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 <stdbool.h> 11 #include <stdlib.h> 12 #include <string.h> 13 14 #include "ceed-ref.h" 15 16 //------------------------------------------------------------------------------ 17 // Core ElemRestriction Apply Code 18 //------------------------------------------------------------------------------ 19 static inline int CeedElemRestrictionApplyStridedNoTranspose_Ref_Core(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, 20 CeedInt start, CeedInt stop, CeedInt num_elem, CeedInt elem_size, 21 CeedInt v_offset, const CeedScalar *uu, CeedScalar *vv) { 22 // No offsets provided, identity restriction 23 bool has_backend_strides; 24 25 CeedCallBackend(CeedElemRestrictionHasBackendStrides(r, &has_backend_strides)); 26 if (has_backend_strides) { 27 // CPU backend strides are {1, elem_size, elem_size*num_comp} 28 // This if branch is left separate to allow better inlining 29 for (CeedInt e = start * block_size; e < stop * block_size; e += block_size) { 30 CeedPragmaSIMD for (CeedInt k = 0; k < num_comp; k++) { 31 CeedPragmaSIMD for (CeedInt n = 0; n < elem_size; n++) { 32 CeedPragmaSIMD for (CeedInt j = 0; j < block_size; j++) { 33 vv[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] = 34 uu[n + k * elem_size + CeedIntMin(e + j, num_elem - 1) * elem_size * num_comp]; 35 } 36 } 37 } 38 } 39 } else { 40 // User provided strides 41 CeedInt strides[3]; 42 43 CeedCallBackend(CeedElemRestrictionGetStrides(r, &strides)); 44 for (CeedInt e = start * block_size; e < stop * block_size; e += block_size) { 45 CeedPragmaSIMD for (CeedInt k = 0; k < num_comp; k++) { 46 CeedPragmaSIMD for (CeedInt n = 0; n < elem_size; n++) { 47 CeedPragmaSIMD for (CeedInt j = 0; j < block_size; j++) { 48 vv[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] = 49 uu[n * strides[0] + k * strides[1] + CeedIntMin(e + j, num_elem - 1) * strides[2]]; 50 } 51 } 52 } 53 } 54 } 55 return CEED_ERROR_SUCCESS; 56 } 57 58 static inline int CeedElemRestrictionApplyStandardNoTranspose_Ref_Core(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, 59 const CeedInt comp_stride, CeedInt start, CeedInt stop, CeedInt num_elem, 60 CeedInt elem_size, CeedInt v_offset, const CeedScalar *uu, CeedScalar *vv) { 61 // Default restriction with offsets 62 CeedElemRestriction_Ref *impl; 63 64 CeedCallBackend(CeedElemRestrictionGetData(r, &impl)); 65 for (CeedInt e = start * block_size; e < stop * block_size; e += block_size) { 66 CeedPragmaSIMD for (CeedInt k = 0; k < num_comp; k++) { 67 CeedPragmaSIMD for (CeedInt i = 0; i < elem_size * block_size; i++) { 68 vv[elem_size * (k * block_size + e * num_comp) + i - v_offset] = uu[impl->offsets[i + e * elem_size] + k * comp_stride]; 69 } 70 } 71 } 72 return CEED_ERROR_SUCCESS; 73 } 74 75 static inline int CeedElemRestrictionApplyOrientedNoTranspose_Ref_Core(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, 76 const CeedInt comp_stride, CeedInt start, CeedInt stop, CeedInt num_elem, 77 CeedInt elem_size, CeedInt v_offset, const CeedScalar *uu, CeedScalar *vv) { 78 // Restriction with orientations 79 CeedElemRestriction_Ref *impl; 80 81 CeedCallBackend(CeedElemRestrictionGetData(r, &impl)); 82 for (CeedInt e = start * block_size; e < stop * block_size; e += block_size) { 83 CeedPragmaSIMD for (CeedInt k = 0; k < num_comp; k++) { 84 CeedPragmaSIMD for (CeedInt i = 0; i < elem_size * block_size; i++) { 85 vv[elem_size * (k * block_size + e * num_comp) + i - v_offset] = 86 uu[impl->offsets[i + e * elem_size] + k * comp_stride] * (impl->orients[i + e * elem_size] ? -1.0 : 1.0); 87 } 88 } 89 } 90 return CEED_ERROR_SUCCESS; 91 } 92 93 static inline int CeedElemRestrictionApplyCurlOrientedNoTranspose_Ref_Core(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, 94 const CeedInt comp_stride, CeedInt start, CeedInt stop, CeedInt num_elem, 95 CeedInt elem_size, CeedInt v_offset, const CeedScalar *uu, 96 CeedScalar *vv) { 97 // Restriction with tridiagonal transformation 98 CeedElemRestriction_Ref *impl; 99 100 CeedCallBackend(CeedElemRestrictionGetData(r, &impl)); 101 for (CeedInt e = start * block_size; e < stop * block_size; e += block_size) { 102 CeedPragmaSIMD for (CeedInt k = 0; k < num_comp; k++) { 103 CeedInt n = 0; 104 CeedPragmaSIMD for (CeedInt j = 0; j < block_size; j++) { 105 vv[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] = 106 uu[impl->offsets[j + n * block_size + e * elem_size] + k * comp_stride] * 107 impl->curl_orients[j + (3 * n + 1) * block_size + e * 3 * elem_size] + 108 uu[impl->offsets[j + (n + 1) * block_size + e * elem_size] + k * comp_stride] * 109 impl->curl_orients[j + (3 * n + 2) * block_size + e * 3 * elem_size]; 110 } 111 for (n = 1; n < elem_size - 1; n++) { 112 CeedPragmaSIMD for (CeedInt j = 0; j < block_size; j++) { 113 vv[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] = 114 uu[impl->offsets[j + (n - 1) * block_size + e * elem_size] + k * comp_stride] * 115 impl->curl_orients[j + (3 * n + 0) * block_size + e * 3 * elem_size] + 116 uu[impl->offsets[j + n * block_size + e * elem_size] + k * comp_stride] * 117 impl->curl_orients[j + (3 * n + 1) * block_size + e * 3 * elem_size] + 118 uu[impl->offsets[j + (n + 1) * block_size + e * elem_size] + k * comp_stride] * 119 impl->curl_orients[j + (3 * n + 2) * block_size + e * 3 * elem_size]; 120 } 121 } 122 CeedPragmaSIMD for (CeedInt j = 0; j < block_size; j++) { 123 vv[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] = 124 uu[impl->offsets[j + (n - 1) * block_size + e * elem_size] + k * comp_stride] * 125 impl->curl_orients[j + (3 * n + 0) * block_size + e * 3 * elem_size] + 126 uu[impl->offsets[j + n * block_size + e * elem_size] + k * comp_stride] * 127 impl->curl_orients[j + (3 * n + 1) * block_size + e * 3 * elem_size]; 128 } 129 } 130 } 131 return CEED_ERROR_SUCCESS; 132 } 133 134 static inline int CeedElemRestrictionApplyCurlOrientedUnsignedNoTranspose_Ref_Core(CeedElemRestriction r, const CeedInt num_comp, 135 const CeedInt block_size, const CeedInt comp_stride, CeedInt start, 136 CeedInt stop, CeedInt num_elem, CeedInt elem_size, 137 CeedInt v_offset, const CeedScalar *uu, CeedScalar *vv) { 138 // Restriction with (unsigned) tridiagonal transformation 139 CeedElemRestriction_Ref *impl; 140 141 CeedCallBackend(CeedElemRestrictionGetData(r, &impl)); 142 for (CeedInt e = start * block_size; e < stop * block_size; e += block_size) { 143 CeedPragmaSIMD for (CeedInt k = 0; k < num_comp; k++) { 144 CeedInt n = 0; 145 146 CeedPragmaSIMD for (CeedInt j = 0; j < block_size; j++) { 147 vv[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] = 148 uu[impl->offsets[j + n * block_size + e * elem_size] + k * comp_stride] * 149 abs(impl->curl_orients[j + (3 * n + 1) * block_size + e * 3 * elem_size]) + 150 uu[impl->offsets[j + (n + 1) * block_size + e * elem_size] + k * comp_stride] * 151 abs(impl->curl_orients[j + (3 * n + 2) * block_size + e * 3 * elem_size]); 152 } 153 for (n = 1; n < elem_size - 1; n++) { 154 CeedPragmaSIMD for (CeedInt j = 0; j < block_size; j++) { 155 vv[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] = 156 uu[impl->offsets[j + (n - 1) * block_size + e * elem_size] + k * comp_stride] * 157 abs(impl->curl_orients[j + (3 * n + 0) * block_size + e * 3 * elem_size]) + 158 uu[impl->offsets[j + n * block_size + e * elem_size] + k * comp_stride] * 159 abs(impl->curl_orients[j + (3 * n + 1) * block_size + e * 3 * elem_size]) + 160 uu[impl->offsets[j + (n + 1) * block_size + e * elem_size] + k * comp_stride] * 161 abs(impl->curl_orients[j + (3 * n + 2) * block_size + e * 3 * elem_size]); 162 } 163 } 164 CeedPragmaSIMD for (CeedInt j = 0; j < block_size; j++) { 165 vv[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] = 166 uu[impl->offsets[j + (n - 1) * block_size + e * elem_size] + k * comp_stride] * 167 abs(impl->curl_orients[j + (3 * n + 0) * block_size + e * 3 * elem_size]) + 168 uu[impl->offsets[j + n * block_size + e * elem_size] + k * comp_stride] * 169 abs(impl->curl_orients[j + (3 * n + 1) * block_size + e * 3 * elem_size]); 170 } 171 } 172 } 173 return CEED_ERROR_SUCCESS; 174 } 175 176 static inline int CeedElemRestrictionApplyStridedTranspose_Ref_Core(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, 177 CeedInt start, CeedInt stop, CeedInt num_elem, CeedInt elem_size, 178 CeedInt v_offset, const CeedScalar *uu, CeedScalar *vv) { 179 // No offsets provided, identity restriction 180 bool has_backend_strides; 181 182 CeedCallBackend(CeedElemRestrictionHasBackendStrides(r, &has_backend_strides)); 183 if (has_backend_strides) { 184 // CPU backend strides are {1, elem_size, elem_size*num_comp} 185 // This if brach is left separate to allow better inlining 186 for (CeedInt e = start * block_size; e < stop * block_size; e += block_size) { 187 CeedPragmaSIMD for (CeedInt k = 0; k < num_comp; k++) { 188 CeedPragmaSIMD for (CeedInt n = 0; n < elem_size; n++) { 189 CeedPragmaSIMD for (CeedInt j = 0; j < CeedIntMin(block_size, num_elem - e); j++) { 190 vv[n + k * elem_size + (e + j) * elem_size * num_comp] += uu[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset]; 191 } 192 } 193 } 194 } 195 } else { 196 // User provided strides 197 CeedInt strides[3]; 198 199 CeedCallBackend(CeedElemRestrictionGetStrides(r, &strides)); 200 for (CeedInt e = start * block_size; e < stop * block_size; e += block_size) { 201 CeedPragmaSIMD for (CeedInt k = 0; k < num_comp; k++) { 202 CeedPragmaSIMD for (CeedInt n = 0; n < elem_size; n++) { 203 CeedPragmaSIMD for (CeedInt j = 0; j < CeedIntMin(block_size, num_elem - e); j++) { 204 vv[n * strides[0] + k * strides[1] + (e + j) * strides[2]] += 205 uu[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset]; 206 } 207 } 208 } 209 } 210 } 211 return CEED_ERROR_SUCCESS; 212 } 213 214 static inline int CeedElemRestrictionApplyStandardTranspose_Ref_Core(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, 215 const CeedInt comp_stride, CeedInt start, CeedInt stop, CeedInt num_elem, 216 CeedInt elem_size, CeedInt v_offset, const CeedScalar *uu, CeedScalar *vv) { 217 // Default restriction with offsets 218 CeedElemRestriction_Ref *impl; 219 220 CeedCallBackend(CeedElemRestrictionGetData(r, &impl)); 221 for (CeedInt e = start * block_size; e < stop * block_size; e += block_size) { 222 for (CeedInt k = 0; k < num_comp; k++) { 223 for (CeedInt i = 0; i < elem_size * block_size; i += block_size) { 224 // Iteration bound set to discard padding elements 225 for (CeedInt j = i; j < i + CeedIntMin(block_size, num_elem - e); j++) { 226 vv[impl->offsets[j + e * elem_size] + k * comp_stride] += uu[elem_size * (k * block_size + e * num_comp) + j - v_offset]; 227 } 228 } 229 } 230 } 231 return CEED_ERROR_SUCCESS; 232 } 233 234 static inline int CeedElemRestrictionApplyOrientedTranspose_Ref_Core(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, 235 const CeedInt comp_stride, CeedInt start, CeedInt stop, CeedInt num_elem, 236 CeedInt elem_size, CeedInt v_offset, const CeedScalar *uu, CeedScalar *vv) { 237 // Restriction with orientations 238 CeedElemRestriction_Ref *impl; 239 240 CeedCallBackend(CeedElemRestrictionGetData(r, &impl)); 241 for (CeedInt e = start * block_size; e < stop * block_size; e += block_size) { 242 for (CeedInt k = 0; k < num_comp; k++) { 243 for (CeedInt i = 0; i < elem_size * block_size; i += block_size) { 244 // Iteration bound set to discard padding elements 245 for (CeedInt j = i; j < i + CeedIntMin(block_size, num_elem - e); j++) { 246 vv[impl->offsets[j + e * elem_size] + k * comp_stride] += 247 uu[elem_size * (k * block_size + e * num_comp) + j - v_offset] * (impl->orients[j + e * elem_size] ? -1.0 : 1.0); 248 } 249 } 250 } 251 } 252 return CEED_ERROR_SUCCESS; 253 } 254 255 static inline int CeedElemRestrictionApplyCurlOrientedTranspose_Ref_Core(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, 256 const CeedInt comp_stride, CeedInt start, CeedInt stop, CeedInt num_elem, 257 CeedInt elem_size, CeedInt v_offset, const CeedScalar *uu, CeedScalar *vv) { 258 // Restriction with tridiagonal transformation 259 CeedElemRestriction_Ref *impl; 260 261 CeedCallBackend(CeedElemRestrictionGetData(r, &impl)); 262 for (CeedInt e = start * block_size; e < stop * block_size; e += block_size) { 263 for (CeedInt k = 0; k < num_comp; k++) { 264 // Iteration bound set to discard padding elements 265 CeedInt block_end = CeedIntMin(block_size, num_elem - e), n = 0; 266 for (CeedInt j = 0; j < block_end; j++) { 267 vv[impl->offsets[j + n * block_size + e * elem_size] + k * comp_stride] += 268 uu[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] * 269 impl->curl_orients[j + (3 * n + 1) * block_size + e * 3 * elem_size] + 270 uu[e * elem_size * num_comp + (k * elem_size + n + 1) * block_size + j - v_offset] * 271 impl->curl_orients[j + (3 * n + 3) * block_size + e * 3 * elem_size]; 272 } 273 for (n = 1; n < elem_size - 1; n++) { 274 for (CeedInt j = 0; j < block_end; j++) { 275 vv[impl->offsets[j + n * block_size + e * elem_size] + k * comp_stride] += 276 uu[e * elem_size * num_comp + (k * elem_size + n - 1) * block_size + j - v_offset] * 277 impl->curl_orients[j + (3 * n - 1) * block_size + e * 3 * elem_size] + 278 uu[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] * 279 impl->curl_orients[j + (3 * n + 1) * block_size + e * 3 * elem_size] + 280 uu[e * elem_size * num_comp + (k * elem_size + n + 1) * block_size + j - v_offset] * 281 impl->curl_orients[j + (3 * n + 3) * block_size + e * 3 * elem_size]; 282 } 283 } 284 for (CeedInt j = 0; j < block_end; j++) { 285 vv[impl->offsets[j + n * block_size + e * elem_size] + k * comp_stride] += 286 uu[e * elem_size * num_comp + (k * elem_size + n - 1) * block_size + j - v_offset] * 287 impl->curl_orients[j + (3 * n - 1) * block_size + e * 3 * elem_size] + 288 uu[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] * 289 impl->curl_orients[j + (3 * n + 1) * block_size + e * 3 * elem_size]; 290 } 291 } 292 } 293 return CEED_ERROR_SUCCESS; 294 } 295 296 static inline int CeedElemRestrictionApplyCurlOrientedUnsignedTranspose_Ref_Core(CeedElemRestriction r, const CeedInt num_comp, 297 const CeedInt block_size, const CeedInt comp_stride, CeedInt start, 298 CeedInt stop, CeedInt num_elem, CeedInt elem_size, CeedInt v_offset, 299 const CeedScalar *uu, CeedScalar *vv) { 300 // Restriction with (unsigned) tridiagonal transformation 301 CeedElemRestriction_Ref *impl; 302 303 CeedCallBackend(CeedElemRestrictionGetData(r, &impl)); 304 for (CeedInt e = start * block_size; e < stop * block_size; e += block_size) { 305 for (CeedInt k = 0; k < num_comp; k++) { 306 // Iteration bound set to discard padding elements 307 CeedInt n = 0; 308 const CeedInt block_end = CeedIntMin(block_size, num_elem - e); 309 310 for (CeedInt j = 0; j < block_end; j++) { 311 vv[impl->offsets[j + n * block_size + e * elem_size] + k * comp_stride] += 312 uu[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] * 313 abs(impl->curl_orients[j + (3 * n + 1) * block_size + e * 3 * elem_size]) + 314 uu[e * elem_size * num_comp + (k * elem_size + n + 1) * block_size + j - v_offset] * 315 abs(impl->curl_orients[j + (3 * n + 3) * block_size + e * 3 * elem_size]); 316 } 317 for (n = 1; n < elem_size - 1; n++) { 318 for (CeedInt j = 0; j < block_end; j++) { 319 vv[impl->offsets[j + n * block_size + e * elem_size] + k * comp_stride] += 320 uu[e * elem_size * num_comp + (k * elem_size + n - 1) * block_size + j - v_offset] * 321 abs(impl->curl_orients[j + (3 * n - 1) * block_size + e * 3 * elem_size]) + 322 uu[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] * 323 abs(impl->curl_orients[j + (3 * n + 1) * block_size + e * 3 * elem_size]) + 324 uu[e * elem_size * num_comp + (k * elem_size + n + 1) * block_size + j - v_offset] * 325 abs(impl->curl_orients[j + (3 * n + 3) * block_size + e * 3 * elem_size]); 326 } 327 } 328 for (CeedInt j = 0; j < block_end; j++) { 329 vv[impl->offsets[j + n * block_size + e * elem_size] + k * comp_stride] += 330 uu[e * elem_size * num_comp + (k * elem_size + n - 1) * block_size + j - v_offset] * 331 abs(impl->curl_orients[j + (3 * n - 1) * block_size + e * 3 * elem_size]) + 332 uu[e * elem_size * num_comp + (k * elem_size + n) * block_size + j - v_offset] * 333 abs(impl->curl_orients[j + (3 * n + 1) * block_size + e * 3 * elem_size]); 334 } 335 } 336 } 337 return CEED_ERROR_SUCCESS; 338 } 339 340 static inline int CeedElemRestrictionApply_Ref_Core(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, 341 const CeedInt comp_stride, CeedInt start, CeedInt stop, CeedTransposeMode t_mode, bool use_signs, 342 bool use_orients, CeedVector u, CeedVector v, CeedRequest *request) { 343 CeedInt num_elem, elem_size, v_offset; 344 CeedRestrictionType rstr_type; 345 const CeedScalar *uu; 346 CeedScalar *vv; 347 348 CeedCallBackend(CeedElemRestrictionGetNumElements(r, &num_elem)); 349 CeedCallBackend(CeedElemRestrictionGetElementSize(r, &elem_size)); 350 v_offset = start * block_size * elem_size * num_comp; 351 CeedCallBackend(CeedElemRestrictionGetType(r, &rstr_type)); 352 CeedCallBackend(CeedVectorGetArrayRead(u, CEED_MEM_HOST, &uu)); 353 354 if (t_mode == CEED_TRANSPOSE) { 355 // Sum into for transpose mode, E-vector to L-vector 356 CeedCallBackend(CeedVectorGetArray(v, CEED_MEM_HOST, &vv)); 357 } else { 358 // Overwrite for notranspose mode, L-vector to E-vector 359 CeedCallBackend(CeedVectorGetArrayWrite(v, CEED_MEM_HOST, &vv)); 360 } 361 if (t_mode == CEED_TRANSPOSE) { 362 // Restriction from E-vector to L-vector 363 // Performing v += r^T * u 364 // uu has shape [elem_size, num_comp, num_elem], row-major 365 // vv has shape [nnodes, num_comp] 366 // Sum into for transpose mode 367 switch (rstr_type) { 368 case CEED_RESTRICTION_STRIDED: 369 CeedElemRestrictionApplyStridedTranspose_Ref_Core(r, num_comp, block_size, start, stop, num_elem, elem_size, v_offset, uu, vv); 370 break; 371 case CEED_RESTRICTION_STANDARD: 372 CeedElemRestrictionApplyStandardTranspose_Ref_Core(r, num_comp, block_size, comp_stride, start, stop, num_elem, elem_size, v_offset, uu, vv); 373 break; 374 case CEED_RESTRICTION_ORIENTED: 375 if (use_signs) { 376 CeedElemRestrictionApplyOrientedTranspose_Ref_Core(r, num_comp, block_size, comp_stride, start, stop, num_elem, elem_size, v_offset, uu, 377 vv); 378 } else { 379 CeedElemRestrictionApplyStandardTranspose_Ref_Core(r, num_comp, block_size, comp_stride, start, stop, num_elem, elem_size, v_offset, uu, 380 vv); 381 } 382 break; 383 case CEED_RESTRICTION_CURL_ORIENTED: 384 if (use_signs && use_orients) { 385 CeedElemRestrictionApplyCurlOrientedTranspose_Ref_Core(r, num_comp, block_size, comp_stride, start, stop, num_elem, elem_size, v_offset, uu, 386 vv); 387 } else if (use_orients) { 388 CeedElemRestrictionApplyCurlOrientedUnsignedTranspose_Ref_Core(r, num_comp, block_size, comp_stride, start, stop, num_elem, elem_size, 389 v_offset, uu, vv); 390 } else { 391 CeedElemRestrictionApplyStandardTranspose_Ref_Core(r, num_comp, block_size, comp_stride, start, stop, num_elem, elem_size, v_offset, uu, 392 vv); 393 } 394 break; 395 } 396 } else { 397 // Restriction from L-vector to E-vector 398 // Perform: v = r * u 399 // vv has shape [elem_size, num_comp, num_elem], row-major 400 // uu has shape [nnodes, num_comp] 401 // Overwrite for notranspose mode 402 switch (rstr_type) { 403 case CEED_RESTRICTION_STRIDED: 404 CeedElemRestrictionApplyStridedNoTranspose_Ref_Core(r, num_comp, block_size, start, stop, num_elem, elem_size, v_offset, uu, vv); 405 break; 406 case CEED_RESTRICTION_STANDARD: 407 CeedElemRestrictionApplyStandardNoTranspose_Ref_Core(r, num_comp, block_size, comp_stride, start, stop, num_elem, elem_size, v_offset, uu, 408 vv); 409 break; 410 case CEED_RESTRICTION_ORIENTED: 411 if (use_signs) { 412 CeedElemRestrictionApplyOrientedNoTranspose_Ref_Core(r, num_comp, block_size, comp_stride, start, stop, num_elem, elem_size, v_offset, uu, 413 vv); 414 } else { 415 CeedElemRestrictionApplyStandardNoTranspose_Ref_Core(r, num_comp, block_size, comp_stride, start, stop, num_elem, elem_size, v_offset, uu, 416 vv); 417 } 418 break; 419 case CEED_RESTRICTION_CURL_ORIENTED: 420 if (use_signs && use_orients) { 421 CeedElemRestrictionApplyCurlOrientedNoTranspose_Ref_Core(r, num_comp, block_size, comp_stride, start, stop, num_elem, elem_size, v_offset, 422 uu, vv); 423 } else if (use_orients) { 424 CeedElemRestrictionApplyCurlOrientedUnsignedNoTranspose_Ref_Core(r, num_comp, block_size, comp_stride, start, stop, num_elem, elem_size, 425 v_offset, uu, vv); 426 } else { 427 CeedElemRestrictionApplyStandardNoTranspose_Ref_Core(r, num_comp, block_size, comp_stride, start, stop, num_elem, elem_size, v_offset, uu, 428 vv); 429 } 430 break; 431 } 432 } 433 CeedCallBackend(CeedVectorRestoreArrayRead(u, &uu)); 434 CeedCallBackend(CeedVectorRestoreArray(v, &vv)); 435 if (request != CEED_REQUEST_IMMEDIATE && request != CEED_REQUEST_ORDERED) *request = NULL; 436 return CEED_ERROR_SUCCESS; 437 } 438 439 //------------------------------------------------------------------------------ 440 // ElemRestriction Apply - Common Sizes 441 //------------------------------------------------------------------------------ 442 static int CeedElemRestrictionApply_Ref_110(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, const CeedInt comp_stride, 443 CeedInt start, CeedInt stop, CeedTransposeMode t_mode, bool use_signs, bool use_orients, CeedVector u, 444 CeedVector v, CeedRequest *request) { 445 return CeedElemRestrictionApply_Ref_Core(r, 1, 1, comp_stride, start, stop, t_mode, use_signs, use_orients, u, v, request); 446 } 447 448 static int CeedElemRestrictionApply_Ref_111(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, const CeedInt comp_stride, 449 CeedInt start, CeedInt stop, CeedTransposeMode t_mode, bool use_signs, bool use_orients, CeedVector u, 450 CeedVector v, CeedRequest *request) { 451 return CeedElemRestrictionApply_Ref_Core(r, 1, 1, 1, start, stop, t_mode, use_signs, use_orients, u, v, request); 452 } 453 454 static int CeedElemRestrictionApply_Ref_180(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, const CeedInt comp_stride, 455 CeedInt start, CeedInt stop, CeedTransposeMode t_mode, bool use_signs, bool use_orients, CeedVector u, 456 CeedVector v, CeedRequest *request) { 457 return CeedElemRestrictionApply_Ref_Core(r, 1, 8, comp_stride, start, stop, t_mode, use_signs, use_orients, u, v, request); 458 } 459 460 static int CeedElemRestrictionApply_Ref_181(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, const CeedInt comp_stride, 461 CeedInt start, CeedInt stop, CeedTransposeMode t_mode, bool use_signs, bool use_orients, CeedVector u, 462 CeedVector v, CeedRequest *request) { 463 return CeedElemRestrictionApply_Ref_Core(r, 1, 8, 1, start, stop, t_mode, use_signs, use_orients, u, v, request); 464 } 465 466 static int CeedElemRestrictionApply_Ref_310(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, const CeedInt comp_stride, 467 CeedInt start, CeedInt stop, CeedTransposeMode t_mode, bool use_signs, bool use_orients, CeedVector u, 468 CeedVector v, CeedRequest *request) { 469 return CeedElemRestrictionApply_Ref_Core(r, 3, 1, comp_stride, start, stop, t_mode, use_signs, use_orients, u, v, request); 470 } 471 472 static int CeedElemRestrictionApply_Ref_311(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, const CeedInt comp_stride, 473 CeedInt start, CeedInt stop, CeedTransposeMode t_mode, bool use_signs, bool use_orients, CeedVector u, 474 CeedVector v, CeedRequest *request) { 475 return CeedElemRestrictionApply_Ref_Core(r, 3, 1, 1, start, stop, t_mode, use_signs, use_orients, u, v, request); 476 } 477 478 static int CeedElemRestrictionApply_Ref_380(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, const CeedInt comp_stride, 479 CeedInt start, CeedInt stop, CeedTransposeMode t_mode, bool use_signs, bool use_orients, CeedVector u, 480 CeedVector v, CeedRequest *request) { 481 return CeedElemRestrictionApply_Ref_Core(r, 3, 8, comp_stride, start, stop, t_mode, use_signs, use_orients, u, v, request); 482 } 483 484 static int CeedElemRestrictionApply_Ref_381(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, const CeedInt comp_stride, 485 CeedInt start, CeedInt stop, CeedTransposeMode t_mode, bool use_signs, bool use_orients, CeedVector u, 486 CeedVector v, CeedRequest *request) { 487 return CeedElemRestrictionApply_Ref_Core(r, 3, 8, 1, start, stop, t_mode, use_signs, use_orients, u, v, request); 488 } 489 490 // LCOV_EXCL_START 491 static int CeedElemRestrictionApply_Ref_510(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, const CeedInt comp_stride, 492 CeedInt start, CeedInt stop, CeedTransposeMode t_mode, bool use_signs, bool use_orients, CeedVector u, 493 CeedVector v, CeedRequest *request) { 494 return CeedElemRestrictionApply_Ref_Core(r, 5, 1, comp_stride, start, stop, t_mode, use_signs, use_orients, u, v, request); 495 } 496 // LCOV_EXCL_STOP 497 498 static int CeedElemRestrictionApply_Ref_511(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, const CeedInt comp_stride, 499 CeedInt start, CeedInt stop, CeedTransposeMode t_mode, bool use_signs, bool use_orients, CeedVector u, 500 CeedVector v, CeedRequest *request) { 501 return CeedElemRestrictionApply_Ref_Core(r, 5, 1, 1, start, stop, t_mode, use_signs, use_orients, u, v, request); 502 } 503 504 // LCOV_EXCL_START 505 static int CeedElemRestrictionApply_Ref_580(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, const CeedInt comp_stride, 506 CeedInt start, CeedInt stop, CeedTransposeMode t_mode, bool use_signs, bool use_orients, CeedVector u, 507 CeedVector v, CeedRequest *request) { 508 return CeedElemRestrictionApply_Ref_Core(r, 5, 8, comp_stride, start, stop, t_mode, use_signs, use_orients, u, v, request); 509 } 510 // LCOV_EXCL_STOP 511 512 static int CeedElemRestrictionApply_Ref_581(CeedElemRestriction r, const CeedInt num_comp, const CeedInt block_size, const CeedInt comp_stride, 513 CeedInt start, CeedInt stop, CeedTransposeMode t_mode, bool use_signs, bool use_orients, CeedVector u, 514 CeedVector v, CeedRequest *request) { 515 return CeedElemRestrictionApply_Ref_Core(r, 5, 8, 1, start, stop, t_mode, use_signs, use_orients, u, v, request); 516 } 517 518 //------------------------------------------------------------------------------ 519 // ElemRestriction Apply 520 //------------------------------------------------------------------------------ 521 static int CeedElemRestrictionApply_Ref(CeedElemRestriction r, CeedTransposeMode t_mode, CeedVector u, CeedVector v, CeedRequest *request) { 522 CeedInt num_block, block_size, num_comp, comp_stride; 523 CeedElemRestriction_Ref *impl; 524 525 CeedCallBackend(CeedElemRestrictionGetNumBlocks(r, &num_block)); 526 CeedCallBackend(CeedElemRestrictionGetBlockSize(r, &block_size)); 527 CeedCallBackend(CeedElemRestrictionGetNumComponents(r, &num_comp)); 528 CeedCallBackend(CeedElemRestrictionGetCompStride(r, &comp_stride)); 529 CeedCallBackend(CeedElemRestrictionGetData(r, &impl)); 530 return impl->Apply(r, num_comp, block_size, comp_stride, 0, num_block, t_mode, true, true, u, v, request); 531 } 532 533 //------------------------------------------------------------------------------ 534 // ElemRestriction Apply Unsigned 535 //------------------------------------------------------------------------------ 536 static int CeedElemRestrictionApplyUnsigned_Ref(CeedElemRestriction r, CeedTransposeMode t_mode, CeedVector u, CeedVector v, CeedRequest *request) { 537 CeedInt num_block, block_size, num_comp, comp_stride; 538 CeedElemRestriction_Ref *impl; 539 540 CeedCallBackend(CeedElemRestrictionGetNumBlocks(r, &num_block)); 541 CeedCallBackend(CeedElemRestrictionGetBlockSize(r, &block_size)); 542 CeedCallBackend(CeedElemRestrictionGetNumComponents(r, &num_comp)); 543 CeedCallBackend(CeedElemRestrictionGetCompStride(r, &comp_stride)); 544 CeedCallBackend(CeedElemRestrictionGetData(r, &impl)); 545 return impl->Apply(r, num_comp, block_size, comp_stride, 0, num_block, t_mode, false, true, u, v, request); 546 } 547 548 //------------------------------------------------------------------------------ 549 // ElemRestriction Apply Unoriented 550 //------------------------------------------------------------------------------ 551 static int CeedElemRestrictionApplyUnoriented_Ref(CeedElemRestriction r, CeedTransposeMode t_mode, CeedVector u, CeedVector v, CeedRequest *request) { 552 CeedInt num_block, block_size, num_comp, comp_stride; 553 CeedElemRestriction_Ref *impl; 554 555 CeedCallBackend(CeedElemRestrictionGetNumBlocks(r, &num_block)); 556 CeedCallBackend(CeedElemRestrictionGetBlockSize(r, &block_size)); 557 CeedCallBackend(CeedElemRestrictionGetNumComponents(r, &num_comp)); 558 CeedCallBackend(CeedElemRestrictionGetCompStride(r, &comp_stride)); 559 CeedCallBackend(CeedElemRestrictionGetData(r, &impl)); 560 return impl->Apply(r, num_comp, block_size, comp_stride, 0, num_block, t_mode, false, false, u, v, request); 561 } 562 563 //------------------------------------------------------------------------------ 564 // ElemRestriction Apply Block 565 //------------------------------------------------------------------------------ 566 static int CeedElemRestrictionApplyBlock_Ref(CeedElemRestriction r, CeedInt block, CeedTransposeMode t_mode, CeedVector u, CeedVector v, 567 CeedRequest *request) { 568 CeedInt block_size, num_comp, comp_stride; 569 CeedElemRestriction_Ref *impl; 570 571 CeedCallBackend(CeedElemRestrictionGetBlockSize(r, &block_size)); 572 CeedCallBackend(CeedElemRestrictionGetNumComponents(r, &num_comp)); 573 CeedCallBackend(CeedElemRestrictionGetCompStride(r, &comp_stride)); 574 CeedCallBackend(CeedElemRestrictionGetData(r, &impl)); 575 return impl->Apply(r, num_comp, block_size, comp_stride, block, block + 1, t_mode, true, true, u, v, request); 576 } 577 578 //------------------------------------------------------------------------------ 579 // ElemRestriction Get Offsets 580 //------------------------------------------------------------------------------ 581 static int CeedElemRestrictionGetOffsets_Ref(CeedElemRestriction rstr, CeedMemType mem_type, const CeedInt **offsets) { 582 Ceed ceed; 583 CeedElemRestriction_Ref *impl; 584 585 CeedCallBackend(CeedElemRestrictionGetData(rstr, &impl)); 586 CeedCallBackend(CeedElemRestrictionGetCeed(rstr, &ceed)); 587 588 CeedCheck(mem_type == CEED_MEM_HOST, ceed, CEED_ERROR_BACKEND, "Can only provide to HOST memory"); 589 590 *offsets = impl->offsets; 591 return CEED_ERROR_SUCCESS; 592 } 593 594 //------------------------------------------------------------------------------ 595 // ElemRestriction Get Orientations 596 //------------------------------------------------------------------------------ 597 static int CeedElemRestrictionGetOrientations_Ref(CeedElemRestriction rstr, CeedMemType mem_type, const bool **orients) { 598 Ceed ceed; 599 CeedElemRestriction_Ref *impl; 600 601 CeedCallBackend(CeedElemRestrictionGetData(rstr, &impl)); 602 CeedCallBackend(CeedElemRestrictionGetCeed(rstr, &ceed)); 603 604 CeedCheck(mem_type == CEED_MEM_HOST, ceed, CEED_ERROR_BACKEND, "Can only provide to HOST memory"); 605 606 *orients = impl->orients; 607 return CEED_ERROR_SUCCESS; 608 } 609 610 //------------------------------------------------------------------------------ 611 // ElemRestriction Get Curl-Conforming Orientations 612 //------------------------------------------------------------------------------ 613 static int CeedElemRestrictionGetCurlOrientations_Ref(CeedElemRestriction rstr, CeedMemType mem_type, const CeedInt8 **curl_orients) { 614 Ceed ceed; 615 CeedElemRestriction_Ref *impl; 616 617 CeedCallBackend(CeedElemRestrictionGetData(rstr, &impl)); 618 CeedCallBackend(CeedElemRestrictionGetCeed(rstr, &ceed)); 619 620 CeedCheck(mem_type == CEED_MEM_HOST, ceed, CEED_ERROR_BACKEND, "Can only provide to HOST memory"); 621 622 *curl_orients = impl->curl_orients; 623 return CEED_ERROR_SUCCESS; 624 } 625 626 //------------------------------------------------------------------------------ 627 // ElemRestriction Destroy 628 //------------------------------------------------------------------------------ 629 static int CeedElemRestrictionDestroy_Ref(CeedElemRestriction r) { 630 CeedElemRestriction_Ref *impl; 631 632 CeedCallBackend(CeedElemRestrictionGetData(r, &impl)); 633 CeedCallBackend(CeedFree(&impl->offsets_allocated)); 634 CeedCallBackend(CeedFree(&impl->orients_allocated)); 635 CeedCallBackend(CeedFree(&impl->curl_orients_allocated)); 636 CeedCallBackend(CeedFree(&impl)); 637 return CEED_ERROR_SUCCESS; 638 } 639 640 //------------------------------------------------------------------------------ 641 // ElemRestriction Create 642 //------------------------------------------------------------------------------ 643 int CeedElemRestrictionCreate_Ref(CeedMemType mem_type, CeedCopyMode copy_mode, const CeedInt *offsets, const bool *orients, 644 const CeedInt8 *curl_orients, CeedElemRestriction r) { 645 Ceed ceed; 646 CeedInt num_elem, elem_size, num_block, block_size, num_comp, comp_stride; 647 CeedRestrictionType rstr_type; 648 CeedElemRestriction_Ref *impl; 649 650 CeedCallBackend(CeedElemRestrictionGetCeed(r, &ceed)); 651 CeedCallBackend(CeedElemRestrictionGetNumElements(r, &num_elem)); 652 CeedCallBackend(CeedElemRestrictionGetElementSize(r, &elem_size)); 653 CeedCallBackend(CeedElemRestrictionGetNumBlocks(r, &num_block)); 654 CeedCallBackend(CeedElemRestrictionGetBlockSize(r, &block_size)); 655 CeedCallBackend(CeedElemRestrictionGetNumComponents(r, &num_comp)); 656 CeedCallBackend(CeedElemRestrictionGetCompStride(r, &comp_stride)); 657 CeedInt layout[3] = {1, elem_size, elem_size * num_comp}; 658 659 CeedCheck(mem_type == CEED_MEM_HOST, ceed, CEED_ERROR_BACKEND, "Only MemType = HOST supported"); 660 CeedCallBackend(CeedCalloc(1, &impl)); 661 662 // Offsets data 663 CeedCallBackend(CeedElemRestrictionGetType(r, &rstr_type)); 664 if (rstr_type != CEED_RESTRICTION_STRIDED) { 665 const char *resource; 666 667 // Check indices for ref or memcheck backends 668 CeedCallBackend(CeedGetResource(ceed, &resource)); 669 if (!strcmp(resource, "/cpu/self/ref/serial") || !strcmp(resource, "/cpu/self/ref/blocked") || !strcmp(resource, "/cpu/self/memcheck/serial") || 670 !strcmp(resource, "/cpu/self/memcheck/blocked")) { 671 CeedSize l_size; 672 673 CeedCallBackend(CeedElemRestrictionGetLVectorSize(r, &l_size)); 674 for (CeedInt i = 0; i < num_elem * elem_size; i++) { 675 CeedCheck(offsets[i] >= 0 && offsets[i] + (num_comp - 1) * comp_stride < l_size, ceed, CEED_ERROR_BACKEND, 676 "Restriction offset %" CeedInt_FMT " (%" CeedInt_FMT ") out of range [0, %" CeedInt_FMT "]", i, offsets[i], l_size); 677 } 678 } 679 680 // Copy data 681 switch (copy_mode) { 682 case CEED_COPY_VALUES: 683 CeedCallBackend(CeedMalloc(num_elem * elem_size, &impl->offsets_allocated)); 684 memcpy(impl->offsets_allocated, offsets, num_elem * elem_size * sizeof(offsets[0])); 685 impl->offsets = impl->offsets_allocated; 686 break; 687 case CEED_OWN_POINTER: 688 impl->offsets_allocated = (CeedInt *)offsets; 689 impl->offsets = impl->offsets_allocated; 690 break; 691 case CEED_USE_POINTER: 692 impl->offsets = offsets; 693 } 694 695 // Orientation data 696 if (rstr_type == CEED_RESTRICTION_ORIENTED) { 697 CeedCheck(orients != NULL, ceed, CEED_ERROR_BACKEND, "No orients array provided for oriented restriction"); 698 switch (copy_mode) { 699 case CEED_COPY_VALUES: 700 CeedCallBackend(CeedMalloc(num_elem * elem_size, &impl->orients_allocated)); 701 memcpy(impl->orients_allocated, orients, num_elem * elem_size * sizeof(orients[0])); 702 impl->orients = impl->orients_allocated; 703 break; 704 case CEED_OWN_POINTER: 705 impl->orients_allocated = (bool *)orients; 706 impl->orients = impl->orients_allocated; 707 break; 708 case CEED_USE_POINTER: 709 impl->orients = orients; 710 } 711 } else if (rstr_type == CEED_RESTRICTION_CURL_ORIENTED) { 712 CeedCheck(curl_orients != NULL, ceed, CEED_ERROR_BACKEND, "No curl_orients array provided for oriented restriction"); 713 switch (copy_mode) { 714 case CEED_COPY_VALUES: 715 CeedCallBackend(CeedMalloc(num_elem * 3 * elem_size, &impl->curl_orients_allocated)); 716 memcpy(impl->curl_orients_allocated, curl_orients, num_elem * 3 * elem_size * sizeof(curl_orients[0])); 717 impl->curl_orients = impl->curl_orients_allocated; 718 break; 719 case CEED_OWN_POINTER: 720 impl->curl_orients_allocated = (CeedInt8 *)curl_orients; 721 impl->curl_orients = impl->curl_orients_allocated; 722 break; 723 case CEED_USE_POINTER: 724 impl->curl_orients = curl_orients; 725 } 726 } 727 } 728 729 CeedCallBackend(CeedElemRestrictionSetData(r, impl)); 730 CeedCallBackend(CeedElemRestrictionSetELayout(r, layout)); 731 CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", r, "Apply", CeedElemRestrictionApply_Ref)); 732 CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", r, "ApplyUnsigned", CeedElemRestrictionApplyUnsigned_Ref)); 733 CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", r, "ApplyUnoriented", CeedElemRestrictionApplyUnoriented_Ref)); 734 CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", r, "ApplyBlock", CeedElemRestrictionApplyBlock_Ref)); 735 CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", r, "GetOffsets", CeedElemRestrictionGetOffsets_Ref)); 736 CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", r, "GetOrientations", CeedElemRestrictionGetOrientations_Ref)); 737 CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", r, "GetCurlOrientations", CeedElemRestrictionGetCurlOrientations_Ref)); 738 CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", r, "Destroy", CeedElemRestrictionDestroy_Ref)); 739 740 // Set apply function based upon num_comp, block_size, and comp_stride 741 CeedInt index = -1; 742 743 if (block_size < 10) index = 100 * num_comp + 10 * block_size + (comp_stride == 1); 744 switch (index) { 745 case 110: 746 impl->Apply = CeedElemRestrictionApply_Ref_110; 747 break; 748 case 111: 749 impl->Apply = CeedElemRestrictionApply_Ref_111; 750 break; 751 case 180: 752 impl->Apply = CeedElemRestrictionApply_Ref_180; 753 break; 754 case 181: 755 impl->Apply = CeedElemRestrictionApply_Ref_181; 756 break; 757 case 310: 758 impl->Apply = CeedElemRestrictionApply_Ref_310; 759 break; 760 case 311: 761 impl->Apply = CeedElemRestrictionApply_Ref_311; 762 break; 763 case 380: 764 impl->Apply = CeedElemRestrictionApply_Ref_380; 765 break; 766 case 381: 767 impl->Apply = CeedElemRestrictionApply_Ref_381; 768 break; 769 // LCOV_EXCL_START 770 case 510: 771 impl->Apply = CeedElemRestrictionApply_Ref_510; 772 break; 773 // LCOV_EXCL_STOP 774 case 511: 775 impl->Apply = CeedElemRestrictionApply_Ref_511; 776 break; 777 // LCOV_EXCL_START 778 case 580: 779 impl->Apply = CeedElemRestrictionApply_Ref_580; 780 break; 781 // LCOV_EXCL_STOP 782 case 581: 783 impl->Apply = CeedElemRestrictionApply_Ref_581; 784 break; 785 default: 786 impl->Apply = CeedElemRestrictionApply_Ref_Core; 787 break; 788 } 789 return CEED_ERROR_SUCCESS; 790 } 791 792 //------------------------------------------------------------------------------ 793