 if (eval_mode != CEED_EVAL_WEIGHT) CeedCallBackend(CeedVectorRestoreArrayRead(u, &d_u)); 171 return CEED_ERROR_SUCCESS; 172 } 173 174 //------------------------------------------------------------------------------ 175 // Destroy tensor basis 176 //------------------------------------------------------------------------------ 177 static int CeedBasisDestroy_Hip(CeedBasis basis) { 178 Ceed ceed; 179 CeedBasis_Hip *data; 180 181 CeedCallBackend(CeedBasisGetCeed(basis, &ceed)); 182 CeedCallBackend(CeedBasisGetData(basis, &data)); 183 CeedCallHip(ceed, hipModuleUnload(data->module)); 184 CeedCallHip(ceed, hipFree(data->d_q_weight_1d)); 185 CeedCallHip(ceed, hipFree(data->d_interp_1d)); 186 CeedCallHip(ceed, hipFree(data->d_grad_1d)); 187 CeedCallBackend(CeedFree(&data)); 188 return CEED_ERROR_SUCCESS; 189 } 190 191 //------------------------------------------------------------------------------ 192 // Destroy non-tensor basis 193 //------------------------------------------------------------------------------ 194 static int CeedBasisDestroyNonTensor_Hip(CeedBasis basis) { 195 Ceed ceed; 196 CeedBasisNonTensor_Hip *data; 197 198 CeedCallBackend(CeedBasisGetCeed(basis, &ceed)); 199 CeedCallBackend(CeedBasisGetData(basis, &data)); 200 CeedCallHip(ceed, hipModuleUnload(data->module)); 201 CeedCallHip(ceed, hipFree(data->d_q_weight)); 202 CeedCallHip(ceed, hipFree(data->d_interp)); 203 CeedCallHip(ceed, hipFree(data->d_grad)); 204 CeedCallHip(ceed, hipFree(data->d_div)); 205 CeedCallHip(ceed, hipFree(data->d_curl)); 206 CeedCallBackend(CeedFree(&data)); 207 return CEED_ERROR_SUCCESS; 208 } 209 210 //------------------------------------------------------------------------------ 211 // Create tensor 212 //------------------------------------------------------------------------------ 213 int CeedBasisCreateTensorH1_Hip(CeedInt dim, CeedInt P_1d, CeedInt Q_1d, const CeedScalar *interp_1d, const CeedScalar *grad_1d, 214 const CeedScalar *q_ref_1d, const CeedScalar *q_weight_1d, CeedBasis basis) { 215 Ceed ceed; 216 char *basis_kernel_path, *basis_kernel_source; 217 CeedInt num_comp; 218 const CeedInt q_bytes = Q_1d * sizeof(CeedScalar); 219 const CeedInt interp_bytes = q_bytes * P_1d; 220 CeedBasis_Hip *data; 221 222 CeedCallBackend(CeedBasisGetCeed(basis, &ceed)); 223 CeedCallBackend(CeedCalloc(1, &data)); 224 225 // Copy data to GPU 226 CeedCallHip(ceed, hipMalloc((void **)&data->d_q_weight_1d, q_bytes)); 227 CeedCallHip(ceed, hipMemcpy(data->d_q_weight_1d, q_weight_1d, q_bytes, hipMemcpyHostToDevice)); 228 CeedCallHip(ceed, hipMalloc((void **)&data->d_interp_1d, interp_bytes)); 229 CeedCallHip(ceed, hipMemcpy(data->d_interp_1d, interp_1d, interp_bytes, hipMemcpyHostToDevice)); 230 CeedCallHip(ceed, hipMalloc((void **)&data->d_grad_1d, interp_bytes)); 231 CeedCallHip(ceed, hipMemcpy(data->d_grad_1d, grad_1d, interp_bytes, hipMemcpyHostToDevice)); 232 233 // Compile basis kernels 234 CeedCallBackend(CeedBasisGetNumComponents(basis, &num_comp)); 235 CeedCallBackend(CeedGetJitAbsolutePath(ceed, "ceed/jit-source/hip/hip-ref-basis-tensor.h", &basis_kernel_path)); 236 CeedDebug256(ceed, CEED_DEBUG_COLOR_SUCCESS, "----- Loading Basis Kernel Source -----\n"); 237 CeedCallBackend(CeedLoadSourceToBuffer(ceed, basis_kernel_path, &basis_kernel_source)); 238 CeedDebug256(ceed, CEED_DEBUG_COLOR_SUCCESS, "----- Loading Basis Kernel Source Complete! -----\n"); 239 CeedCallBackend(CeedCompile_Hip(ceed, basis_kernel_source, &data->module, 7, "BASIS_Q_1D", Q_1d, "BASIS_P_1D", P_1d, "BASIS_BUF_LEN", 240 num_comp * CeedIntPow(Q_1d > P_1d ? Q_1d : P_1d, dim), "BASIS_DIM", dim, "BASIS_NUM_COMP", num_comp, 241 "BASIS_NUM_NODES", CeedIntPow(P_1d, dim), "BASIS_NUM_QPTS", CeedIntPow(Q_1d, dim))); 242 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "Interp", &data->Interp)); 243 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "Grad", &data->Grad)); 244 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "Weight", &data->Weight)); 245 CeedCallBackend(CeedFree(&basis_kernel_path)); 246 CeedCallBackend(CeedFree(&basis_kernel_source)); 247 248 CeedCallBackend(CeedBasisSetData(basis, data)); 249 250 // Register backend functions 251 CeedCallBackend(CeedSetBackendFunction(ceed, "Basis", basis, "Apply", CeedBasisApply_Hip)); 252 CeedCallBackend(CeedSetBackendFunction(ceed, "Basis", basis, "Destroy", CeedBasisDestroy_Hip)); 253 return CEED_ERROR_SUCCESS; 254 } 255 256 //------------------------------------------------------------------------------ 257 // Create non-tensor H^1 258 //------------------------------------------------------------------------------ 259 int CeedBasisCreateH1_Hip(CeedElemTopology topo, CeedInt dim, CeedInt num_nodes, CeedInt num_qpts, const CeedScalar *interp, const CeedScalar *grad, 260 const CeedScalar *q_ref, const CeedScalar *q_weight, CeedBasis basis) { 261 Ceed ceed; 262 char *basis_kernel_path, *basis_kernel_source; 263 CeedInt num_comp, q_comp_interp, q_comp_grad; 264 const CeedInt q_bytes = num_qpts * sizeof(CeedScalar); 265 CeedBasisNonTensor_Hip *data; 266 267 CeedCallBackend(CeedBasisGetCeed(basis, &ceed)); 268 CeedCallBackend(CeedCalloc(1, &data)); 269 270 // Copy basis data to GPU 271 CeedCallBackend(CeedBasisGetNumQuadratureComponents(basis, CEED_EVAL_INTERP, &q_comp_interp)); 272 CeedCallBackend(CeedBasisGetNumQuadratureComponents(basis, CEED_EVAL_GRAD, &q_comp_grad)); 273 CeedCallHip(ceed, hipMalloc((void **)&data->d_q_weight, q_bytes)); 274 CeedCallHip(ceed, hipMemcpy(data->d_q_weight, q_weight, q_bytes, hipMemcpyHostToDevice)); 275 if (interp) { 276 const CeedInt interp_bytes = q_bytes * num_nodes * q_comp_interp; 277 278 CeedCallHip(ceed, hipMalloc((void **)&data->d_interp, interp_bytes)); 279 CeedCallHip(ceed, hipMemcpy(data->d_interp, interp, interp_bytes, hipMemcpyHostToDevice)); 280 } 281 if (grad) { 282 const CeedInt grad_bytes = q_bytes * num_nodes * q_comp_grad; 283 284 CeedCallHip(ceed, hipMalloc((void **)&data->d_grad, grad_bytes)); 285 CeedCallHip(ceed, hipMemcpy(data->d_grad, grad, grad_bytes, hipMemcpyHostToDevice)); 286 } 287 288 // Compile basis kernels 289 CeedCallBackend(CeedBasisGetNumComponents(basis, &num_comp)); 290 CeedCallBackend(CeedGetJitAbsolutePath(ceed, "ceed/jit-source/hip/hip-ref-basis-nontensor.h", &basis_kernel_path)); 291 CeedDebug256(ceed, CEED_DEBUG_COLOR_SUCCESS, "----- Loading Basis Kernel Source -----\n"); 292 CeedCallBackend(CeedLoadSourceToBuffer(ceed, basis_kernel_path, &basis_kernel_source)); 293 CeedDebug256(ceed, CEED_DEBUG_COLOR_SUCCESS, "----- Loading Basis Kernel Source Complete! -----\n"); 294 CeedCallBackend(CeedCompile_Hip(ceed, basis_kernel_source, &data->module, 5, "BASIS_Q", num_qpts, "BASIS_P", num_nodes, "BASIS_Q_COMP_INTERP", 295 q_comp_interp, "BASIS_Q_COMP_DERIV", q_comp_grad, "BASIS_NUM_COMP", num_comp)); 296 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "Interp", &data->Interp)); 297 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "InterpTranspose", &data->InterpTranspose)); 298 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "Deriv", &data->Deriv)); 299 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "DerivTranspose", &data->DerivTranspose)); 300 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "Weight", &data->Weight)); 301 CeedCallBackend(CeedFree(&basis_kernel_path)); 302 CeedCallBackend(CeedFree(&basis_kernel_source)); 303 304 CeedCallBackend(CeedBasisSetData(basis, data)); 305 306 // Register backend functions 307 CeedCallBackend(CeedSetBackendFunction(ceed, "Basis", basis, "Apply", CeedBasisApplyNonTensor_Hip)); 308 CeedCallBackend(CeedSetBackendFunction(ceed, "Basis", basis, "Destroy", CeedBasisDestroyNonTensor_Hip)); 309 return CEED_ERROR_SUCCESS; 310 } 311 312 //------------------------------------------------------------------------------ 313 // Create non-tensor H(div) 314 //------------------------------------------------------------------------------ 315 int CeedBasisCreateHdiv_Hip(CeedElemTopology topo, CeedInt dim, CeedInt num_nodes, CeedInt num_qpts, const CeedScalar *interp, const CeedScalar *div, 316 const CeedScalar *q_ref, const CeedScalar *q_weight, CeedBasis basis) { 317 Ceed ceed; 318 char *basis_kernel_path, *basis_kernel_source; 319 CeedInt num_comp, q_comp_interp, q_comp_div; 320 const CeedInt q_bytes = num_qpts * sizeof(CeedScalar); 321 CeedBasisNonTensor_Hip *data; 322 323 CeedCallBackend(CeedBasisGetCeed(basis, &ceed)); 324 CeedCallBackend(CeedCalloc(1, &data)); 325 326 // Copy basis data to GPU 327 CeedCallBackend(CeedBasisGetNumQuadratureComponents(basis, CEED_EVAL_INTERP, &q_comp_interp)); 328 CeedCallBackend(CeedBasisGetNumQuadratureComponents(basis, CEED_EVAL_DIV, &q_comp_div)); 329 CeedCallHip(ceed, hipMalloc((void **)&data->d_q_weight, q_bytes)); 330 CeedCallHip(ceed, hipMemcpy(data->d_q_weight, q_weight, q_bytes, hipMemcpyHostToDevice)); 331 if (interp) { 332 const CeedInt interp_bytes = q_bytes * num_nodes * q_comp_interp; 333 334 CeedCallHip(ceed, hipMalloc((void **)&data->d_interp, interp_bytes)); 335 CeedCallHip(ceed, hipMemcpy(data->d_interp, interp, interp_bytes, hipMemcpyHostToDevice)); 336 } 337 if (div) { 338 const CeedInt div_bytes = q_bytes * num_nodes * q_comp_div; 339 340 CeedCallHip(ceed, hipMalloc((void **)&data->d_div, div_bytes)); 341 CeedCallHip(ceed, hipMemcpy(data->d_div, div, div_bytes, hipMemcpyHostToDevice)); 342 } 343 344 // Compile basis kernels 345 CeedCallBackend(CeedBasisGetNumComponents(basis, &num_comp)); 346 CeedCallBackend(CeedGetJitAbsolutePath(ceed, "ceed/jit-source/hip/hip-ref-basis-nontensor.h", &basis_kernel_path)); 347 CeedDebug256(ceed, CEED_DEBUG_COLOR_SUCCESS, "----- Loading Basis Kernel Source -----\n"); 348 CeedCallBackend(CeedLoadSourceToBuffer(ceed, basis_kernel_path, &basis_kernel_source)); 349 CeedDebug256(ceed, CEED_DEBUG_COLOR_SUCCESS, "----- Loading Basis Kernel Source Complete! -----\n"); 350 CeedCallBackend(CeedCompile_Hip(ceed, basis_kernel_source, &data->module, 5, "BASIS_Q", num_qpts, "BASIS_P", num_nodes, "BASIS_Q_COMP_INTERP", 351 q_comp_interp, "BASIS_Q_COMP_DERIV", q_comp_div, "BASIS_NUM_COMP", num_comp)); 352 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "Interp", &data->Interp)); 353 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "InterpTranspose", &data->InterpTranspose)); 354 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "Deriv", &data->Deriv)); 355 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "DerivTranspose", &data->DerivTranspose)); 356 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "Weight", &data->Weight)); 357 CeedCallBackend(CeedFree(&basis_kernel_path)); 358 CeedCallBackend(CeedFree(&basis_kernel_source)); 359 360 CeedCallBackend(CeedBasisSetData(basis, data)); 361 362 // Register backend functions 363 CeedCallBackend(CeedSetBackendFunction(ceed, "Basis", basis, "Apply", CeedBasisApplyNonTensor_Hip)); 364 CeedCallBackend(CeedSetBackendFunction(ceed, "Basis", basis, "Destroy", CeedBasisDestroyNonTensor_Hip)); 365 return CEED_ERROR_SUCCESS; 366 } 367 368 //------------------------------------------------------------------------------ 369 // Create non-tensor H(curl) 370 //------------------------------------------------------------------------------ 371 int CeedBasisCreateHcurl_Hip(CeedElemTopology topo, CeedInt dim, CeedInt num_nodes, CeedInt num_qpts, const CeedScalar *interp, 372 const CeedScalar *curl, const CeedScalar *q_ref, const CeedScalar *q_weight, CeedBasis basis) { 373 Ceed ceed; 374 char *basis_kernel_path, *basis_kernel_source; 375 CeedInt num_comp, q_comp_interp, q_comp_curl; 376 const CeedInt q_bytes = num_qpts * sizeof(CeedScalar); 377 CeedBasisNonTensor_Hip *data; 378 379 CeedCallBackend(CeedBasisGetCeed(basis, &ceed)); 380 CeedCallBackend(CeedCalloc(1, &data)); 381 382 // Copy basis data to GPU 383 CeedCallBackend(CeedBasisGetNumQuadratureComponents(basis, CEED_EVAL_INTERP, &q_comp_interp)); 384 CeedCallBackend(CeedBasisGetNumQuadratureComponents(basis, CEED_EVAL_CURL, &q_comp_curl)); 385 CeedCallHip(ceed, hipMalloc((void **)&data->d_q_weight, q_bytes)); 386 CeedCallHip(ceed, hipMemcpy(data->d_q_weight, q_weight, q_bytes, hipMemcpyHostToDevice)); 387 if (interp) { 388 const CeedInt interp_bytes = q_bytes * num_nodes * q_comp_interp; 389 390 CeedCallHip(ceed, hipMalloc((void **)&data->d_interp, interp_bytes)); 391 CeedCallHip(ceed, hipMemcpy(data->d_interp, interp, interp_bytes, hipMemcpyHostToDevice)); 392 } 393 if (curl) { 394 const CeedInt curl_bytes = q_bytes * num_nodes * q_comp_curl; 395 396 CeedCallHip(ceed, hipMalloc((void **)&data->d_curl, curl_bytes)); 397 CeedCallHip(ceed, hipMemcpy(data->d_curl, curl, curl_bytes, hipMemcpyHostToDevice)); 398 } 399 400 // Compile basis kernels 401 CeedCallBackend(CeedBasisGetNumComponents(basis, &num_comp)); 402 CeedCallBackend(CeedGetJitAbsolutePath(ceed, "ceed/jit-source/hip/hip-ref-basis-nontensor.h", &basis_kernel_path)); 403 CeedDebug256(ceed, CEED_DEBUG_COLOR_SUCCESS, "----- Loading Basis Kernel Source -----\n"); 404 CeedCallBackend(CeedLoadSourceToBuffer(ceed, basis_kernel_path, &basis_kernel_source)); 405 CeedDebug256(ceed, CEED_DEBUG_COLOR_SUCCESS, "----- Loading Basis Kernel Source Complete! -----\n"); 406 CeedCallBackend(CeedCompile_Hip(ceed, basis_kernel_source, &data->module, 5, "BASIS_Q", num_qpts, "BASIS_P", num_nodes, "BASIS_Q_COMP_INTERP", 407 q_comp_interp, "BASIS_Q_COMP_DERIV", q_comp_curl, "BASIS_NUM_COMP", num_comp)); 408 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "Interp", &data->Interp)); 409 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "InterpTranspose", &data->InterpTranspose)); 410 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "Deriv", &data->Deriv)); 411 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "DerivTranspose", &data->DerivTranspose)); 412 CeedCallBackend(CeedGetKernel_Hip(ceed, data->module, "Weight", &data->Weight)); 413 CeedCallBackend(CeedFree(&basis_kernel_path)); 414 CeedCallBackend(CeedFree(&basis_kernel_source)); 415 416 CeedCallBackend(CeedBasisSetData(basis, data)); 417 418 // Register backend functions 419 CeedCallBackend(CeedSetBackendFunction(ceed, "Basis", basis, "Apply", CeedBasisApplyNonTensor_Hip)); 420 CeedCallBackend(CeedSetBackendFunction(ceed, "Basis", basis, "Destroy", CeedBasisDestroyNonTensor_Hip)); 421 return CEED_ERROR_SUCCESS; 422 } 423 424 //------------------------------------------------------------------------------ 425