xref: /libCEED/backends/cuda-ref/ceed-cuda-ref-restriction.c (revision b3d03e38aa94038a8b9805419605106d8c381b6f)
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 <cuda.h>
12 #include <cuda_runtime.h>
13 #include <stdbool.h>
14 #include <stddef.h>
15 #include <string.h>
16 
17 #include "../cuda/ceed-cuda-common.h"
18 #include "../cuda/ceed-cuda-compile.h"
19 #include "ceed-cuda-ref.h"
20 
21 //------------------------------------------------------------------------------
22 // Core apply restriction code
23 //------------------------------------------------------------------------------
24 static inline int CeedElemRestrictionApply_Cuda_Core(CeedElemRestriction rstr, CeedTransposeMode t_mode, bool use_signs, bool use_orients,
25                                                      CeedVector u, CeedVector v, CeedRequest *request) {
26   Ceed                      ceed;
27   CeedInt                   num_elem, elem_size;
28   CeedRestrictionType       rstr_type;
29   const CeedScalar         *d_u;
30   CeedScalar               *d_v;
31   CeedElemRestriction_Cuda *impl;
32   CUfunction                kernel;
33 
34   CeedCallBackend(CeedElemRestrictionGetData(rstr, &impl));
35   CeedCallBackend(CeedElemRestrictionGetCeed(rstr, &ceed));
36   CeedCallBackend(CeedElemRestrictionGetNumElements(rstr, &num_elem));
37   CeedCallBackend(CeedElemRestrictionGetElementSize(rstr, &elem_size));
38   CeedCallBackend(CeedElemRestrictionGetType(rstr, &rstr_type));
39   const CeedInt num_nodes = impl->num_nodes;
40 
41   // Get vectors
42   CeedCallBackend(CeedVectorGetArrayRead(u, CEED_MEM_DEVICE, &d_u));
43   if (t_mode == CEED_TRANSPOSE) {
44     // Sum into for transpose mode, e-vec to l-vec
45     CeedCallBackend(CeedVectorGetArray(v, CEED_MEM_DEVICE, &d_v));
46   } else {
47     // Overwrite for notranspose mode, l-vec to e-vec
48     CeedCallBackend(CeedVectorGetArrayWrite(v, CEED_MEM_DEVICE, &d_v));
49   }
50 
51   // Restrict
52   if (t_mode == CEED_NOTRANSPOSE) {
53     // L-vector -> E-vector
54     const CeedInt block_size = elem_size < 1024 ? (elem_size > 32 ? elem_size : 32) : 1024;
55     const CeedInt grid       = CeedDivUpInt(num_nodes, block_size);
56 
57     switch (rstr_type) {
58       case CEED_RESTRICTION_STRIDED: {
59         kernel       = impl->StridedNoTranspose;
60         void *args[] = {&num_elem, &d_u, &d_v};
61 
62         CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
63       } break;
64       case CEED_RESTRICTION_STANDARD: {
65         kernel       = impl->OffsetNoTranspose;
66         void *args[] = {&num_elem, &impl->d_ind, &d_u, &d_v};
67 
68         CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
69       } break;
70       case CEED_RESTRICTION_ORIENTED: {
71         if (use_signs) {
72           kernel       = impl->OrientedNoTranspose;
73           void *args[] = {&num_elem, &impl->d_ind, &impl->d_orients, &d_u, &d_v};
74 
75           CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
76         } else {
77           kernel       = impl->OffsetNoTranspose;
78           void *args[] = {&num_elem, &impl->d_ind, &d_u, &d_v};
79 
80           CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
81         }
82       } break;
83       case CEED_RESTRICTION_CURL_ORIENTED: {
84         if (use_signs && use_orients) {
85           kernel       = impl->CurlOrientedNoTranspose;
86           void *args[] = {&num_elem, &impl->d_ind, &impl->d_curl_orients, &d_u, &d_v};
87 
88           CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
89         } else if (use_orients) {
90           kernel       = impl->CurlOrientedUnsignedNoTranspose;
91           void *args[] = {&num_elem, &impl->d_ind, &impl->d_curl_orients, &d_u, &d_v};
92 
93           CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
94         } else {
95           kernel       = impl->OffsetNoTranspose;
96           void *args[] = {&num_elem, &impl->d_ind, &d_u, &d_v};
97 
98           CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
99         }
100       } break;
101       case CEED_RESTRICTION_POINTS: {
102         // LCOV_EXCL_START
103         return CeedError(ceed, CEED_ERROR_UNSUPPORTED, "Backend does not implement restriction CeedElemRestrictionAtPoints");
104         // LCOV_EXCL_STOP
105       } break;
106     }
107   } else {
108     // E-vector -> L-vector
109     const CeedInt block_size = 32;
110     const CeedInt grid       = CeedDivUpInt(num_nodes, block_size);
111 
112     switch (rstr_type) {
113       case CEED_RESTRICTION_STRIDED: {
114         kernel       = impl->StridedTranspose;
115         void *args[] = {&num_elem, &d_u, &d_v};
116 
117         CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
118       } break;
119       case CEED_RESTRICTION_STANDARD: {
120         if (impl->OffsetTranspose) {
121           kernel       = impl->OffsetTranspose;
122           void *args[] = {&num_elem, &impl->d_ind, &d_u, &d_v};
123 
124           CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
125         } else {
126           kernel       = impl->OffsetTransposeDet;
127           void *args[] = {&impl->d_l_vec_indices, &impl->d_t_indices, &impl->d_t_offsets, &d_u, &d_v};
128 
129           CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
130         }
131       } break;
132       case CEED_RESTRICTION_ORIENTED: {
133         if (use_signs) {
134           kernel       = impl->OrientedTranspose;
135           void *args[] = {&num_elem, &impl->d_ind, &impl->d_orients, &d_u, &d_v};
136 
137           CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
138         } else {
139           if (impl->OffsetTranspose) {
140             kernel       = impl->OffsetTranspose;
141             void *args[] = {&num_elem, &impl->d_ind, &d_u, &d_v};
142 
143             CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
144           } else {
145             kernel       = impl->OffsetTransposeDet;
146             void *args[] = {&impl->d_l_vec_indices, &impl->d_t_indices, &impl->d_t_offsets, &d_u, &d_v};
147 
148             CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
149           }
150         }
151       } break;
152       case CEED_RESTRICTION_CURL_ORIENTED: {
153         if (use_signs && use_orients) {
154           kernel       = impl->CurlOrientedTranspose;
155           void *args[] = {&num_elem, &impl->d_ind, &impl->d_curl_orients, &d_u, &d_v};
156 
157           CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
158         } else if (use_orients) {
159           kernel       = impl->CurlOrientedUnsignedTranspose;
160           void *args[] = {&num_elem, &impl->d_ind, &impl->d_curl_orients, &d_u, &d_v};
161 
162           CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
163         } else {
164           if (impl->OffsetTranspose) {
165             kernel       = impl->OffsetTranspose;
166             void *args[] = {&num_elem, &impl->d_ind, &d_u, &d_v};
167 
168             CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
169           } else {
170             kernel       = impl->OffsetTransposeDet;
171             void *args[] = {&impl->d_l_vec_indices, &impl->d_t_indices, &impl->d_t_offsets, &d_u, &d_v};
172 
173             CeedCallBackend(CeedRunKernel_Cuda(ceed, kernel, grid, block_size, args));
174           }
175         }
176       } break;
177       case CEED_RESTRICTION_POINTS: {
178         // LCOV_EXCL_START
179         return CeedError(ceed, CEED_ERROR_UNSUPPORTED, "Backend does not implement restriction CeedElemRestrictionAtPoints");
180         // LCOV_EXCL_STOP
181       } break;
182     }
183   }
184 
185   if (request != CEED_REQUEST_IMMEDIATE && request != CEED_REQUEST_ORDERED) *request = NULL;
186 
187   // Restore arrays
188   CeedCallBackend(CeedVectorRestoreArrayRead(u, &d_u));
189   CeedCallBackend(CeedVectorRestoreArray(v, &d_v));
190   return CEED_ERROR_SUCCESS;
191 }
192 
193 //------------------------------------------------------------------------------
194 // Apply restriction
195 //------------------------------------------------------------------------------
196 static int CeedElemRestrictionApply_Cuda(CeedElemRestriction rstr, CeedTransposeMode t_mode, CeedVector u, CeedVector v, CeedRequest *request) {
197   return CeedElemRestrictionApply_Cuda_Core(rstr, t_mode, true, true, u, v, request);
198 }
199 
200 //------------------------------------------------------------------------------
201 // Apply unsigned restriction
202 //------------------------------------------------------------------------------
203 static int CeedElemRestrictionApplyUnsigned_Cuda(CeedElemRestriction rstr, CeedTransposeMode t_mode, CeedVector u, CeedVector v,
204                                                  CeedRequest *request) {
205   return CeedElemRestrictionApply_Cuda_Core(rstr, t_mode, false, true, u, v, request);
206 }
207 
208 //------------------------------------------------------------------------------
209 // Apply unoriented restriction
210 //------------------------------------------------------------------------------
211 static int CeedElemRestrictionApplyUnoriented_Cuda(CeedElemRestriction rstr, CeedTransposeMode t_mode, CeedVector u, CeedVector v,
212                                                    CeedRequest *request) {
213   return CeedElemRestrictionApply_Cuda_Core(rstr, t_mode, false, false, u, v, request);
214 }
215 
216 //------------------------------------------------------------------------------
217 // Get offsets
218 //------------------------------------------------------------------------------
219 static int CeedElemRestrictionGetOffsets_Cuda(CeedElemRestriction rstr, CeedMemType mem_type, const CeedInt **offsets) {
220   CeedElemRestriction_Cuda *impl;
221 
222   CeedCallBackend(CeedElemRestrictionGetData(rstr, &impl));
223   switch (mem_type) {
224     case CEED_MEM_HOST:
225       *offsets = impl->h_ind;
226       break;
227     case CEED_MEM_DEVICE:
228       *offsets = impl->d_ind;
229       break;
230   }
231   return CEED_ERROR_SUCCESS;
232 }
233 
234 //------------------------------------------------------------------------------
235 // Get orientations
236 //------------------------------------------------------------------------------
237 static int CeedElemRestrictionGetOrientations_Cuda(CeedElemRestriction rstr, CeedMemType mem_type, const bool **orients) {
238   CeedElemRestriction_Cuda *impl;
239   CeedCallBackend(CeedElemRestrictionGetData(rstr, &impl));
240 
241   switch (mem_type) {
242     case CEED_MEM_HOST:
243       *orients = impl->h_orients;
244       break;
245     case CEED_MEM_DEVICE:
246       *orients = impl->d_orients;
247       break;
248   }
249   return CEED_ERROR_SUCCESS;
250 }
251 
252 //------------------------------------------------------------------------------
253 // Get curl-conforming orientations
254 //------------------------------------------------------------------------------
255 static int CeedElemRestrictionGetCurlOrientations_Cuda(CeedElemRestriction rstr, CeedMemType mem_type, const CeedInt8 **curl_orients) {
256   CeedElemRestriction_Cuda *impl;
257   CeedCallBackend(CeedElemRestrictionGetData(rstr, &impl));
258 
259   switch (mem_type) {
260     case CEED_MEM_HOST:
261       *curl_orients = impl->h_curl_orients;
262       break;
263     case CEED_MEM_DEVICE:
264       *curl_orients = impl->d_curl_orients;
265       break;
266   }
267   return CEED_ERROR_SUCCESS;
268 }
269 
270 //------------------------------------------------------------------------------
271 // Destroy restriction
272 //------------------------------------------------------------------------------
273 static int CeedElemRestrictionDestroy_Cuda(CeedElemRestriction rstr) {
274   Ceed                      ceed;
275   CeedElemRestriction_Cuda *impl;
276 
277   CeedCallBackend(CeedElemRestrictionGetData(rstr, &impl));
278   CeedCallBackend(CeedElemRestrictionGetCeed(rstr, &ceed));
279   CeedCallCuda(ceed, cuModuleUnload(impl->module));
280   CeedCallBackend(CeedFree(&impl->h_ind_allocated));
281   CeedCallCuda(ceed, cudaFree(impl->d_ind_allocated));
282   CeedCallCuda(ceed, cudaFree(impl->d_t_offsets));
283   CeedCallCuda(ceed, cudaFree(impl->d_t_indices));
284   CeedCallCuda(ceed, cudaFree(impl->d_l_vec_indices));
285   CeedCallBackend(CeedFree(&impl->h_orients_allocated));
286   CeedCallCuda(ceed, cudaFree(impl->d_orients_allocated));
287   CeedCallBackend(CeedFree(&impl->h_curl_orients_allocated));
288   CeedCallCuda(ceed, cudaFree(impl->d_curl_orients_allocated));
289   CeedCallBackend(CeedFree(&impl));
290   return CEED_ERROR_SUCCESS;
291 }
292 
293 //------------------------------------------------------------------------------
294 // Create transpose offsets and indices
295 //------------------------------------------------------------------------------
296 static int CeedElemRestrictionOffset_Cuda(const CeedElemRestriction rstr, const CeedInt *indices) {
297   Ceed                      ceed;
298   bool                     *is_node;
299   CeedSize                  l_size;
300   CeedInt                   num_elem, elem_size, num_comp, num_nodes = 0;
301   CeedInt                  *ind_to_offset, *l_vec_indices, *t_offsets, *t_indices;
302   CeedElemRestriction_Cuda *impl;
303 
304   CeedCallBackend(CeedElemRestrictionGetCeed(rstr, &ceed));
305   CeedCallBackend(CeedElemRestrictionGetData(rstr, &impl));
306   CeedCallBackend(CeedElemRestrictionGetNumElements(rstr, &num_elem));
307   CeedCallBackend(CeedElemRestrictionGetElementSize(rstr, &elem_size));
308   CeedCallBackend(CeedElemRestrictionGetLVectorSize(rstr, &l_size));
309   CeedCallBackend(CeedElemRestrictionGetNumComponents(rstr, &num_comp));
310   const CeedInt size_indices = num_elem * elem_size;
311 
312   // Count num_nodes
313   CeedCallBackend(CeedCalloc(l_size, &is_node));
314 
315   for (CeedInt i = 0; i < size_indices; i++) is_node[indices[i]] = 1;
316   for (CeedInt i = 0; i < l_size; i++) num_nodes += is_node[i];
317   impl->num_nodes = num_nodes;
318 
319   // L-vector offsets array
320   CeedCallBackend(CeedCalloc(l_size, &ind_to_offset));
321   CeedCallBackend(CeedCalloc(num_nodes, &l_vec_indices));
322   for (CeedInt i = 0, j = 0; i < l_size; i++) {
323     if (is_node[i]) {
324       l_vec_indices[j] = i;
325       ind_to_offset[i] = j++;
326     }
327   }
328   CeedCallBackend(CeedFree(&is_node));
329 
330   // Compute transpose offsets and indices
331   const CeedInt size_offsets = num_nodes + 1;
332 
333   CeedCallBackend(CeedCalloc(size_offsets, &t_offsets));
334   CeedCallBackend(CeedMalloc(size_indices, &t_indices));
335   // Count node multiplicity
336   for (CeedInt e = 0; e < num_elem; ++e) {
337     for (CeedInt i = 0; i < elem_size; ++i) ++t_offsets[ind_to_offset[indices[elem_size * e + i]] + 1];
338   }
339   // Convert to running sum
340   for (CeedInt i = 1; i < size_offsets; ++i) t_offsets[i] += t_offsets[i - 1];
341   // List all E-vec indices associated with L-vec node
342   for (CeedInt e = 0; e < num_elem; ++e) {
343     for (CeedInt i = 0; i < elem_size; ++i) {
344       const CeedInt lid = elem_size * e + i;
345       const CeedInt gid = indices[lid];
346 
347       t_indices[t_offsets[ind_to_offset[gid]]++] = lid;
348     }
349   }
350   // Reset running sum
351   for (int i = size_offsets - 1; i > 0; --i) t_offsets[i] = t_offsets[i - 1];
352   t_offsets[0] = 0;
353 
354   // Copy data to device
355   // -- L-vector indices
356   CeedCallCuda(ceed, cudaMalloc((void **)&impl->d_l_vec_indices, num_nodes * sizeof(CeedInt)));
357   CeedCallCuda(ceed, cudaMemcpy(impl->d_l_vec_indices, l_vec_indices, num_nodes * sizeof(CeedInt), cudaMemcpyHostToDevice));
358   // -- Transpose offsets
359   CeedCallCuda(ceed, cudaMalloc((void **)&impl->d_t_offsets, size_offsets * sizeof(CeedInt)));
360   CeedCallCuda(ceed, cudaMemcpy(impl->d_t_offsets, t_offsets, size_offsets * sizeof(CeedInt), cudaMemcpyHostToDevice));
361   // -- Transpose indices
362   CeedCallCuda(ceed, cudaMalloc((void **)&impl->d_t_indices, size_indices * sizeof(CeedInt)));
363   CeedCallCuda(ceed, cudaMemcpy(impl->d_t_indices, t_indices, size_indices * sizeof(CeedInt), cudaMemcpyHostToDevice));
364 
365   // Cleanup
366   CeedCallBackend(CeedFree(&ind_to_offset));
367   CeedCallBackend(CeedFree(&l_vec_indices));
368   CeedCallBackend(CeedFree(&t_offsets));
369   CeedCallBackend(CeedFree(&t_indices));
370   return CEED_ERROR_SUCCESS;
371 }
372 
373 //------------------------------------------------------------------------------
374 // Create restriction
375 //------------------------------------------------------------------------------
376 int CeedElemRestrictionCreate_Cuda(CeedMemType mem_type, CeedCopyMode copy_mode, const CeedInt *indices, const bool *orients,
377                                    const CeedInt8 *curl_orients, CeedElemRestriction rstr) {
378   Ceed                      ceed, ceed_parent;
379   bool                      is_deterministic;
380   CeedInt                   num_elem, num_comp, elem_size, comp_stride = 1;
381   CeedRestrictionType       rstr_type;
382   char                     *restriction_kernel_path, *restriction_kernel_source;
383   CeedElemRestriction_Cuda *impl;
384 
385   CeedCallBackend(CeedElemRestrictionGetCeed(rstr, &ceed));
386   CeedCallBackend(CeedGetParent(ceed, &ceed_parent));
387   CeedCallBackend(CeedIsDeterministic(ceed_parent, &is_deterministic));
388   CeedCallBackend(CeedElemRestrictionGetNumElements(rstr, &num_elem));
389   CeedCallBackend(CeedElemRestrictionGetNumComponents(rstr, &num_comp));
390   CeedCallBackend(CeedElemRestrictionGetElementSize(rstr, &elem_size));
391   const CeedInt size       = num_elem * elem_size;
392   CeedInt       strides[3] = {1, size, elem_size};
393   CeedInt       layout[3]  = {1, elem_size * num_elem, elem_size};
394 
395   // Stride data
396   CeedCallBackend(CeedElemRestrictionGetType(rstr, &rstr_type));
397   if (rstr_type == CEED_RESTRICTION_STRIDED) {
398     bool has_backend_strides;
399 
400     CeedCallBackend(CeedElemRestrictionHasBackendStrides(rstr, &has_backend_strides));
401     if (!has_backend_strides) {
402       CeedCallBackend(CeedElemRestrictionGetStrides(rstr, &strides));
403     }
404   } else {
405     CeedCallBackend(CeedElemRestrictionGetCompStride(rstr, &comp_stride));
406   }
407 
408   CeedCallBackend(CeedCalloc(1, &impl));
409   impl->num_nodes                = size;
410   impl->h_ind                    = NULL;
411   impl->h_ind_allocated          = NULL;
412   impl->d_ind                    = NULL;
413   impl->d_ind_allocated          = NULL;
414   impl->d_t_indices              = NULL;
415   impl->d_t_offsets              = NULL;
416   impl->h_orients                = NULL;
417   impl->h_orients_allocated      = NULL;
418   impl->d_orients                = NULL;
419   impl->d_orients_allocated      = NULL;
420   impl->h_curl_orients           = NULL;
421   impl->h_curl_orients_allocated = NULL;
422   impl->d_curl_orients           = NULL;
423   impl->d_curl_orients_allocated = NULL;
424   CeedCallBackend(CeedElemRestrictionSetData(rstr, impl));
425   CeedCallBackend(CeedElemRestrictionSetELayout(rstr, layout));
426 
427   // Set up device offset/orientation arrays
428   if (rstr_type != CEED_RESTRICTION_STRIDED) {
429     switch (mem_type) {
430       case CEED_MEM_HOST: {
431         switch (copy_mode) {
432           case CEED_OWN_POINTER:
433             impl->h_ind_allocated = (CeedInt *)indices;
434             impl->h_ind           = (CeedInt *)indices;
435             break;
436           case CEED_USE_POINTER:
437             impl->h_ind = (CeedInt *)indices;
438             break;
439           case CEED_COPY_VALUES:
440             CeedCallBackend(CeedMalloc(size, &impl->h_ind_allocated));
441             memcpy(impl->h_ind_allocated, indices, size * sizeof(CeedInt));
442             impl->h_ind = impl->h_ind_allocated;
443             break;
444         }
445         CeedCallCuda(ceed, cudaMalloc((void **)&impl->d_ind, size * sizeof(CeedInt)));
446         impl->d_ind_allocated = impl->d_ind;  // We own the device memory
447         CeedCallCuda(ceed, cudaMemcpy(impl->d_ind, indices, size * sizeof(CeedInt), cudaMemcpyHostToDevice));
448         if (is_deterministic) CeedCallBackend(CeedElemRestrictionOffset_Cuda(rstr, indices));
449       } break;
450       case CEED_MEM_DEVICE: {
451         switch (copy_mode) {
452           case CEED_COPY_VALUES:
453             CeedCallCuda(ceed, cudaMalloc((void **)&impl->d_ind, size * sizeof(CeedInt)));
454             impl->d_ind_allocated = impl->d_ind;  // We own the device memory
455             CeedCallCuda(ceed, cudaMemcpy(impl->d_ind, indices, size * sizeof(CeedInt), cudaMemcpyDeviceToDevice));
456             break;
457           case CEED_OWN_POINTER:
458             impl->d_ind           = (CeedInt *)indices;
459             impl->d_ind_allocated = impl->d_ind;
460             break;
461           case CEED_USE_POINTER:
462             impl->d_ind = (CeedInt *)indices;
463             break;
464         }
465         CeedCallBackend(CeedMalloc(size, &impl->h_ind_allocated));
466         CeedCallCuda(ceed, cudaMemcpy(impl->h_ind_allocated, impl->d_ind, size * sizeof(CeedInt), cudaMemcpyDeviceToHost));
467         impl->h_ind = impl->h_ind_allocated;
468         if (is_deterministic) CeedCallBackend(CeedElemRestrictionOffset_Cuda(rstr, indices));
469       } break;
470     }
471 
472     // Orientation data
473     if (rstr_type == CEED_RESTRICTION_ORIENTED) {
474       switch (mem_type) {
475         case CEED_MEM_HOST: {
476           switch (copy_mode) {
477             case CEED_OWN_POINTER:
478               impl->h_orients_allocated = (bool *)orients;
479               impl->h_orients           = (bool *)orients;
480               break;
481             case CEED_USE_POINTER:
482               impl->h_orients = (bool *)orients;
483               break;
484             case CEED_COPY_VALUES:
485               CeedCallBackend(CeedMalloc(size, &impl->h_orients_allocated));
486               memcpy(impl->h_orients_allocated, orients, size * sizeof(bool));
487               impl->h_orients = impl->h_orients_allocated;
488               break;
489           }
490           CeedCallCuda(ceed, cudaMalloc((void **)&impl->d_orients, size * sizeof(bool)));
491           impl->d_orients_allocated = impl->d_orients;  // We own the device memory
492           CeedCallCuda(ceed, cudaMemcpy(impl->d_orients, orients, size * sizeof(bool), cudaMemcpyHostToDevice));
493         } break;
494         case CEED_MEM_DEVICE: {
495           switch (copy_mode) {
496             case CEED_COPY_VALUES:
497               CeedCallCuda(ceed, cudaMalloc((void **)&impl->d_orients, size * sizeof(bool)));
498               impl->d_orients_allocated = impl->d_orients;  // We own the device memory
499               CeedCallCuda(ceed, cudaMemcpy(impl->d_orients, orients, size * sizeof(bool), cudaMemcpyDeviceToDevice));
500               break;
501             case CEED_OWN_POINTER:
502               impl->d_orients           = (bool *)orients;
503               impl->d_orients_allocated = impl->d_orients;
504               break;
505             case CEED_USE_POINTER:
506               impl->d_orients = (bool *)orients;
507               break;
508           }
509           CeedCallBackend(CeedMalloc(size, &impl->h_orients_allocated));
510           CeedCallCuda(ceed, cudaMemcpy(impl->h_orients_allocated, impl->d_orients, size * sizeof(bool), cudaMemcpyDeviceToHost));
511           impl->h_orients = impl->h_orients_allocated;
512         } break;
513       }
514     } else if (rstr_type == CEED_RESTRICTION_CURL_ORIENTED) {
515       switch (mem_type) {
516         case CEED_MEM_HOST: {
517           switch (copy_mode) {
518             case CEED_OWN_POINTER:
519               impl->h_curl_orients_allocated = (CeedInt8 *)curl_orients;
520               impl->h_curl_orients           = (CeedInt8 *)curl_orients;
521               break;
522             case CEED_USE_POINTER:
523               impl->h_curl_orients = (CeedInt8 *)curl_orients;
524               break;
525             case CEED_COPY_VALUES:
526               CeedCallBackend(CeedMalloc(3 * size, &impl->h_curl_orients_allocated));
527               memcpy(impl->h_curl_orients_allocated, curl_orients, 3 * size * sizeof(CeedInt8));
528               impl->h_curl_orients = impl->h_curl_orients_allocated;
529               break;
530           }
531           CeedCallCuda(ceed, cudaMalloc((void **)&impl->d_curl_orients, 3 * size * sizeof(CeedInt8)));
532           impl->d_curl_orients_allocated = impl->d_curl_orients;  // We own the device memory
533           CeedCallCuda(ceed, cudaMemcpy(impl->d_curl_orients, curl_orients, 3 * size * sizeof(CeedInt8), cudaMemcpyHostToDevice));
534         } break;
535         case CEED_MEM_DEVICE: {
536           switch (copy_mode) {
537             case CEED_COPY_VALUES:
538               CeedCallCuda(ceed, cudaMalloc((void **)&impl->d_curl_orients, 3 * size * sizeof(CeedInt8)));
539               impl->d_curl_orients_allocated = impl->d_curl_orients;  // We own the device memory
540               CeedCallCuda(ceed, cudaMemcpy(impl->d_curl_orients, curl_orients, 3 * size * sizeof(CeedInt8), cudaMemcpyDeviceToDevice));
541               break;
542             case CEED_OWN_POINTER:
543               impl->d_curl_orients           = (CeedInt8 *)curl_orients;
544               impl->d_curl_orients_allocated = impl->d_curl_orients;
545               break;
546             case CEED_USE_POINTER:
547               impl->d_curl_orients = (CeedInt8 *)curl_orients;
548               break;
549           }
550           CeedCallBackend(CeedMalloc(3 * size, &impl->h_curl_orients_allocated));
551           CeedCallCuda(ceed, cudaMemcpy(impl->h_curl_orients_allocated, impl->d_curl_orients, 3 * size * sizeof(CeedInt8), cudaMemcpyDeviceToHost));
552           impl->h_curl_orients = impl->h_curl_orients_allocated;
553         } break;
554       }
555     }
556   }
557 
558   // Compile CUDA kernels
559   CeedCallBackend(CeedGetJitAbsolutePath(ceed, "ceed/jit-source/cuda/cuda-ref-restriction.h", &restriction_kernel_path));
560   CeedDebug256(ceed, CEED_DEBUG_COLOR_SUCCESS, "----- Loading Restriction Kernel Source -----\n");
561   CeedCallBackend(CeedLoadSourceToBuffer(ceed, restriction_kernel_path, &restriction_kernel_source));
562   CeedDebug256(ceed, CEED_DEBUG_COLOR_SUCCESS, "----- Loading Restriction Kernel Source Complete! -----\n");
563   CeedCallBackend(CeedCompile_Cuda(ceed, restriction_kernel_source, &impl->module, 8, "RSTR_ELEM_SIZE", elem_size, "RSTR_NUM_ELEM", num_elem,
564                                    "RSTR_NUM_COMP", num_comp, "RSTR_NUM_NODES", impl->num_nodes, "RSTR_COMP_STRIDE", comp_stride, "RSTR_STRIDE_NODES",
565                                    strides[0], "RSTR_STRIDE_COMP", strides[1], "RSTR_STRIDE_ELEM", strides[2]));
566   CeedCallBackend(CeedGetKernel_Cuda(ceed, impl->module, "StridedNoTranspose", &impl->StridedNoTranspose));
567   CeedCallBackend(CeedGetKernel_Cuda(ceed, impl->module, "StridedTranspose", &impl->StridedTranspose));
568   CeedCallBackend(CeedGetKernel_Cuda(ceed, impl->module, "OffsetNoTranspose", &impl->OffsetNoTranspose));
569   if (!is_deterministic) {
570     CeedCallBackend(CeedGetKernel_Cuda(ceed, impl->module, "OffsetTranspose", &impl->OffsetTranspose));
571   } else {
572     CeedCallBackend(CeedGetKernel_Cuda(ceed, impl->module, "OffsetTransposeDet", &impl->OffsetTransposeDet));
573   }
574   CeedCallBackend(CeedGetKernel_Cuda(ceed, impl->module, "OrientedNoTranspose", &impl->OrientedNoTranspose));
575   CeedCallBackend(CeedGetKernel_Cuda(ceed, impl->module, "OrientedTranspose", &impl->OrientedTranspose));
576   CeedCallBackend(CeedGetKernel_Cuda(ceed, impl->module, "CurlOrientedNoTranspose", &impl->CurlOrientedNoTranspose));
577   CeedCallBackend(CeedGetKernel_Cuda(ceed, impl->module, "CurlOrientedUnsignedNoTranspose", &impl->CurlOrientedUnsignedNoTranspose));
578   CeedCallBackend(CeedGetKernel_Cuda(ceed, impl->module, "CurlOrientedTranspose", &impl->CurlOrientedTranspose));
579   CeedCallBackend(CeedGetKernel_Cuda(ceed, impl->module, "CurlOrientedUnsignedTranspose", &impl->CurlOrientedUnsignedTranspose));
580   CeedCallBackend(CeedFree(&restriction_kernel_path));
581   CeedCallBackend(CeedFree(&restriction_kernel_source));
582 
583   // Register backend functions
584   CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", rstr, "Apply", CeedElemRestrictionApply_Cuda));
585   CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", rstr, "ApplyUnsigned", CeedElemRestrictionApplyUnsigned_Cuda));
586   CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", rstr, "ApplyUnoriented", CeedElemRestrictionApplyUnoriented_Cuda));
587   CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", rstr, "GetOffsets", CeedElemRestrictionGetOffsets_Cuda));
588   CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", rstr, "GetOrientations", CeedElemRestrictionGetOrientations_Cuda));
589   CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", rstr, "GetCurlOrientations", CeedElemRestrictionGetCurlOrientations_Cuda));
590   CeedCallBackend(CeedSetBackendFunction(ceed, "ElemRestriction", rstr, "Destroy", CeedElemRestrictionDestroy_Cuda));
591   return CEED_ERROR_SUCCESS;
592 }
593 
594 //------------------------------------------------------------------------------
595