1 2 /* 3 Defines projective product routines where A is a MPIAIJ matrix 4 C = P^T * A * P 5 */ 6 7 #include <../src/mat/impls/aij/seq/aij.h> /*I "petscmat.h" I*/ 8 #include <../src/mat/utils/freespace.h> 9 #include <../src/mat/impls/aij/mpi/mpiaij.h> 10 #include <petscbt.h> 11 #include <petsctime.h> 12 13 /* #define PTAP_PROFILE */ 14 15 extern PetscErrorCode MatDestroy_MPIAIJ(Mat); 16 #undef __FUNCT__ 17 #define __FUNCT__ "MatDestroy_MPIAIJ_PtAP" 18 PetscErrorCode MatDestroy_MPIAIJ_PtAP(Mat A) 19 { 20 PetscErrorCode ierr; 21 Mat_MPIAIJ *a =(Mat_MPIAIJ*)A->data; 22 Mat_PtAPMPI *ptap=a->ptap; 23 24 PetscFunctionBegin; 25 if (ptap) { 26 Mat_Merge_SeqsToMPI *merge=ptap->merge; 27 ierr = PetscFree2(ptap->startsj_s,ptap->startsj_r);CHKERRQ(ierr); 28 ierr = PetscFree(ptap->bufa);CHKERRQ(ierr); 29 ierr = MatDestroy(&ptap->P_loc);CHKERRQ(ierr); 30 ierr = MatDestroy(&ptap->P_oth);CHKERRQ(ierr); 31 ierr = MatDestroy(&ptap->A_loc);CHKERRQ(ierr); /* used by MatTransposeMatMult() */ 32 if (ptap->api) {ierr = PetscFree(ptap->api);CHKERRQ(ierr);} 33 if (ptap->apj) {ierr = PetscFree(ptap->apj);CHKERRQ(ierr);} 34 if (ptap->apa) {ierr = PetscFree(ptap->apa);CHKERRQ(ierr);} 35 if (merge) { 36 ierr = PetscFree(merge->id_r);CHKERRQ(ierr); 37 ierr = PetscFree(merge->len_s);CHKERRQ(ierr); 38 ierr = PetscFree(merge->len_r);CHKERRQ(ierr); 39 ierr = PetscFree(merge->bi);CHKERRQ(ierr); 40 ierr = PetscFree(merge->bj);CHKERRQ(ierr); 41 ierr = PetscFree(merge->buf_ri[0]);CHKERRQ(ierr); 42 ierr = PetscFree(merge->buf_ri);CHKERRQ(ierr); 43 ierr = PetscFree(merge->buf_rj[0]);CHKERRQ(ierr); 44 ierr = PetscFree(merge->buf_rj);CHKERRQ(ierr); 45 ierr = PetscFree(merge->coi);CHKERRQ(ierr); 46 ierr = PetscFree(merge->coj);CHKERRQ(ierr); 47 ierr = PetscFree(merge->owners_co);CHKERRQ(ierr); 48 ierr = PetscLayoutDestroy(&merge->rowmap);CHKERRQ(ierr); 49 ierr = merge->destroy(A);CHKERRQ(ierr); 50 ierr = PetscFree(ptap->merge);CHKERRQ(ierr); 51 } 52 ierr = PetscFree(ptap);CHKERRQ(ierr); 53 } 54 PetscFunctionReturn(0); 55 } 56 57 #undef __FUNCT__ 58 #define __FUNCT__ "MatDuplicate_MPIAIJ_MatPtAP" 59 PetscErrorCode MatDuplicate_MPIAIJ_MatPtAP(Mat A, MatDuplicateOption op, Mat *M) 60 { 61 PetscErrorCode ierr; 62 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 63 Mat_PtAPMPI *ptap = a->ptap; 64 Mat_Merge_SeqsToMPI *merge = ptap->merge; 65 66 PetscFunctionBegin; 67 ierr = (*merge->duplicate)(A,op,M);CHKERRQ(ierr); 68 69 (*M)->ops->destroy = merge->destroy; 70 (*M)->ops->duplicate = merge->duplicate; 71 PetscFunctionReturn(0); 72 } 73 74 #undef __FUNCT__ 75 #define __FUNCT__ "MatPtAP_MPIAIJ_MPIAIJ" 76 PetscErrorCode MatPtAP_MPIAIJ_MPIAIJ(Mat A,Mat P,MatReuse scall,PetscReal fill,Mat *C) 77 { 78 PetscErrorCode ierr; 79 80 PetscFunctionBegin; 81 if (scall == MAT_INITIAL_MATRIX) { 82 ierr = PetscLogEventBegin(MAT_PtAPSymbolic,A,P,0,0);CHKERRQ(ierr); 83 ierr = MatPtAPSymbolic_MPIAIJ_MPIAIJ(A,P,fill,C);CHKERRQ(ierr); 84 ierr = PetscLogEventEnd(MAT_PtAPSymbolic,A,P,0,0);CHKERRQ(ierr); 85 } 86 ierr = PetscLogEventBegin(MAT_PtAPNumeric,A,P,0,0);CHKERRQ(ierr); 87 ierr = MatPtAPNumeric_MPIAIJ_MPIAIJ(A,P,*C);CHKERRQ(ierr); 88 ierr = PetscLogEventEnd(MAT_PtAPNumeric,A,P,0,0);CHKERRQ(ierr); 89 PetscFunctionReturn(0); 90 } 91 92 #undef __FUNCT__ 93 #define __FUNCT__ "MatPtAPSymbolic_MPIAIJ_MPIAIJ" 94 PetscErrorCode MatPtAPSymbolic_MPIAIJ_MPIAIJ(Mat A,Mat P,PetscReal fill,Mat *C) 95 { 96 PetscErrorCode ierr; 97 Mat Cmpi; 98 Mat_PtAPMPI *ptap; 99 PetscFreeSpaceList free_space=NULL,current_space=NULL; 100 Mat_MPIAIJ *a =(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c; 101 Mat_SeqAIJ *ad =(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 102 Mat_SeqAIJ *p_loc,*p_oth; 103 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*pdti,*pdtj,*poti,*potj,*ptJ; 104 PetscInt *adi=ad->i,*aj,*aoi=ao->i,nnz; 105 PetscInt *lnk,*owners_co,*coi,*coj,i,k,pnz,row; 106 PetscInt am=A->rmap->n,pN=P->cmap->N,pm=P->rmap->n,pn=P->cmap->n; 107 PetscBT lnkbt; 108 MPI_Comm comm; 109 PetscMPIInt size,rank,tagi,tagj,*len_si,*len_s,*len_ri,icompleted=0; 110 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 111 PetscInt len,proc,*dnz,*onz,*owners; 112 PetscInt nzi,*pti,*ptj; 113 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextci; 114 MPI_Request *swaits,*rwaits; 115 MPI_Status *sstatus,rstatus; 116 Mat_Merge_SeqsToMPI *merge; 117 PetscInt *api,*apj,*Jptr,apnz,*prmap=p->garray,pon,nspacedouble=0,j,ap_rmax=0; 118 PetscReal afill=1.0,afill_tmp; 119 PetscInt rmax; 120 121 PetscFunctionBegin; 122 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 123 124 /* check if matrix local sizes are compatible */ 125 if (A->rmap->rstart != P->rmap->rstart || A->rmap->rend != P->rmap->rend) { 126 SETERRQ4(comm,PETSC_ERR_ARG_SIZ,"Matrix local dimensions are incompatible, Arow (%D, %D) != Prow (%D,%D)",A->rmap->rstart,A->rmap->rend,P->rmap->rstart,P->rmap->rend); 127 } 128 if (A->cmap->rstart != P->rmap->rstart || A->cmap->rend != P->rmap->rend) { 129 SETERRQ4(comm,PETSC_ERR_ARG_SIZ,"Matrix local dimensions are incompatible, Acol (%D, %D) != Prow (%D,%D)",A->cmap->rstart,A->cmap->rend,P->rmap->rstart,P->rmap->rend); 130 } 131 132 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 133 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 134 135 /* create struct Mat_PtAPMPI and attached it to C later */ 136 ierr = PetscNew(&ptap);CHKERRQ(ierr); 137 ierr = PetscNew(&merge);CHKERRQ(ierr); 138 ptap->merge = merge; 139 ptap->reuse = MAT_INITIAL_MATRIX; 140 141 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 142 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 143 144 /* get P_loc by taking all local rows of P */ 145 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 146 147 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 148 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 149 pi_loc = p_loc->i; pj_loc = p_loc->j; 150 pi_oth = p_oth->i; pj_oth = p_oth->j; 151 152 /* (1) compute symbolic AP = A_loc*P = A_diag*P_loc + A_off*P_oth */ 153 /*-------------------------------------------------------------------*/ 154 ierr = PetscMalloc1(am+1,&api);CHKERRQ(ierr); 155 api[0] = 0; 156 157 /* create and initialize a linked list */ 158 ierr = PetscLLCondensedCreate(pN,pN,&lnk,&lnkbt);CHKERRQ(ierr); 159 160 /* Initial FreeSpace size is fill*(nnz(A) + nnz(P)) -OOM for ex56, np=8k on Intrepid! */ 161 ierr = PetscFreeSpaceGet((PetscInt)(fill*(adi[am]+aoi[am]+pi_loc[pm])),&free_space);CHKERRQ(ierr); 162 163 current_space = free_space; 164 165 for (i=0; i<am; i++) { 166 /* diagonal portion of A */ 167 nzi = adi[i+1] - adi[i]; 168 aj = ad->j + adi[i]; 169 for (j=0; j<nzi; j++) { 170 row = aj[j]; 171 pnz = pi_loc[row+1] - pi_loc[row]; 172 Jptr = pj_loc + pi_loc[row]; 173 /* add non-zero cols of P into the sorted linked list lnk */ 174 ierr = PetscLLCondensedAddSorted(pnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 175 } 176 /* off-diagonal portion of A */ 177 nzi = aoi[i+1] - aoi[i]; 178 aj = ao->j + aoi[i]; 179 for (j=0; j<nzi; j++) { 180 row = aj[j]; 181 pnz = pi_oth[row+1] - pi_oth[row]; 182 Jptr = pj_oth + pi_oth[row]; 183 ierr = PetscLLCondensedAddSorted(pnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 184 } 185 apnz = lnk[0]; 186 api[i+1] = api[i] + apnz; 187 if (ap_rmax < apnz) ap_rmax = apnz; 188 189 /* if free space is not available, double the total space in the list */ 190 if (current_space->local_remaining<apnz) { 191 ierr = PetscFreeSpaceGet(apnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 192 nspacedouble++; 193 } 194 195 /* Copy data into free space, then initialize lnk */ 196 ierr = PetscLLCondensedClean(pN,apnz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 197 198 current_space->array += apnz; 199 current_space->local_used += apnz; 200 current_space->local_remaining -= apnz; 201 } 202 203 /* Allocate space for apj, initialize apj, and */ 204 /* destroy list of free space and other temporary array(s) */ 205 ierr = PetscMalloc1(api[am]+1,&apj);CHKERRQ(ierr); 206 ierr = PetscFreeSpaceContiguous(&free_space,apj);CHKERRQ(ierr); 207 afill_tmp = (PetscReal)api[am]/(adi[am]+aoi[am]+pi_loc[pm]+1); 208 if (afill_tmp > afill) afill = afill_tmp; 209 210 /* (2) determine symbolic Co=(p->B)^T*AP - send to others */ 211 /*--------------------------------------------------------*/ 212 ierr = MatGetSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 213 214 /* then, compute symbolic Co = (p->B)^T*AP */ 215 pon = (p->B)->cmap->n; /* total num of rows to be sent to other processors 216 >= (num of nonzero rows of C_seq) - pn */ 217 ierr = PetscMalloc1(pon+1,&coi);CHKERRQ(ierr); 218 coi[0] = 0; 219 220 /* set initial free space to be fill*(nnz(p->B) + nnz(AP)) */ 221 nnz = fill*(poti[pon] + api[am]); 222 ierr = PetscFreeSpaceGet(nnz,&free_space);CHKERRQ(ierr); 223 current_space = free_space; 224 225 for (i=0; i<pon; i++) { 226 pnz = poti[i+1] - poti[i]; 227 ptJ = potj + poti[i]; 228 for (j=0; j<pnz; j++) { 229 row = ptJ[j]; /* row of AP == col of Pot */ 230 apnz = api[row+1] - api[row]; 231 Jptr = apj + api[row]; 232 /* add non-zero cols of AP into the sorted linked list lnk */ 233 ierr = PetscLLCondensedAddSorted(apnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 234 } 235 nnz = lnk[0]; 236 237 /* If free space is not available, double the total space in the list */ 238 if (current_space->local_remaining<nnz) { 239 ierr = PetscFreeSpaceGet(nnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 240 nspacedouble++; 241 } 242 243 /* Copy data into free space, and zero out denserows */ 244 ierr = PetscLLCondensedClean(pN,nnz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 245 246 current_space->array += nnz; 247 current_space->local_used += nnz; 248 current_space->local_remaining -= nnz; 249 250 coi[i+1] = coi[i] + nnz; 251 } 252 printf("[%d] coi[%d] = %d\n",rank,pon,coi[pon]); //coi[pon] can be 0!!! 253 254 ierr = PetscMalloc1(coi[pon],&coj);CHKERRQ(ierr); 255 ierr = PetscFreeSpaceContiguous(&free_space,coj);CHKERRQ(ierr); 256 afill_tmp = (PetscReal)coi[pon]/(poti[pon] + api[am]+1); 257 if (afill_tmp > afill) afill = afill_tmp; 258 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 259 260 /* (3) send j-array (coj) of Co to other processors */ 261 /*--------------------------------------------------*/ 262 ierr = PetscCalloc1(size,&merge->len_s);CHKERRQ(ierr); 263 len_s = merge->len_s; 264 merge->nsend = 0; 265 266 267 /* determine row ownership */ 268 ierr = PetscLayoutCreate(comm,&merge->rowmap);CHKERRQ(ierr); 269 merge->rowmap->n = pn; 270 merge->rowmap->bs = 1; 271 272 ierr = PetscLayoutSetUp(merge->rowmap);CHKERRQ(ierr); 273 owners = merge->rowmap->range; 274 275 /* determine the number of messages to send, their lengths */ 276 ierr = PetscMalloc2(size,&len_si,size,&sstatus);CHKERRQ(ierr); 277 ierr = PetscMemzero(len_si,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 278 ierr = PetscMalloc1(size+2,&owners_co);CHKERRQ(ierr); 279 280 proc = 0; 281 for (i=0; i<pon; i++) { 282 while (prmap[i] >= owners[proc+1]) proc++; 283 len_si[proc]++; /* num of rows in Co(=Pt*AP) to be sent to [proc] -- could be empty row!!! */ 284 len_s[proc] += coi[i+1] - coi[i]; /* num of nonzeros in Co to be sent to [proc] */ 285 } 286 287 len = 0; /* max length of buf_si[] */ 288 owners_co[0] = 0; 289 for (proc=0; proc<size; proc++) { 290 owners_co[proc+1] = owners_co[proc] + len_si[proc]; 291 if (len_s[proc]) { 292 merge->nsend++; 293 len_si[proc] = 2*(len_si[proc] + 1); 294 len += len_si[proc]; 295 } 296 } 297 298 /* determine the number and length of messages to receive for coi and coj */ 299 ierr = PetscGatherNumberOfMessages(comm,NULL,len_s,&merge->nrecv);CHKERRQ(ierr); 300 printf("[%d] nsend %d, nrecv %d\n",rank,merge->nsend,merge->nrecv); 301 ierr = PetscGatherMessageLengths2(comm,merge->nsend,merge->nrecv,len_s,len_si,&merge->id_r,&merge->len_r,&len_ri);CHKERRQ(ierr); 302 303 /* post the Irecv and Isend of coj */ 304 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 305 ierr = PetscPostIrecvInt(comm,tagj,merge->nrecv,merge->id_r,merge->len_r,&buf_rj,&rwaits);CHKERRQ(ierr); 306 ierr = PetscMalloc1(merge->nsend+1,&swaits);CHKERRQ(ierr); 307 for (proc=0, k=0; proc<size; proc++) { 308 if (!len_s[proc]) continue; 309 i = owners_co[proc]; 310 ierr = MPI_Isend(coj+coi[i],len_s[proc],MPIU_INT,proc,tagj,comm,swaits+k);CHKERRQ(ierr); 311 k++; 312 } 313 314 /* receives and sends of coj are complete */ 315 for (i=0; i<merge->nrecv; i++) { 316 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 317 } 318 ierr = PetscFree(rwaits);CHKERRQ(ierr); 319 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 320 321 /* (4) send and recv coi */ 322 /*-----------------------*/ 323 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 324 ierr = PetscPostIrecvInt(comm,tagi,merge->nrecv,merge->id_r,len_ri,&buf_ri,&rwaits);CHKERRQ(ierr); 325 ierr = PetscMalloc1(len+1,&buf_s);CHKERRQ(ierr); 326 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 327 for (proc=0,k=0; proc<size; proc++) { 328 if (!len_s[proc]) continue; 329 /* form outgoing message for i-structure: 330 buf_si[0]: nrows to be sent 331 [1:nrows]: row index (global) 332 [nrows+1:2*nrows+1]: i-structure index 333 */ 334 /*-------------------------------------------*/ 335 nrows = len_si[proc]/2 - 1; 336 buf_si_i = buf_si + nrows+1; 337 buf_si[0] = nrows; 338 buf_si_i[0] = 0; 339 nrows = 0; 340 for (i=owners_co[proc]; i<owners_co[proc+1]; i++) { 341 nzi = coi[i+1] - coi[i]; 342 343 buf_si_i[nrows+1] = buf_si_i[nrows] + nzi; /* i-structure */ 344 buf_si[nrows+1] = prmap[i] -owners[proc]; /* local row index */ 345 nrows++; 346 } 347 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,swaits+k);CHKERRQ(ierr); 348 k++; 349 buf_si += len_si[proc]; 350 } 351 i = merge->nrecv; 352 while (i--) { 353 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 354 } 355 ierr = PetscFree(rwaits);CHKERRQ(ierr); 356 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 357 358 ierr = PetscFree2(len_si,sstatus);CHKERRQ(ierr); 359 ierr = PetscFree(len_ri);CHKERRQ(ierr); 360 ierr = PetscFree(swaits);CHKERRQ(ierr); 361 ierr = PetscFree(buf_s);CHKERRQ(ierr); 362 363 /* (5) compute the local portion of C (mpi mat) */ 364 /*----------------------------------------------*/ 365 ierr = MatGetSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 366 367 /* allocate pti array and free space for accumulating nonzero column info */ 368 ierr = PetscMalloc1(pn+1,&pti);CHKERRQ(ierr); 369 pti[0] = 0; 370 371 /* set initial free space to be fill*(nnz(P) + nnz(AP)) */ 372 nnz = fill*(pi_loc[pm] + api[am]); 373 ierr = PetscFreeSpaceGet(nnz,&free_space);CHKERRQ(ierr); 374 current_space = free_space; 375 376 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextci);CHKERRQ(ierr); 377 for (k=0; k<merge->nrecv; k++) { 378 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 379 nrows = *buf_ri_k[k]; 380 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 381 nextci[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 382 } 383 ierr = MatPreallocateInitialize(comm,pn,pn,dnz,onz);CHKERRQ(ierr); 384 rmax = 0; 385 for (i=0; i<pn; i++) { 386 /* add pdt[i,:]*AP into lnk */ 387 pnz = pdti[i+1] - pdti[i]; 388 ptJ = pdtj + pdti[i]; 389 for (j=0; j<pnz; j++) { 390 row = ptJ[j]; /* row of AP == col of Pt */ 391 apnz = api[row+1] - api[row]; 392 Jptr = apj + api[row]; 393 /* add non-zero cols of AP into the sorted linked list lnk */ 394 ierr = PetscLLCondensedAddSorted(apnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 395 } 396 397 /* add received col data into lnk */ 398 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 399 if (i == *nextrow[k]) { /* i-th row */ 400 nzi = *(nextci[k]+1) - *nextci[k]; 401 Jptr = buf_rj[k] + *nextci[k]; 402 ierr = PetscLLCondensedAddSorted(nzi,Jptr,lnk,lnkbt);CHKERRQ(ierr); 403 nextrow[k]++; nextci[k]++; 404 } 405 } 406 nnz = lnk[0]; 407 408 /* if free space is not available, make more free space */ 409 if (current_space->local_remaining<nnz) { 410 ierr = PetscFreeSpaceGet(nnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 411 nspacedouble++; 412 } 413 /* copy data into free space, then initialize lnk */ 414 ierr = PetscLLCondensedClean(pN,nnz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 415 ierr = MatPreallocateSet(i+owners[rank],nnz,current_space->array,dnz,onz);CHKERRQ(ierr); 416 417 current_space->array += nnz; 418 current_space->local_used += nnz; 419 current_space->local_remaining -= nnz; 420 421 pti[i+1] = pti[i] + nnz; 422 if (nnz > rmax) rmax = nnz; 423 } 424 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 425 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 426 427 ierr = PetscMalloc1(pti[pn]+1,&ptj);CHKERRQ(ierr); 428 ierr = PetscFreeSpaceContiguous(&free_space,ptj);CHKERRQ(ierr); 429 afill_tmp = (PetscReal)pti[pn]/(pi_loc[pm] + api[am]+1); 430 if (afill_tmp > afill) afill = afill_tmp; 431 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 432 433 /* (6) create symbolic parallel matrix Cmpi */ 434 /*------------------------------------------*/ 435 ierr = MatCreate(comm,&Cmpi);CHKERRQ(ierr); 436 ierr = MatSetSizes(Cmpi,pn,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 437 ierr = MatSetBlockSizes(Cmpi,PetscAbs(P->cmap->bs),PetscAbs(P->cmap->bs));CHKERRQ(ierr); 438 ierr = MatSetType(Cmpi,MATMPIAIJ);CHKERRQ(ierr); 439 ierr = MatMPIAIJSetPreallocation(Cmpi,0,dnz,0,onz);CHKERRQ(ierr); 440 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 441 442 merge->bi = pti; /* Cseq->i */ 443 merge->bj = ptj; /* Cseq->j */ 444 merge->coi = coi; /* Co->i */ 445 merge->coj = coj; /* Co->j */ 446 merge->buf_ri = buf_ri; 447 merge->buf_rj = buf_rj; 448 merge->owners_co = owners_co; 449 merge->destroy = Cmpi->ops->destroy; 450 merge->duplicate = Cmpi->ops->duplicate; 451 452 /* Cmpi is not ready for use - assembly will be done by MatPtAPNumeric() */ 453 Cmpi->assembled = PETSC_FALSE; 454 Cmpi->ops->destroy = MatDestroy_MPIAIJ_PtAP; 455 Cmpi->ops->duplicate = MatDuplicate_MPIAIJ_MatPtAP; 456 457 /* attach the supporting struct to Cmpi for reuse */ 458 c = (Mat_MPIAIJ*)Cmpi->data; 459 c->ptap = ptap; 460 ptap->api = api; 461 ptap->apj = apj; 462 ptap->rmax = ap_rmax; 463 *C = Cmpi; 464 465 /* flag 'scalable' determines which implementations to be used: 466 0: do dense axpy in MatPtAPNumeric() - fast, but requires storage of a nonscalable dense array apa; 467 1: do sparse axpy in MatPtAPNumeric() - might slow, uses a sparse array apa */ 468 /* set default scalable */ 469 ptap->scalable = PETSC_TRUE; 470 471 ierr = PetscOptionsGetBool(((PetscObject)Cmpi)->prefix,"-matptap_scalable",&ptap->scalable,NULL);CHKERRQ(ierr); 472 if (!ptap->scalable) { /* Do dense axpy */ 473 ierr = PetscCalloc1(pN,&ptap->apa);CHKERRQ(ierr); 474 } else { 475 ierr = PetscCalloc1(ap_rmax+1,&ptap->apa);CHKERRQ(ierr); 476 } 477 478 #if defined(PETSC_USE_INFO) 479 if (pti[pn] != 0) { 480 ierr = PetscInfo3(Cmpi,"Reallocs %D; Fill ratio: given %g needed %g.\n",nspacedouble,(double)fill,(double)afill);CHKERRQ(ierr); 481 ierr = PetscInfo1(Cmpi,"Use MatPtAP(A,P,MatReuse,%g,&C) for best performance.\n",(double)afill);CHKERRQ(ierr); 482 } else { 483 ierr = PetscInfo(Cmpi,"Empty matrix product\n");CHKERRQ(ierr); 484 } 485 #endif 486 PetscFunctionReturn(0); 487 } 488 489 #undef __FUNCT__ 490 #define __FUNCT__ "MatPtAPNumeric_MPIAIJ_MPIAIJ" 491 PetscErrorCode MatPtAPNumeric_MPIAIJ_MPIAIJ(Mat A,Mat P,Mat C) 492 { 493 PetscErrorCode ierr; 494 Mat_MPIAIJ *a =(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c=(Mat_MPIAIJ*)C->data; 495 Mat_SeqAIJ *ad=(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 496 Mat_SeqAIJ *pd=(Mat_SeqAIJ*)(p->A)->data,*po=(Mat_SeqAIJ*)(p->B)->data; 497 Mat_SeqAIJ *p_loc,*p_oth; 498 Mat_PtAPMPI *ptap; 499 Mat_Merge_SeqsToMPI *merge; 500 PetscInt *adi=ad->i,*aoi=ao->i,*adj,*aoj,*apJ,nextp; 501 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*pJ,*pj; 502 PetscInt i,j,k,anz,pnz,apnz,nextap,row,*cj; 503 MatScalar *ada,*aoa,*apa,*pa,*ca,*pa_loc,*pa_oth,valtmp; 504 PetscInt am =A->rmap->n,cm=C->rmap->n,pon=(p->B)->cmap->n; 505 MPI_Comm comm; 506 PetscMPIInt size,rank,taga,*len_s; 507 PetscInt *owners,proc,nrows,**buf_ri_k,**nextrow,**nextci; 508 PetscInt **buf_ri,**buf_rj; 509 PetscInt cnz=0,*bj_i,*bi,*bj,bnz,nextcj; /* bi,bj,ba: local array of C(mpi mat) */ 510 MPI_Request *s_waits,*r_waits; 511 MPI_Status *status; 512 MatScalar **abuf_r,*ba_i,*pA,*coa,*ba; 513 PetscInt *api,*apj,*coi,*coj; 514 PetscInt *poJ=po->j,*pdJ=pd->j,pcstart=P->cmap->rstart,pcend=P->cmap->rend; 515 PetscBool scalable; 516 #if defined(PTAP_PROFILE) 517 PetscLogDouble t0,t1,t2,t3,t4,et2_AP=0.0,et2_PtAP=0.0,t2_0,t2_1,t2_2; 518 #endif 519 520 PetscFunctionBegin; 521 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 522 #if defined(PTAP_PROFILE) 523 ierr = PetscTime(&t0);CHKERRQ(ierr); 524 #endif 525 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 526 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 527 528 ptap = c->ptap; 529 if (!ptap) SETERRQ(PetscObjectComm((PetscObject)C),PETSC_ERR_ARG_INCOMP,"MatPtAP() has not been called to create matrix C yet, cannot use MAT_REUSE_MATRIX"); 530 merge = ptap->merge; 531 apa = ptap->apa; 532 scalable = ptap->scalable; 533 534 /* 1) get P_oth = ptap->P_oth and P_loc = ptap->P_loc */ 535 /*-----------------------------------------------------*/ 536 if (ptap->reuse == MAT_INITIAL_MATRIX) { 537 /* P_oth and P_loc are obtained in MatPtASymbolic(), skip calling MatGetBrowsOfAoCols() and MatMPIAIJGetLocalMat() */ 538 ptap->reuse = MAT_REUSE_MATRIX; 539 } else { /* update numerical values of P_oth and P_loc */ 540 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_REUSE_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 541 ierr = MatMPIAIJGetLocalMat(P,MAT_REUSE_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 542 } 543 #if defined(PTAP_PROFILE) 544 ierr = PetscTime(&t1);CHKERRQ(ierr); 545 #endif 546 547 /* 2) compute numeric C_seq = P_loc^T*A_loc*P - dominating part */ 548 /*--------------------------------------------------------------*/ 549 /* get data from symbolic products */ 550 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 551 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 552 pi_loc=p_loc->i; pj_loc=p_loc->j; pJ=pj_loc; pa_loc=p_loc->a; 553 pi_oth=p_oth->i; pj_oth=p_oth->j; pa_oth=p_oth->a; 554 555 coi = merge->coi; coj = merge->coj; 556 ierr = PetscCalloc1(coi[pon]+1,&coa);CHKERRQ(ierr); 557 558 bi = merge->bi; bj = merge->bj; 559 owners = merge->rowmap->range; 560 ierr = PetscCalloc1(bi[cm]+1,&ba);CHKERRQ(ierr); /* ba: Cseq->a */ 561 562 api = ptap->api; apj = ptap->apj; 563 564 if (!scalable) { /* Do dense axpy on apa (length of pN, stores A[i,:]*P) - nonscalable, but fast */ 565 ierr = PetscInfo(C,"Using non-scalable dense axpy\n");CHKERRQ(ierr); 566 /*-----------------------------------------------------------------------------------------------------*/ 567 for (i=0; i<am; i++) { 568 #if defined(PTAP_PROFILE) 569 ierr = PetscTime(&t2_0);CHKERRQ(ierr); 570 #endif 571 /* 2-a) form i-th sparse row of A_loc*P = Ad*P_loc + Ao*P_oth */ 572 /*------------------------------------------------------------*/ 573 apJ = apj + api[i]; 574 575 /* diagonal portion of A */ 576 anz = adi[i+1] - adi[i]; 577 adj = ad->j + adi[i]; 578 ada = ad->a + adi[i]; 579 for (j=0; j<anz; j++) { 580 row = adj[j]; 581 pnz = pi_loc[row+1] - pi_loc[row]; 582 pj = pj_loc + pi_loc[row]; 583 pa = pa_loc + pi_loc[row]; 584 585 /* perform dense axpy */ 586 valtmp = ada[j]; 587 for (k=0; k<pnz; k++) { 588 apa[pj[k]] += valtmp*pa[k]; 589 } 590 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 591 } 592 593 /* off-diagonal portion of A */ 594 anz = aoi[i+1] - aoi[i]; 595 aoj = ao->j + aoi[i]; 596 aoa = ao->a + aoi[i]; 597 for (j=0; j<anz; j++) { 598 row = aoj[j]; 599 pnz = pi_oth[row+1] - pi_oth[row]; 600 pj = pj_oth + pi_oth[row]; 601 pa = pa_oth + pi_oth[row]; 602 603 /* perform dense axpy */ 604 valtmp = aoa[j]; 605 for (k=0; k<pnz; k++) { 606 apa[pj[k]] += valtmp*pa[k]; 607 } 608 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 609 } 610 #if defined(PTAP_PROFILE) 611 ierr = PetscTime(&t2_1);CHKERRQ(ierr); 612 et2_AP += t2_1 - t2_0; 613 #endif 614 615 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 616 /*--------------------------------------------------------------*/ 617 apnz = api[i+1] - api[i]; 618 /* put the value into Co=(p->B)^T*AP (off-diagonal part, send to others) */ 619 pnz = po->i[i+1] - po->i[i]; 620 poJ = po->j + po->i[i]; 621 pA = po->a + po->i[i]; 622 for (j=0; j<pnz; j++) { 623 row = poJ[j]; 624 cnz = coi[row+1] - coi[row]; 625 cj = coj + coi[row]; 626 ca = coa + coi[row]; 627 /* perform dense axpy */ 628 valtmp = pA[j]; 629 for (k=0; k<cnz; k++) { 630 ca[k] += valtmp*apa[cj[k]]; 631 } 632 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 633 } 634 635 /* put the value into Cd (diagonal part) */ 636 pnz = pd->i[i+1] - pd->i[i]; 637 pdJ = pd->j + pd->i[i]; 638 pA = pd->a + pd->i[i]; 639 for (j=0; j<pnz; j++) { 640 row = pdJ[j]; 641 cnz = bi[row+1] - bi[row]; 642 cj = bj + bi[row]; 643 ca = ba + bi[row]; 644 /* perform dense axpy */ 645 valtmp = pA[j]; 646 for (k=0; k<cnz; k++) { 647 ca[k] += valtmp*apa[cj[k]]; 648 } 649 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 650 } 651 652 /* zero the current row of A*P */ 653 for (k=0; k<apnz; k++) apa[apJ[k]] = 0.0; 654 #if defined(PTAP_PROFILE) 655 ierr = PetscTime(&t2_2);CHKERRQ(ierr); 656 et2_PtAP += t2_2 - t2_1; 657 #endif 658 } 659 } else { /* Do sparse axpy on apa (length of ap_rmax, stores A[i,:]*P) - scalable, but slower */ 660 ierr = PetscInfo(C,"Using scalable sparse axpy\n");CHKERRQ(ierr); 661 /*-----------------------------------------------------------------------------------------*/ 662 pA=pa_loc; 663 for (i=0; i<am; i++) { 664 #if defined(PTAP_PROFILE) 665 ierr = PetscTime(&t2_0);CHKERRQ(ierr); 666 #endif 667 /* form i-th sparse row of A*P */ 668 apnz = api[i+1] - api[i]; 669 apJ = apj + api[i]; 670 /* diagonal portion of A */ 671 anz = adi[i+1] - adi[i]; 672 adj = ad->j + adi[i]; 673 ada = ad->a + adi[i]; 674 for (j=0; j<anz; j++) { 675 row = adj[j]; 676 pnz = pi_loc[row+1] - pi_loc[row]; 677 pj = pj_loc + pi_loc[row]; 678 pa = pa_loc + pi_loc[row]; 679 valtmp = ada[j]; 680 nextp = 0; 681 for (k=0; nextp<pnz; k++) { 682 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 683 apa[k] += valtmp*pa[nextp++]; 684 } 685 } 686 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 687 } 688 /* off-diagonal portion of A */ 689 anz = aoi[i+1] - aoi[i]; 690 aoj = ao->j + aoi[i]; 691 aoa = ao->a + aoi[i]; 692 for (j=0; j<anz; j++) { 693 row = aoj[j]; 694 pnz = pi_oth[row+1] - pi_oth[row]; 695 pj = pj_oth + pi_oth[row]; 696 pa = pa_oth + pi_oth[row]; 697 valtmp = aoa[j]; 698 nextp = 0; 699 for (k=0; nextp<pnz; k++) { 700 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 701 apa[k] += valtmp*pa[nextp++]; 702 } 703 } 704 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 705 } 706 #if defined(PTAP_PROFILE) 707 ierr = PetscTime(&t2_1);CHKERRQ(ierr); 708 et2_AP += t2_1 - t2_0; 709 #endif 710 711 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 712 /*--------------------------------------------------------------*/ 713 pnz = pi_loc[i+1] - pi_loc[i]; 714 pJ = pj_loc + pi_loc[i]; 715 for (j=0; j<pnz; j++) { 716 nextap = 0; 717 row = pJ[j]; /* global index */ 718 if (row < pcstart || row >=pcend) { /* put the value into Co */ 719 row = *poJ; 720 cj = coj + coi[row]; 721 ca = coa + coi[row]; poJ++; 722 } else { /* put the value into Cd */ 723 row = *pdJ; 724 cj = bj + bi[row]; 725 ca = ba + bi[row]; pdJ++; 726 } 727 valtmp = pA[j]; 728 for (k=0; nextap<apnz; k++) { 729 if (cj[k]==apJ[nextap]) ca[k] += valtmp*apa[nextap++]; 730 } 731 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 732 } 733 pA += pnz; 734 /* zero the current row info for A*P */ 735 ierr = PetscMemzero(apa,apnz*sizeof(MatScalar));CHKERRQ(ierr); 736 #if defined(PTAP_PROFILE) 737 ierr = PetscTime(&t2_2);CHKERRQ(ierr); 738 et2_PtAP += t2_2 - t2_1; 739 #endif 740 } 741 } 742 #if defined(PTAP_PROFILE) 743 ierr = PetscTime(&t2);CHKERRQ(ierr); 744 #endif 745 746 /* 3) send and recv matrix values coa */ 747 /*------------------------------------*/ 748 buf_ri = merge->buf_ri; 749 buf_rj = merge->buf_rj; 750 len_s = merge->len_s; 751 ierr = PetscCommGetNewTag(comm,&taga);CHKERRQ(ierr); 752 ierr = PetscPostIrecvScalar(comm,taga,merge->nrecv,merge->id_r,merge->len_r,&abuf_r,&r_waits);CHKERRQ(ierr); 753 754 ierr = PetscMalloc2(merge->nsend+1,&s_waits,size,&status);CHKERRQ(ierr); 755 for (proc=0,k=0; proc<size; proc++) { 756 if (!len_s[proc]) continue; 757 i = merge->owners_co[proc]; 758 ierr = MPI_Isend(coa+coi[i],len_s[proc],MPIU_MATSCALAR,proc,taga,comm,s_waits+k);CHKERRQ(ierr); 759 k++; 760 } 761 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,r_waits,status);CHKERRQ(ierr);} 762 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,s_waits,status);CHKERRQ(ierr);} 763 764 ierr = PetscFree2(s_waits,status);CHKERRQ(ierr); 765 ierr = PetscFree(r_waits);CHKERRQ(ierr); 766 ierr = PetscFree(coa);CHKERRQ(ierr); 767 #if defined(PTAP_PROFILE) 768 ierr = PetscTime(&t3);CHKERRQ(ierr); 769 #endif 770 771 /* 4) insert local Cseq and received values into Cmpi */ 772 /*------------------------------------------------------*/ 773 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextci);CHKERRQ(ierr); 774 for (k=0; k<merge->nrecv; k++) { 775 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 776 nrows = *(buf_ri_k[k]); 777 nextrow[k] = buf_ri_k[k]+1; /* next row number of k-th recved i-structure */ 778 nextci[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 779 } 780 781 for (i=0; i<cm; i++) { 782 row = owners[rank] + i; /* global row index of C_seq */ 783 bj_i = bj + bi[i]; /* col indices of the i-th row of C */ 784 ba_i = ba + bi[i]; 785 bnz = bi[i+1] - bi[i]; 786 /* add received vals into ba */ 787 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 788 /* i-th row */ 789 if (i == *nextrow[k]) { 790 cnz = *(nextci[k]+1) - *nextci[k]; 791 cj = buf_rj[k] + *(nextci[k]); 792 ca = abuf_r[k] + *(nextci[k]); 793 nextcj = 0; 794 for (j=0; nextcj<cnz; j++) { 795 if (bj_i[j] == cj[nextcj]) { /* bcol == ccol */ 796 ba_i[j] += ca[nextcj++]; 797 } 798 } 799 nextrow[k]++; nextci[k]++; 800 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 801 } 802 } 803 ierr = MatSetValues(C,1,&row,bnz,bj_i,ba_i,INSERT_VALUES);CHKERRQ(ierr); 804 } 805 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 806 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 807 808 ierr = PetscFree(ba);CHKERRQ(ierr); 809 ierr = PetscFree(abuf_r[0]);CHKERRQ(ierr); 810 ierr = PetscFree(abuf_r);CHKERRQ(ierr); 811 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 812 #if defined(PTAP_PROFILE) 813 ierr = PetscTime(&t4);CHKERRQ(ierr); 814 if (rank==1) PetscPrintf(MPI_COMM_SELF," [%d] PtAPNum %g/P + %g/PtAP( %g + %g ) + %g/comm + %g/Cloc = %g\n\n",rank,t1-t0,t2-t1,et2_AP,et2_PtAP,t3-t2,t4-t3,t4-t0);CHKERRQ(ierr); 815 #endif 816 PetscFunctionReturn(0); 817 } 818