Actual source code: mpikok.kokkos.cxx
1: /*
2: This file contains routines for Parallel vector operations.
3: */
4: #include <petsc_kokkos.hpp>
5: #include <petscvec_kokkos.hpp>
6: #include <petsc/private/deviceimpl.h>
7: #include <petsc/private/vecimpl.h>
8: #include <../src/vec/vec/impls/mpi/pvecimpl.h>
9: #include <../src/vec/vec/impls/seq/kokkos/veckokkosimpl.hpp>
10: #include <petscsf.h>
12: static PetscErrorCode VecDestroy_MPIKokkos(Vec v)
13: {
14: PetscFunctionBegin;
15: delete static_cast<Vec_Kokkos *>(v->spptr);
16: PetscCall(VecDestroy_MPI(v));
17: PetscFunctionReturn(PETSC_SUCCESS);
18: }
20: static PetscErrorCode VecNorm_MPIKokkos(Vec xin, NormType type, PetscReal *z)
21: {
22: PetscFunctionBegin;
23: PetscCall(VecNorm_MPI_Default(xin, type, z, VecNorm_SeqKokkos));
24: PetscFunctionReturn(PETSC_SUCCESS);
25: }
27: static PetscErrorCode VecErrorWeightedNorms_MPIKokkos(Vec U, Vec Y, Vec E, NormType wnormtype, PetscReal atol, Vec vatol, PetscReal rtol, Vec vrtol, PetscReal ignore_max, PetscReal *norm, PetscInt *norm_loc, PetscReal *norma, PetscInt *norma_loc, PetscReal *normr, PetscInt *normr_loc)
28: {
29: PetscFunctionBegin;
30: PetscCall(VecErrorWeightedNorms_MPI_Default(U, Y, E, wnormtype, atol, vatol, rtol, vrtol, ignore_max, norm, norm_loc, norma, norma_loc, normr, normr_loc, VecErrorWeightedNorms_SeqKokkos));
31: PetscFunctionReturn(PETSC_SUCCESS);
32: }
34: /* z = y^H x */
35: static PetscErrorCode VecDot_MPIKokkos(Vec xin, Vec yin, PetscScalar *z)
36: {
37: PetscFunctionBegin;
38: PetscCall(VecXDot_MPI_Default(xin, yin, z, VecDot_SeqKokkos));
39: PetscFunctionReturn(PETSC_SUCCESS);
40: }
42: /* z = y^T x */
43: static PetscErrorCode VecTDot_MPIKokkos(Vec xin, Vec yin, PetscScalar *z)
44: {
45: PetscFunctionBegin;
46: PetscCall(VecXDot_MPI_Default(xin, yin, z, VecTDot_SeqKokkos));
47: PetscFunctionReturn(PETSC_SUCCESS);
48: }
50: static PetscErrorCode VecMDot_MPIKokkos(Vec xin, PetscInt nv, const Vec y[], PetscScalar *z)
51: {
52: PetscFunctionBegin;
53: PetscCall(VecMXDot_MPI_Default(xin, nv, y, z, VecMDot_SeqKokkos));
54: PetscFunctionReturn(PETSC_SUCCESS);
55: }
57: static PetscErrorCode VecMTDot_MPIKokkos(Vec xin, PetscInt nv, const Vec y[], PetscScalar *z)
58: {
59: PetscFunctionBegin;
60: PetscCall(VecMXDot_MPI_Default(xin, nv, y, z, VecMTDot_SeqKokkos));
61: PetscFunctionReturn(PETSC_SUCCESS);
62: }
64: static PetscErrorCode VecMDot_MPIKokkos_GEMV(Vec xin, PetscInt nv, const Vec y[], PetscScalar *z)
65: {
66: PetscFunctionBegin;
67: PetscCall(VecMXDot_MPI_Default(xin, nv, y, z, VecMDot_SeqKokkos_GEMV));
68: PetscFunctionReturn(PETSC_SUCCESS);
69: }
71: static PetscErrorCode VecMTDot_MPIKokkos_GEMV(Vec xin, PetscInt nv, const Vec y[], PetscScalar *z)
72: {
73: PetscFunctionBegin;
74: PetscCall(VecMXDot_MPI_Default(xin, nv, y, z, VecMTDot_SeqKokkos_GEMV));
75: PetscFunctionReturn(PETSC_SUCCESS);
76: }
78: static PetscErrorCode VecMax_MPIKokkos(Vec xin, PetscInt *idx, PetscReal *z)
79: {
80: const MPI_Op ops[] = {MPIU_MAXLOC, MPIU_MAX};
82: PetscFunctionBegin;
83: PetscCall(VecMinMax_MPI_Default(xin, idx, z, VecMax_SeqKokkos, ops));
84: PetscFunctionReturn(PETSC_SUCCESS);
85: }
87: static PetscErrorCode VecMin_MPIKokkos(Vec xin, PetscInt *idx, PetscReal *z)
88: {
89: const MPI_Op ops[] = {MPIU_MINLOC, MPIU_MIN};
91: PetscFunctionBegin;
92: PetscCall(VecMinMax_MPI_Default(xin, idx, z, VecMin_SeqKokkos, ops));
93: PetscFunctionReturn(PETSC_SUCCESS);
94: }
96: static PetscErrorCode VecDuplicate_MPIKokkos(Vec win, Vec *vv)
97: {
98: Vec v;
99: Vec_Kokkos *veckok;
100: Vec_MPI *wdata = (Vec_MPI *)win->data;
102: PetscScalarKokkosDualView w_dual;
104: PetscFunctionBegin;
105: PetscCallCXX(w_dual = PetscScalarKokkosDualView("w_dual", win->map->n + wdata->nghost)); // Kokkos init's v_dual to zero
107: /* Reuse VecDuplicate_MPI, which contains a lot of stuff */
108: PetscCall(VecDuplicateWithArray_MPI(win, w_dual.view_host().data(), &v)); /* after the call, v is a VECMPI */
109: PetscCall(PetscObjectChangeTypeName((PetscObject)v, VECMPIKOKKOS));
110: v->ops[0] = win->ops[0];
112: /* Build the Vec_Kokkos struct */
113: veckok = new Vec_Kokkos(v->map->n, w_dual.view_host().data(), w_dual.view_device().data());
114: veckok->w_dual = w_dual;
115: v->spptr = veckok;
116: v->offloadmask = PETSC_OFFLOAD_KOKKOS;
117: *vv = v;
118: PetscFunctionReturn(PETSC_SUCCESS);
119: }
121: static PetscErrorCode VecDotNorm2_MPIKokkos(Vec s, Vec t, PetscScalar *dp, PetscScalar *nm)
122: {
123: PetscFunctionBegin;
124: PetscCall(VecDotNorm2_MPI_Default(s, t, dp, nm, VecDotNorm2_SeqKokkos));
125: PetscFunctionReturn(PETSC_SUCCESS);
126: }
128: static PetscErrorCode VecGetSubVector_MPIKokkos(Vec x, IS is, Vec *y)
129: {
130: PetscFunctionBegin;
131: PetscCall(VecGetSubVector_Kokkos_Private(x, PETSC_TRUE, is, y));
132: PetscFunctionReturn(PETSC_SUCCESS);
133: }
135: static PetscErrorCode VecSetPreallocationCOO_MPIKokkos(Vec x, PetscCount ncoo, const PetscInt coo_i[])
136: {
137: const auto vecmpi = static_cast<Vec_MPI *>(x->data);
138: const auto veckok = static_cast<Vec_Kokkos *>(x->spptr);
139: PetscInt m;
141: PetscFunctionBegin;
142: PetscCall(VecGetLocalSize(x, &m));
143: PetscCall(VecSetPreallocationCOO_MPI(x, ncoo, coo_i));
144: PetscCall(veckok->SetUpCOO(vecmpi, m));
145: PetscFunctionReturn(PETSC_SUCCESS);
146: }
148: static PetscErrorCode VecSetValuesCOO_MPIKokkos(Vec x, const PetscScalar v[], InsertMode imode)
149: {
150: const auto vecmpi = static_cast<Vec_MPI *>(x->data);
151: const auto veckok = static_cast<Vec_Kokkos *>(x->spptr);
152: const PetscCountKokkosView &jmap1 = veckok->jmap1_d;
153: const PetscCountKokkosView &perm1 = veckok->perm1_d;
154: const PetscCountKokkosView &imap2 = veckok->imap2_d;
155: const PetscCountKokkosView &jmap2 = veckok->jmap2_d;
156: const PetscCountKokkosView &perm2 = veckok->perm2_d;
157: const PetscCountKokkosView &Cperm = veckok->Cperm_d;
158: PetscScalarKokkosView &sendbuf = veckok->sendbuf_d;
159: PetscScalarKokkosView &recvbuf = veckok->recvbuf_d;
160: PetscScalarKokkosView xv;
161: ConstPetscScalarKokkosView vv;
162: PetscMemType memtype;
163: PetscInt m;
165: PetscFunctionBegin;
166: PetscCall(VecGetLocalSize(x, &m));
167: PetscCall(PetscGetMemType(v, &memtype));
168: if (PetscMemTypeHost(memtype)) { /* If user gave v[] in host, we might need to copy it to device if any */
169: vv = Kokkos::create_mirror_view_and_copy(DefaultMemorySpace(), PetscScalarKokkosViewHost(const_cast<PetscScalar *>(v), vecmpi->coo_n));
170: } else {
171: vv = ConstPetscScalarKokkosView(v, vecmpi->coo_n); /* Directly use v[]'s memory */
172: }
174: /* Pack entries to be sent to remote */
175: Kokkos::parallel_for(Kokkos::RangePolicy<>(PetscGetKokkosExecutionSpace(), 0, vecmpi->sendlen), KOKKOS_LAMBDA(const PetscCount i) { sendbuf(i) = vv(Cperm(i)); });
176: PetscCall(PetscSFReduceWithMemTypeBegin(vecmpi->coo_sf, MPIU_SCALAR, PETSC_MEMTYPE_KOKKOS, sendbuf.data(), PETSC_MEMTYPE_KOKKOS, recvbuf.data(), MPI_REPLACE));
178: if (imode == INSERT_VALUES) PetscCall(VecGetKokkosViewWrite(x, &xv)); /* write vector */
179: else PetscCall(VecGetKokkosView(x, &xv)); /* read & write vector */
181: Kokkos::parallel_for(
182: Kokkos::RangePolicy<>(PetscGetKokkosExecutionSpace(), 0, m), KOKKOS_LAMBDA(const PetscCount i) {
183: PetscScalar sum = 0.0;
184: for (PetscCount k = jmap1(i); k < jmap1(i + 1); k++) sum += vv(perm1(k));
185: xv(i) = (imode == INSERT_VALUES ? 0.0 : xv(i)) + sum;
186: });
188: PetscCall(PetscSFReduceEnd(vecmpi->coo_sf, MPIU_SCALAR, sendbuf.data(), recvbuf.data(), MPI_REPLACE));
190: /* Add received remote entries */
191: Kokkos::parallel_for(
192: Kokkos::RangePolicy<>(PetscGetKokkosExecutionSpace(), 0, vecmpi->nnz2), KOKKOS_LAMBDA(PetscCount i) {
193: for (PetscCount k = jmap2(i); k < jmap2(i + 1); k++) xv(imap2(i)) += recvbuf(perm2(k));
194: });
196: if (imode == INSERT_VALUES) PetscCall(VecRestoreKokkosViewWrite(x, &xv));
197: else PetscCall(VecRestoreKokkosView(x, &xv));
198: PetscFunctionReturn(PETSC_SUCCESS);
199: }
201: PetscErrorCode VecSetOps_MPIKokkos(Vec v)
202: {
203: PetscFunctionBegin;
204: v->ops->abs = VecAbs_SeqKokkos;
205: v->ops->reciprocal = VecReciprocal_SeqKokkos;
206: v->ops->pointwisemult = VecPointwiseMult_SeqKokkos;
207: v->ops->setrandom = VecSetRandom_SeqKokkos;
208: v->ops->dotnorm2 = VecDotNorm2_MPIKokkos;
209: v->ops->waxpy = VecWAXPY_SeqKokkos;
210: v->ops->norm = VecNorm_MPIKokkos;
211: v->ops->min = VecMin_MPIKokkos;
212: v->ops->max = VecMax_MPIKokkos;
213: v->ops->sum = VecSum_SeqKokkos;
214: v->ops->shift = VecShift_SeqKokkos;
215: v->ops->scale = VecScale_SeqKokkos;
216: v->ops->copy = VecCopy_SeqKokkos;
217: v->ops->set = VecSet_SeqKokkos;
218: v->ops->swap = VecSwap_SeqKokkos;
219: v->ops->axpy = VecAXPY_SeqKokkos;
220: v->ops->axpby = VecAXPBY_SeqKokkos;
221: v->ops->maxpy = VecMAXPY_SeqKokkos;
222: v->ops->aypx = VecAYPX_SeqKokkos;
223: v->ops->axpbypcz = VecAXPBYPCZ_SeqKokkos;
224: v->ops->pointwisedivide = VecPointwiseDivide_SeqKokkos;
225: v->ops->placearray = VecPlaceArray_SeqKokkos;
226: v->ops->replacearray = VecReplaceArray_SeqKokkos;
227: v->ops->resetarray = VecResetArray_SeqKokkos;
229: v->ops->dot = VecDot_MPIKokkos;
230: v->ops->tdot = VecTDot_MPIKokkos;
231: v->ops->mdot = VecMDot_MPIKokkos;
232: v->ops->mtdot = VecMTDot_MPIKokkos;
234: v->ops->dot_local = VecDot_SeqKokkos;
235: v->ops->tdot_local = VecTDot_SeqKokkos;
236: v->ops->mdot_local = VecMDot_SeqKokkos;
237: v->ops->mtdot_local = VecMTDot_SeqKokkos;
239: v->ops->norm_local = VecNorm_SeqKokkos;
240: v->ops->duplicate = VecDuplicate_MPIKokkos;
241: v->ops->destroy = VecDestroy_MPIKokkos;
242: v->ops->getlocalvector = VecGetLocalVector_SeqKokkos;
243: v->ops->restorelocalvector = VecRestoreLocalVector_SeqKokkos;
244: v->ops->getlocalvectorread = VecGetLocalVector_SeqKokkos;
245: v->ops->restorelocalvectorread = VecRestoreLocalVector_SeqKokkos;
246: v->ops->getarraywrite = VecGetArrayWrite_SeqKokkos;
247: v->ops->getarray = VecGetArray_SeqKokkos;
248: v->ops->restorearray = VecRestoreArray_SeqKokkos;
249: v->ops->getarrayandmemtype = VecGetArrayAndMemType_SeqKokkos;
250: v->ops->restorearrayandmemtype = VecRestoreArrayAndMemType_SeqKokkos;
251: v->ops->getarraywriteandmemtype = VecGetArrayWriteAndMemType_SeqKokkos;
252: v->ops->getsubvector = VecGetSubVector_MPIKokkos;
253: v->ops->restoresubvector = VecRestoreSubVector_SeqKokkos;
255: v->ops->setpreallocationcoo = VecSetPreallocationCOO_MPIKokkos;
256: v->ops->setvaluescoo = VecSetValuesCOO_MPIKokkos;
258: v->ops->errorwnorm = VecErrorWeightedNorms_MPIKokkos;
259: PetscFunctionReturn(PETSC_SUCCESS);
260: }
262: PETSC_INTERN PetscErrorCode VecConvert_MPI_MPIKokkos_inplace(Vec v)
263: {
264: Vec_MPI *vecmpi;
266: PetscFunctionBegin;
267: PetscCall(PetscKokkosInitializeCheck());
268: PetscCall(PetscLayoutSetUp(v->map));
269: PetscCall(PetscObjectChangeTypeName((PetscObject)v, VECMPIKOKKOS));
270: PetscCall(VecSetOps_MPIKokkos(v));
271: PetscCheck(!v->spptr, PETSC_COMM_SELF, PETSC_ERR_PLIB, "v->spptr not NULL");
272: vecmpi = static_cast<Vec_MPI *>(v->data);
273: PetscCallCXX(v->spptr = new Vec_Kokkos(v->map->n, vecmpi->array, NULL));
274: v->offloadmask = PETSC_OFFLOAD_KOKKOS;
275: PetscFunctionReturn(PETSC_SUCCESS);
276: }
278: // Duplicate a VECMPIKOKKOS
279: static PetscErrorCode VecDuplicateVecs_MPIKokkos_GEMV(Vec w, PetscInt m, Vec *V[])
280: {
281: PetscInt64 lda; // use 64-bit as we will do "m * lda"
282: PetscScalar *array_h, *array_d;
283: PetscLayout map;
284: Vec_MPI *wmpi = (Vec_MPI *)w->data;
285: PetscScalarKokkosDualView w_dual;
287: PetscFunctionBegin;
288: PetscCall(PetscKokkosInitializeCheck()); // as we'll call kokkos_malloc()
289: if (wmpi->nghost) { // currently only do GEMV optimiation for vectors without ghosts
290: w->ops->duplicatevecs = VecDuplicateVecs_Default;
291: PetscCall(VecDuplicateVecs(w, m, V));
292: } else {
293: PetscCall(PetscMalloc1(m, V));
294: PetscCall(VecGetLayout(w, &map));
295: VecGetLocalSizeAligned(w, 64, &lda); // get in lda the 64-bytes aligned local size
297: // See comments in VecCreate_SeqKokkos() on why we use DualView to allocate the memory
298: PetscCallCXX(w_dual = PetscScalarKokkosDualView("VecDuplicateVecs", m * lda)); // Kokkos init's w_dual to zero
300: // create the m vectors with raw arrays
301: array_h = w_dual.view_host().data();
302: array_d = w_dual.view_device().data();
303: for (PetscInt i = 0; i < m; i++) {
304: Vec v;
305: PetscCall(VecCreateMPIKokkosWithLayoutAndArrays_Private(map, &array_h[i * lda], &array_d[i * lda], &v));
306: PetscCallCXX(static_cast<Vec_Kokkos *>(v->spptr)->v_dual.modify_host()); // as we only init'ed array_h
307: PetscCall(PetscObjectListDuplicate(((PetscObject)w)->olist, &((PetscObject)v)->olist));
308: PetscCall(PetscFunctionListDuplicate(((PetscObject)w)->qlist, &((PetscObject)v)->qlist));
309: v->ops[0] = w->ops[0];
310: v->stash.donotstash = w->stash.donotstash;
311: v->stash.ignorenegidx = w->stash.ignorenegidx;
312: v->stash.bs = w->stash.bs;
313: v->bstash.bs = w->bstash.bs;
314: (*V)[i] = v;
315: }
317: // let the first vector own the raw arrays, so when it is destroyed it will free the arrays
318: if (m) {
319: Vec v = (*V)[0];
321: static_cast<Vec_Kokkos *>(v->spptr)->w_dual = w_dual; // stash the memory
322: // disable replacearray of the first vector, as freeing its memory also frees others in the group.
323: // But replacearray of others is ok, as they don't own their array.
324: if (m > 1) v->ops->replacearray = VecReplaceArray_Default_GEMV_Error;
325: }
326: }
327: PetscFunctionReturn(PETSC_SUCCESS);
328: }
330: /*MC
331: VECMPIKOKKOS - VECMPIKOKKOS = "mpikokkos" - The basic parallel vector, modified to use Kokkos
333: Options Database Keys:
334: . -vec_type mpikokkos - sets the vector type to VECMPIKOKKOS during a call to VecSetFromOptions()
336: Level: beginner
338: .seealso: `VecCreate()`, `VecSetType()`, `VecSetFromOptions()`, `VecCreateMPIKokkosWithArray()`, `VECMPI`, `VecType`, `VecCreateMPI()`
339: M*/
340: PetscErrorCode VecCreate_MPIKokkos(Vec v)
341: {
342: PetscBool mdot_use_gemv = PETSC_TRUE;
343: PetscBool maxpy_use_gemv = PETSC_FALSE; // default is false as we saw bad performance with vendors' GEMV with tall skinny matrices.
344: PetscScalarKokkosDualView v_dual;
346: PetscFunctionBegin;
347: PetscCall(PetscKokkosInitializeCheck());
348: PetscCall(PetscLayoutSetUp(v->map));
350: PetscCallCXX(v_dual = PetscScalarKokkosDualView("v_dual", v->map->n)); // Kokkos init's v_dual to zero
351: PetscCall(VecCreate_MPI_Private(v, PETSC_FALSE, 0, v_dual.view_host().data()));
353: PetscCall(PetscObjectChangeTypeName((PetscObject)v, VECMPIKOKKOS));
354: PetscCall(VecSetOps_MPIKokkos(v));
355: PetscCheck(!v->spptr, PETSC_COMM_SELF, PETSC_ERR_PLIB, "v->spptr not NULL");
356: PetscCallCXX(v->spptr = new Vec_Kokkos(v_dual));
357: v->offloadmask = PETSC_OFFLOAD_KOKKOS;
358: PetscCall(PetscOptionsGetBool(NULL, NULL, "-vec_mdot_use_gemv", &mdot_use_gemv, NULL));
359: PetscCall(PetscOptionsGetBool(NULL, NULL, "-vec_maxpy_use_gemv", &maxpy_use_gemv, NULL));
361: // allocate multiple vectors together
362: if (mdot_use_gemv || maxpy_use_gemv) v->ops[0].duplicatevecs = VecDuplicateVecs_MPIKokkos_GEMV;
364: if (mdot_use_gemv) {
365: v->ops[0].mdot = VecMDot_MPIKokkos_GEMV;
366: v->ops[0].mtdot = VecMTDot_MPIKokkos_GEMV;
367: v->ops[0].mdot_local = VecMDot_SeqKokkos_GEMV;
368: v->ops[0].mtdot_local = VecMTDot_SeqKokkos_GEMV;
369: }
371: if (maxpy_use_gemv) v->ops[0].maxpy = VecMAXPY_SeqKokkos_GEMV;
372: PetscFunctionReturn(PETSC_SUCCESS);
373: }
375: // Create a VECMPIKOKKOS with layout and arrays
376: PetscErrorCode VecCreateMPIKokkosWithLayoutAndArrays_Private(PetscLayout map, const PetscScalar harray[], const PetscScalar darray[], Vec *v)
377: {
378: Vec w;
380: PetscFunctionBegin;
381: if (map->n > 0) PetscCheck(darray, map->comm, PETSC_ERR_ARG_WRONG, "darray cannot be NULL");
382: #if defined(KOKKOS_ENABLE_UNIFIED_MEMORY)
383: PetscCheck(harray == darray, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "harray and darray must be the same");
384: #endif
385: PetscCall(VecCreateMPIWithLayoutAndArray_Private(map, harray, &w));
386: PetscCall(PetscObjectChangeTypeName((PetscObject)w, VECMPIKOKKOS)); // Change it to VECKOKKOS
387: PetscCall(VecSetOps_MPIKokkos(w));
388: PetscCallCXX(w->spptr = new Vec_Kokkos(map->n, const_cast<PetscScalar *>(harray), const_cast<PetscScalar *>(darray)));
389: w->offloadmask = PETSC_OFFLOAD_KOKKOS;
390: *v = w;
391: PetscFunctionReturn(PETSC_SUCCESS);
392: }
394: /*@C
395: VecCreateMPIKokkosWithArray - Creates a parallel, array-style vector,
396: where the user provides the GPU array space to store the vector values.
398: Collective
400: Input Parameters:
401: + comm - the MPI communicator to use
402: . bs - block size, same meaning as VecSetBlockSize()
403: . n - local vector length, cannot be PETSC_DECIDE
404: . N - global vector length (or PETSC_DECIDE to have calculated)
405: - darray - the user provided GPU array to store the vector values
407: Output Parameter:
408: . v - the vector
410: Notes:
411: Use VecDuplicate() or VecDuplicateVecs() to form additional vectors of the
412: same type as an existing vector.
414: If the user-provided array is NULL, then VecKokkosPlaceArray() can be used
415: at a later stage to SET the array for storing the vector values.
417: PETSc does NOT free the array when the vector is destroyed via VecDestroy().
418: The user should not free the array until the vector is destroyed.
420: Level: intermediate
422: .seealso: `VecCreateSeqKokkosWithArray()`, `VecCreateMPIWithArray()`, `VecCreateSeqWithArray()`,
423: `VecCreate()`, `VecDuplicate()`, `VecDuplicateVecs()`, `VecCreateGhost()`,
424: `VecCreateMPI()`, `VecCreateGhostWithArray()`, `VecPlaceArray()`
426: @*/
427: PetscErrorCode VecCreateMPIKokkosWithArray(MPI_Comm comm, PetscInt bs, PetscInt n, PetscInt N, const PetscScalar darray[], Vec *v)
428: {
429: Vec w;
430: Vec_Kokkos *veckok;
431: Vec_MPI *vecmpi;
432: PetscScalar *harray;
434: PetscFunctionBegin;
435: PetscCheck(n != PETSC_DECIDE, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Must set local size of vector");
436: PetscCall(PetscKokkosInitializeCheck());
437: PetscCall(PetscSplitOwnership(comm, &n, &N));
438: PetscCall(VecCreate(comm, &w));
439: PetscCall(VecSetSizes(w, n, N));
440: PetscCall(VecSetBlockSize(w, bs));
441: PetscCall(PetscLayoutSetUp(w->map));
443: if (std::is_same<DefaultMemorySpace, HostMirrorMemorySpace>::value) {
444: harray = const_cast<PetscScalar *>(darray);
445: } else PetscCall(PetscMalloc1(w->map->n, &harray)); /* If device is not the same as host, allocate the host array ourselves */
447: PetscCall(VecCreate_MPI_Private(w, PETSC_FALSE /*alloc*/, 0 /*nghost*/, harray)); /* Build a sequential vector with provided data */
448: vecmpi = static_cast<Vec_MPI *>(w->data);
450: if (!std::is_same<DefaultMemorySpace, HostMirrorMemorySpace>::value) vecmpi->array_allocated = harray; /* The host array was allocated by petsc */
452: PetscCall(PetscObjectChangeTypeName((PetscObject)w, VECMPIKOKKOS));
453: PetscCall(VecSetOps_MPIKokkos(w));
454: veckok = new Vec_Kokkos(n, harray, const_cast<PetscScalar *>(darray));
455: veckok->v_dual.modify_device(); /* Mark the device is modified */
456: w->spptr = static_cast<void *>(veckok);
457: w->offloadmask = PETSC_OFFLOAD_KOKKOS;
458: *v = w;
459: PetscFunctionReturn(PETSC_SUCCESS);
460: }
462: /*
463: VecCreateMPIKokkosWithArrays_Private - Creates a Kokkos parallel, array-style vector
464: with user-provided arrays on host and device.
466: Collective
468: Input Parameter:
469: + comm - the communicator
470: . bs - the block size
471: . n - the local vector length
472: . N - the global vector length
473: - harray - host memory where the vector elements are to be stored.
474: - darray - device memory where the vector elements are to be stored.
476: Output Parameter:
477: . v - the vector
479: Notes:
480: If there is no device, then harray and darray must be the same.
481: If n is not zero, then harray and darray must be allocated.
482: After the call, the created vector is supposed to be in a synchronized state, i.e.,
483: we suppose harray and darray have the same data.
485: PETSc does NOT free the array when the vector is destroyed via VecDestroy().
486: The user should not free the array until the vector is destroyed.
487: */
488: PetscErrorCode VecCreateMPIKokkosWithArrays_Private(MPI_Comm comm, PetscInt bs, PetscInt n, PetscInt N, const PetscScalar harray[], const PetscScalar darray[], Vec *v)
489: {
490: Vec w;
492: PetscFunctionBegin;
493: PetscCall(PetscKokkosInitializeCheck());
494: if (n) {
495: PetscAssertPointer(harray, 5);
496: PetscCheck(darray, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "darray cannot be NULL");
497: }
498: if (std::is_same<DefaultMemorySpace, HostMirrorMemorySpace>::value) PetscCheck(harray == darray, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "harray and darray must be the same");
499: PetscCall(VecCreateMPIWithArray(comm, bs, n, N, harray, &w));
500: PetscCall(PetscObjectChangeTypeName((PetscObject)w, VECMPIKOKKOS)); /* Change it to Kokkos */
501: PetscCall(VecSetOps_MPIKokkos(w));
502: PetscCallCXX(w->spptr = new Vec_Kokkos(n, const_cast<PetscScalar *>(harray), const_cast<PetscScalar *>(darray)));
503: w->offloadmask = PETSC_OFFLOAD_KOKKOS;
504: *v = w;
505: PetscFunctionReturn(PETSC_SUCCESS);
506: }
508: /*MC
509: VECKOKKOS - VECKOKKOS = "kokkos" - The basic vector, modified to use Kokkos
511: Options Database Keys:
512: . -vec_type kokkos - sets the vector type to VECKOKKOS during a call to VecSetFromOptions()
514: Level: beginner
516: .seealso: `VecCreate()`, `VecSetType()`, `VecSetFromOptions()`, `VecCreateMPIKokkosWithArray()`, `VECMPI`, `VecType`, `VecCreateMPI()`
517: M*/
518: PetscErrorCode VecCreate_Kokkos(Vec v)
519: {
520: PetscMPIInt size;
522: PetscFunctionBegin;
523: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)v), &size));
524: if (size == 1) PetscCall(VecSetType(v, VECSEQKOKKOS));
525: else PetscCall(VecSetType(v, VECMPIKOKKOS));
526: PetscFunctionReturn(PETSC_SUCCESS);
527: }