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 PetscErrorCode MatDestroy_MPIAIJ_PtAP(Mat A) 17 { 18 PetscErrorCode ierr; 19 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 20 Mat_PtAPMPI *ptap=a->ptap; 21 22 PetscFunctionBegin; 23 if (ptap) { 24 Mat_Merge_SeqsToMPI *merge=ptap->merge; 25 ierr = PetscFree2(ptap->startsj_s,ptap->startsj_r);CHKERRQ(ierr); 26 ierr = PetscFree(ptap->bufa);CHKERRQ(ierr); 27 ierr = MatDestroy(&ptap->P_loc);CHKERRQ(ierr); 28 ierr = MatDestroy(&ptap->P_oth);CHKERRQ(ierr); 29 ierr = MatDestroy(&ptap->A_loc);CHKERRQ(ierr); /* used by MatTransposeMatMult() */ 30 ierr = MatDestroy(&ptap->Rd);CHKERRQ(ierr); 31 ierr = MatDestroy(&ptap->Ro);CHKERRQ(ierr); 32 if (ptap->AP_loc) { /* used by alg_rap */ 33 Mat_SeqAIJ *ap = (Mat_SeqAIJ*)(ptap->AP_loc)->data; 34 ierr = PetscFree(ap->i);CHKERRQ(ierr); 35 ierr = PetscFree2(ap->j,ap->a);CHKERRQ(ierr); 36 ierr = MatDestroy(&ptap->AP_loc);CHKERRQ(ierr); 37 } else { /* used by alg_ptap */ 38 ierr = PetscFree(ptap->api);CHKERRQ(ierr); 39 ierr = PetscFree(ptap->apj);CHKERRQ(ierr); 40 } 41 ierr = MatDestroy(&ptap->C_loc);CHKERRQ(ierr); 42 ierr = MatDestroy(&ptap->C_oth);CHKERRQ(ierr); 43 if (ptap->apa) {ierr = PetscFree(ptap->apa);CHKERRQ(ierr);} 44 45 if (merge) { /* used by alg_ptap */ 46 ierr = PetscFree(merge->id_r);CHKERRQ(ierr); 47 ierr = PetscFree(merge->len_s);CHKERRQ(ierr); 48 ierr = PetscFree(merge->len_r);CHKERRQ(ierr); 49 ierr = PetscFree(merge->bi);CHKERRQ(ierr); 50 ierr = PetscFree(merge->bj);CHKERRQ(ierr); 51 ierr = PetscFree(merge->buf_ri[0]);CHKERRQ(ierr); 52 ierr = PetscFree(merge->buf_ri);CHKERRQ(ierr); 53 ierr = PetscFree(merge->buf_rj[0]);CHKERRQ(ierr); 54 ierr = PetscFree(merge->buf_rj);CHKERRQ(ierr); 55 ierr = PetscFree(merge->coi);CHKERRQ(ierr); 56 ierr = PetscFree(merge->coj);CHKERRQ(ierr); 57 ierr = PetscFree(merge->owners_co);CHKERRQ(ierr); 58 ierr = PetscLayoutDestroy(&merge->rowmap);CHKERRQ(ierr); 59 ierr = PetscFree(ptap->merge);CHKERRQ(ierr); 60 } 61 62 ierr = ptap->destroy(A);CHKERRQ(ierr); 63 ierr = PetscFree(ptap);CHKERRQ(ierr); 64 } 65 PetscFunctionReturn(0); 66 } 67 68 PetscErrorCode MatDuplicate_MPIAIJ_MatPtAP(Mat A, MatDuplicateOption op, Mat *M) 69 { 70 PetscErrorCode ierr; 71 Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data; 72 Mat_PtAPMPI *ptap = a->ptap; 73 74 PetscFunctionBegin; 75 ierr = (*ptap->duplicate)(A,op,M);CHKERRQ(ierr); 76 (*M)->ops->destroy = ptap->destroy; 77 (*M)->ops->duplicate = ptap->duplicate; 78 PetscFunctionReturn(0); 79 } 80 81 PETSC_INTERN PetscErrorCode MatPtAP_MPIAIJ_MPIAIJ(Mat A,Mat P,MatReuse scall,PetscReal fill,Mat *C) 82 { 83 PetscErrorCode ierr; 84 PetscBool rap=PETSC_TRUE; /* do R=P^T locally, then C=R*A*P */ 85 MPI_Comm comm; 86 87 PetscFunctionBegin; 88 /* check if matrix local sizes are compatible */ 89 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 90 if (A->rmap->rstart != P->rmap->rstart || A->rmap->rend != P->rmap->rend) 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); 91 if (A->cmap->rstart != P->rmap->rstart || A->cmap->rend != P->rmap->rend) 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); 92 93 ierr = PetscOptionsGetBool(NULL,NULL,"-matrap",&rap,NULL);CHKERRQ(ierr); 94 if (scall == MAT_INITIAL_MATRIX) { 95 ierr = PetscLogEventBegin(MAT_PtAPSymbolic,A,P,0,0);CHKERRQ(ierr); 96 if (rap) { /* do R=P^T locally, then C=R*A*P */ 97 ierr = MatPtAPSymbolic_MPIAIJ_MPIAIJ(A,P,fill,C);CHKERRQ(ierr); 98 } else { /* do P^T*A*P */ 99 ierr = MatPtAPSymbolic_MPIAIJ_MPIAIJ_ptap(A,P,fill,C);CHKERRQ(ierr); 100 } 101 ierr = PetscLogEventEnd(MAT_PtAPSymbolic,A,P,0,0);CHKERRQ(ierr); 102 } 103 ierr = PetscLogEventBegin(MAT_PtAPNumeric,A,P,0,0);CHKERRQ(ierr); 104 ierr = (*(*C)->ops->ptapnumeric)(A,P,*C);CHKERRQ(ierr); 105 ierr = PetscLogEventEnd(MAT_PtAPNumeric,A,P,0,0);CHKERRQ(ierr); 106 PetscFunctionReturn(0); 107 } 108 109 #if defined(PETSC_HAVE_HYPRE) 110 PETSC_INTERN PetscErrorCode MatPtAPSymbolic_AIJ_AIJ_wHYPRE(Mat,Mat,PetscReal,Mat*); 111 #endif 112 113 PetscErrorCode MatPtAPNumeric_MPIAIJ_MPIAIJ_scalable(Mat A,Mat P,Mat C) 114 { 115 PetscErrorCode ierr; 116 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c=(Mat_MPIAIJ*)C->data; 117 Mat_SeqAIJ *ad=(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 118 Mat_SeqAIJ *ap,*p_loc,*p_oth,*c_seq; 119 Mat_PtAPMPI *ptap = c->ptap; 120 Mat AP_loc,C_loc,C_oth; 121 PetscInt i,rstart,rend,cm,ncols,row,*api,*apj,am = A->rmap->n,apnz; 122 PetscScalar *apa; 123 const PetscInt *cols; 124 const PetscScalar *vals; 125 126 PetscFunctionBegin; 127 ierr = MatZeroEntries(C);CHKERRQ(ierr); 128 129 /* 1) get R = Pd^T,Ro = Po^T */ 130 if (ptap->reuse == MAT_REUSE_MATRIX) { 131 ierr = MatTranspose_SeqAIJ(p->A,MAT_REUSE_MATRIX,&ptap->Rd);CHKERRQ(ierr); 132 ierr = MatTranspose_SeqAIJ(p->B,MAT_REUSE_MATRIX,&ptap->Ro);CHKERRQ(ierr); 133 } 134 135 /* 2) get AP_loc */ 136 AP_loc = ptap->AP_loc; 137 ap = (Mat_SeqAIJ*)AP_loc->data; 138 139 /* 2-1) get P_oth = ptap->P_oth and P_loc = ptap->P_loc */ 140 /*-----------------------------------------------------*/ 141 if (ptap->reuse == MAT_REUSE_MATRIX) { 142 /* P_oth and P_loc are obtained in MatPtASymbolic() when reuse == MAT_INITIAL_MATRIX */ 143 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_REUSE_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 144 ierr = MatMPIAIJGetLocalMat(P,MAT_REUSE_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 145 } 146 147 /* 2-2) compute numeric A_loc*P - dominating part */ 148 /* ---------------------------------------------- */ 149 /* get data from symbolic products */ 150 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 151 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 152 api = ap->i; 153 apj = ap->j; 154 for (i=0; i<am; i++) { 155 /* AP[i,:] = A[i,:]*P = Ad*P_loc Ao*P_oth */ 156 apnz = api[i+1] - api[i]; 157 apa = ap->a + api[i]; 158 ierr = PetscMemzero(apa,sizeof(PetscScalar)*apnz);CHKERRQ(ierr); 159 AProw_scalable(i,ad,ao,p_loc,p_oth,api,apj,apa); 160 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 161 } 162 163 /* 3) C_loc = Rd*AP_loc, C_oth = Ro*AP_loc */ 164 ierr = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(ptap->Rd,AP_loc,ptap->C_loc);CHKERRQ(ierr); 165 ierr = MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(ptap->Ro,AP_loc,ptap->C_oth);CHKERRQ(ierr); 166 167 C_loc = ptap->C_loc; 168 C_oth = ptap->C_oth; 169 170 /* add C_loc and Co to to C */ 171 ierr = MatGetOwnershipRange(C,&rstart,&rend);CHKERRQ(ierr); 172 173 /* C_loc -> C */ 174 cm = C_loc->rmap->N; 175 c_seq = (Mat_SeqAIJ*)C_loc->data; 176 cols = c_seq->j; 177 vals = c_seq->a; 178 for (i=0; i<cm; i++) { 179 ncols = c_seq->i[i+1] - c_seq->i[i]; 180 row = rstart + i; 181 ierr = MatSetValues(C,1,&row,ncols,cols,vals,ADD_VALUES);CHKERRQ(ierr); 182 cols += ncols; vals += ncols; 183 } 184 185 /* Co -> C, off-processor part */ 186 cm = C_oth->rmap->N; 187 c_seq = (Mat_SeqAIJ*)C_oth->data; 188 cols = c_seq->j; 189 vals = c_seq->a; 190 for (i=0; i<cm; i++) { 191 ncols = c_seq->i[i+1] - c_seq->i[i]; 192 row = p->garray[i]; 193 ierr = MatSetValues(C,1,&row,ncols,cols,vals,ADD_VALUES);CHKERRQ(ierr); 194 cols += ncols; vals += ncols; 195 } 196 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 197 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 198 199 ptap->reuse = MAT_REUSE_MATRIX; 200 PetscFunctionReturn(0); 201 } 202 203 PetscErrorCode MatPtAPSymbolic_MPIAIJ_MPIAIJ_scalable(Mat A,Mat P,PetscReal fill,Mat *C) 204 { 205 PetscErrorCode ierr; 206 Mat_PtAPMPI *ptap; 207 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c; 208 MPI_Comm comm; 209 PetscMPIInt size,rank; 210 Mat Cmpi,P_loc,P_oth; 211 PetscFreeSpaceList free_space=NULL,current_space=NULL; 212 PetscInt am=A->rmap->n,pm=P->rmap->n,pN=P->cmap->N,pn=P->cmap->n; 213 PetscInt *lnk,i,k,pnz,row,nsend; 214 PetscMPIInt tagi,tagj,*len_si,*len_s,*len_ri,icompleted=0,nrecv; 215 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 216 PetscInt len,proc,*dnz,*onz,*owners,nzi,nspacedouble; 217 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextci; 218 MPI_Request *swaits,*rwaits; 219 MPI_Status *sstatus,rstatus; 220 PetscLayout rowmap; 221 PetscInt *owners_co,*coi,*coj; /* i and j array of (p->B)^T*A*P - used in the communication */ 222 PetscMPIInt *len_r,*id_r; /* array of length of comm->size, store send/recv matrix values */ 223 PetscInt *api,*apj,*Jptr,apnz,*prmap=p->garray,con,j,Crmax,*aj,*ai,*pi; 224 Mat_SeqAIJ *p_loc,*p_oth,*ad=(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data,*c_loc,*c_oth; 225 PetscScalar *apv; 226 PetscTable ta; 227 #if defined(PETSC_USE_INFO) 228 PetscReal apfill; 229 #endif 230 231 PetscFunctionBegin; 232 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 233 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 234 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 235 236 /* create symbolic parallel matrix Cmpi */ 237 ierr = MatCreate(comm,&Cmpi);CHKERRQ(ierr); 238 ierr = MatSetType(Cmpi,MATMPIAIJ);CHKERRQ(ierr); 239 240 /* create struct Mat_PtAPMPI and attached it to C later */ 241 ierr = PetscNew(&ptap);CHKERRQ(ierr); 242 ptap->reuse = MAT_INITIAL_MATRIX; 243 244 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 245 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&P_oth);CHKERRQ(ierr); 246 /* get P_loc by taking all local rows of P */ 247 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&P_loc);CHKERRQ(ierr); 248 249 ptap->P_loc = P_loc; 250 ptap->P_oth = P_oth; 251 252 /* (0) compute Rd = Pd^T, Ro = Po^T */ 253 /* --------------------------------- */ 254 ierr = MatTranspose_SeqAIJ(p->A,MAT_INITIAL_MATRIX,&ptap->Rd);CHKERRQ(ierr); 255 ierr = MatTranspose_SeqAIJ(p->B,MAT_INITIAL_MATRIX,&ptap->Ro);CHKERRQ(ierr); 256 257 /* (1) compute symbolic AP = A_loc*P = Ad*P_loc + Ao*P_oth (api,apj) */ 258 /* ----------------------------------------------------------------- */ 259 p_loc = (Mat_SeqAIJ*)P_loc->data; 260 p_oth = (Mat_SeqAIJ*)P_oth->data; 261 262 /* create and initialize a linked list */ 263 ierr = PetscTableCreate(pn,pN,&ta);CHKERRQ(ierr); /* for compute AP_loc and Cmpi */ 264 MatRowMergeMax_SeqAIJ(p_loc,P_loc->rmap->N,ta); 265 MatRowMergeMax_SeqAIJ(p_oth,P_oth->rmap->N,ta); 266 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); /* Crmax = nnz(sum of Prows) */ 267 268 ierr = PetscLLCondensedCreate_Scalable(Crmax,&lnk);CHKERRQ(ierr); 269 270 /* Initial FreeSpace size is fill*(nnz(A) + nnz(P)) */ 271 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ad->i[am],PetscIntSumTruncate(ao->i[am],p_loc->i[pm]))),&free_space);CHKERRQ(ierr); 272 current_space = free_space; 273 nspacedouble = 0; 274 275 ierr = PetscMalloc1(am+1,&api);CHKERRQ(ierr); 276 api[0] = 0; 277 for (i=0; i<am; i++) { 278 /* diagonal portion: Ad[i,:]*P */ 279 ai = ad->i; pi = p_loc->i; 280 nzi = ai[i+1] - ai[i]; 281 aj = ad->j + ai[i]; 282 for (j=0; j<nzi; j++) { 283 row = aj[j]; 284 pnz = pi[row+1] - pi[row]; 285 Jptr = p_loc->j + pi[row]; 286 /* add non-zero cols of P into the sorted linked list lnk */ 287 ierr = PetscLLCondensedAddSorted_Scalable(pnz,Jptr,lnk);CHKERRQ(ierr); 288 } 289 /* off-diagonal portion: Ao[i,:]*P */ 290 ai = ao->i; pi = p_oth->i; 291 nzi = ai[i+1] - ai[i]; 292 aj = ao->j + ai[i]; 293 for (j=0; j<nzi; j++) { 294 row = aj[j]; 295 pnz = pi[row+1] - pi[row]; 296 Jptr = p_oth->j + pi[row]; 297 ierr = PetscLLCondensedAddSorted_Scalable(pnz,Jptr,lnk);CHKERRQ(ierr); 298 } 299 apnz = lnk[0]; 300 api[i+1] = api[i] + apnz; 301 302 /* if free space is not available, double the total space in the list */ 303 if (current_space->local_remaining<apnz) { 304 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(apnz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 305 nspacedouble++; 306 } 307 308 /* Copy data into free space, then initialize lnk */ 309 ierr = PetscLLCondensedClean_Scalable(apnz,current_space->array,lnk);CHKERRQ(ierr); 310 311 current_space->array += apnz; 312 current_space->local_used += apnz; 313 current_space->local_remaining -= apnz; 314 } 315 /* Allocate space for apj and apv, initialize apj, and */ 316 /* destroy list of free space and other temporary array(s) */ 317 ierr = PetscMalloc2(api[am],&apj,api[am],&apv);CHKERRQ(ierr); 318 ierr = PetscFreeSpaceContiguous(&free_space,apj);CHKERRQ(ierr); 319 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); 320 321 /* Create AP_loc for reuse */ 322 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,am,pN,api,apj,apv,&ptap->AP_loc);CHKERRQ(ierr); 323 324 #if defined(PETSC_USE_INFO) 325 apfill = (PetscReal)api[am]/(ad->i[am]+ao->i[am]+p_loc->i[pm]+1); 326 ptap->AP_loc->info.mallocs = nspacedouble; 327 ptap->AP_loc->info.fill_ratio_given = fill; 328 ptap->AP_loc->info.fill_ratio_needed = apfill; 329 330 if (api[am]) { 331 ierr = PetscInfo3(ptap->AP_loc,"AP_loc reallocs %D; Fill ratio: given %g needed %g.\n",nspacedouble,(double)fill,(double)apfill);CHKERRQ(ierr); 332 ierr = PetscInfo1(ptap->AP_loc,"Use MatPtAP(A,B,MatReuse,%g,&C) for best AP_loc performance.;\n",(double)apfill);CHKERRQ(ierr); 333 } else { 334 ierr = PetscInfo(ptap->AP_loc,"AP_loc is empty \n");CHKERRQ(ierr); 335 } 336 #endif 337 338 /* (2-1) compute symbolic Co = Ro*AP_loc */ 339 /* ------------------------------------ */ 340 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(ptap->Ro,ptap->AP_loc,fill,&ptap->C_oth);CHKERRQ(ierr); 341 342 /* (3) send coj of C_oth to other processors */ 343 /* ------------------------------------------ */ 344 /* determine row ownership */ 345 ierr = PetscLayoutCreate(comm,&rowmap);CHKERRQ(ierr); 346 rowmap->n = pn; 347 rowmap->bs = 1; 348 ierr = PetscLayoutSetUp(rowmap);CHKERRQ(ierr); 349 owners = rowmap->range; 350 351 /* determine the number of messages to send, their lengths */ 352 ierr = PetscMalloc4(size,&len_s,size,&len_si,size,&sstatus,size+2,&owners_co);CHKERRQ(ierr); 353 ierr = PetscMemzero(len_s,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 354 ierr = PetscMemzero(len_si,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 355 356 c_oth = (Mat_SeqAIJ*)ptap->C_oth->data; 357 coi = c_oth->i; coj = c_oth->j; 358 con = ptap->C_oth->rmap->n; 359 proc = 0; 360 for (i=0; i<con; i++) { 361 while (prmap[i] >= owners[proc+1]) proc++; 362 len_si[proc]++; /* num of rows in Co(=Pt*AP) to be sent to [proc] */ 363 len_s[proc] += coi[i+1] - coi[i]; /* num of nonzeros in Co to be sent to [proc] */ 364 } 365 366 len = 0; /* max length of buf_si[], see (4) */ 367 owners_co[0] = 0; 368 nsend = 0; 369 for (proc=0; proc<size; proc++) { 370 owners_co[proc+1] = owners_co[proc] + len_si[proc]; 371 if (len_s[proc]) { 372 nsend++; 373 len_si[proc] = 2*(len_si[proc] + 1); /* length of buf_si to be sent to [proc] */ 374 len += len_si[proc]; 375 } 376 } 377 378 /* determine the number and length of messages to receive for coi and coj */ 379 ierr = PetscGatherNumberOfMessages(comm,NULL,len_s,&nrecv);CHKERRQ(ierr); 380 ierr = PetscGatherMessageLengths2(comm,nsend,nrecv,len_s,len_si,&id_r,&len_r,&len_ri);CHKERRQ(ierr); 381 382 /* post the Irecv and Isend of coj */ 383 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 384 ierr = PetscPostIrecvInt(comm,tagj,nrecv,id_r,len_r,&buf_rj,&rwaits);CHKERRQ(ierr); 385 ierr = PetscMalloc1(nsend+1,&swaits);CHKERRQ(ierr); 386 for (proc=0, k=0; proc<size; proc++) { 387 if (!len_s[proc]) continue; 388 i = owners_co[proc]; 389 ierr = MPI_Isend(coj+coi[i],len_s[proc],MPIU_INT,proc,tagj,comm,swaits+k);CHKERRQ(ierr); 390 k++; 391 } 392 393 /* (2-2) compute symbolic C_loc = Rd*AP_loc */ 394 /* ---------------------------------------- */ 395 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(ptap->Rd,ptap->AP_loc,fill,&ptap->C_loc);CHKERRQ(ierr); 396 c_loc = (Mat_SeqAIJ*)ptap->C_loc->data; 397 398 /* receives coj are complete */ 399 for (i=0; i<nrecv; i++) { 400 ierr = MPI_Waitany(nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 401 } 402 ierr = PetscFree(rwaits);CHKERRQ(ierr); 403 if (nsend) {ierr = MPI_Waitall(nsend,swaits,sstatus);CHKERRQ(ierr);} 404 405 /* add received column indices into ta to update Crmax */ 406 for (k=0; k<nrecv; k++) {/* k-th received message */ 407 Jptr = buf_rj[k]; 408 for (j=0; j<len_r[k]; j++) { 409 ierr = PetscTableAdd(ta,*(Jptr+j)+1,1,INSERT_VALUES);CHKERRQ(ierr); 410 } 411 } 412 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); 413 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 414 415 /* (4) send and recv coi */ 416 /*-----------------------*/ 417 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 418 ierr = PetscPostIrecvInt(comm,tagi,nrecv,id_r,len_ri,&buf_ri,&rwaits);CHKERRQ(ierr); 419 ierr = PetscMalloc1(len+1,&buf_s);CHKERRQ(ierr); 420 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 421 for (proc=0,k=0; proc<size; proc++) { 422 if (!len_s[proc]) continue; 423 /* form outgoing message for i-structure: 424 buf_si[0]: nrows to be sent 425 [1:nrows]: row index (global) 426 [nrows+1:2*nrows+1]: i-structure index 427 */ 428 /*-------------------------------------------*/ 429 nrows = len_si[proc]/2 - 1; /* num of rows in Co to be sent to [proc] */ 430 buf_si_i = buf_si + nrows+1; 431 buf_si[0] = nrows; 432 buf_si_i[0] = 0; 433 nrows = 0; 434 for (i=owners_co[proc]; i<owners_co[proc+1]; i++) { 435 nzi = coi[i+1] - coi[i]; 436 buf_si_i[nrows+1] = buf_si_i[nrows] + nzi; /* i-structure */ 437 buf_si[nrows+1] = prmap[i] -owners[proc]; /* local row index */ 438 nrows++; 439 } 440 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,swaits+k);CHKERRQ(ierr); 441 k++; 442 buf_si += len_si[proc]; 443 } 444 for (i=0; i<nrecv; i++) { 445 ierr = MPI_Waitany(nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 446 } 447 ierr = PetscFree(rwaits);CHKERRQ(ierr); 448 if (nsend) {ierr = MPI_Waitall(nsend,swaits,sstatus);CHKERRQ(ierr);} 449 450 ierr = PetscFree4(len_s,len_si,sstatus,owners_co);CHKERRQ(ierr); 451 ierr = PetscFree(len_ri);CHKERRQ(ierr); 452 ierr = PetscFree(swaits);CHKERRQ(ierr); 453 ierr = PetscFree(buf_s);CHKERRQ(ierr); 454 455 /* (5) compute the local portion of Cmpi */ 456 /* ------------------------------------------ */ 457 /* set initial free space to be Crmax, sufficient for holding nozeros in each row of Cmpi */ 458 ierr = PetscFreeSpaceGet(Crmax,&free_space);CHKERRQ(ierr); 459 current_space = free_space; 460 461 ierr = PetscMalloc3(nrecv,&buf_ri_k,nrecv,&nextrow,nrecv,&nextci);CHKERRQ(ierr); 462 for (k=0; k<nrecv; k++) { 463 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 464 nrows = *buf_ri_k[k]; 465 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 466 nextci[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 467 } 468 469 ierr = MatPreallocateInitialize(comm,pn,pn,dnz,onz);CHKERRQ(ierr); 470 ierr = PetscLLCondensedCreate_Scalable(Crmax,&lnk);CHKERRQ(ierr); 471 for (i=0; i<pn; i++) { 472 /* add C_loc into Cmpi */ 473 nzi = c_loc->i[i+1] - c_loc->i[i]; 474 Jptr = c_loc->j + c_loc->i[i]; 475 ierr = PetscLLCondensedAddSorted_Scalable(nzi,Jptr,lnk);CHKERRQ(ierr); 476 477 /* add received col data into lnk */ 478 for (k=0; k<nrecv; k++) { /* k-th received message */ 479 if (i == *nextrow[k]) { /* i-th row */ 480 nzi = *(nextci[k]+1) - *nextci[k]; 481 Jptr = buf_rj[k] + *nextci[k]; 482 ierr = PetscLLCondensedAddSorted_Scalable(nzi,Jptr,lnk);CHKERRQ(ierr); 483 nextrow[k]++; nextci[k]++; 484 } 485 } 486 nzi = lnk[0]; 487 488 /* copy data into free space, then initialize lnk */ 489 ierr = PetscLLCondensedClean_Scalable(nzi,current_space->array,lnk);CHKERRQ(ierr); 490 ierr = MatPreallocateSet(i+owners[rank],nzi,current_space->array,dnz,onz);CHKERRQ(ierr); 491 } 492 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 493 ierr = PetscLLCondensedDestroy_Scalable(lnk);CHKERRQ(ierr); 494 ierr = PetscFreeSpaceDestroy(free_space);CHKERRQ(ierr); 495 496 /* local sizes and preallocation */ 497 ierr = MatSetSizes(Cmpi,pn,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 498 ierr = MatSetBlockSizes(Cmpi,PetscAbs(P->cmap->bs),PetscAbs(P->cmap->bs));CHKERRQ(ierr); 499 ierr = MatMPIAIJSetPreallocation(Cmpi,0,dnz,0,onz);CHKERRQ(ierr); 500 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 501 502 /* members in merge */ 503 ierr = PetscFree(id_r);CHKERRQ(ierr); 504 ierr = PetscFree(len_r);CHKERRQ(ierr); 505 ierr = PetscFree(buf_ri[0]);CHKERRQ(ierr); 506 ierr = PetscFree(buf_ri);CHKERRQ(ierr); 507 ierr = PetscFree(buf_rj[0]);CHKERRQ(ierr); 508 ierr = PetscFree(buf_rj);CHKERRQ(ierr); 509 ierr = PetscLayoutDestroy(&rowmap);CHKERRQ(ierr); 510 511 /* attach the supporting struct to Cmpi for reuse */ 512 c = (Mat_MPIAIJ*)Cmpi->data; 513 c->ptap = ptap; 514 ptap->duplicate = Cmpi->ops->duplicate; 515 ptap->destroy = Cmpi->ops->destroy; 516 517 /* Cmpi is not ready for use - assembly will be done by MatPtAPNumeric() */ 518 Cmpi->assembled = PETSC_FALSE; 519 Cmpi->ops->destroy = MatDestroy_MPIAIJ_PtAP; 520 Cmpi->ops->duplicate = MatDuplicate_MPIAIJ_MatPtAP; 521 *C = Cmpi; 522 PetscFunctionReturn(0); 523 } 524 525 PetscErrorCode MatPtAPSymbolic_MPIAIJ_MPIAIJ(Mat A,Mat P,PetscReal fill,Mat *C) 526 { 527 PetscErrorCode ierr; 528 Mat_PtAPMPI *ptap; 529 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c; 530 MPI_Comm comm; 531 PetscMPIInt size,rank; 532 Mat Cmpi; 533 PetscFreeSpaceList free_space=NULL,current_space=NULL; 534 PetscInt am=A->rmap->n,pm=P->rmap->n,pN=P->cmap->N,pn=P->cmap->n; 535 PetscInt *lnk,i,k,pnz,row,nsend; 536 PetscBT lnkbt; 537 PetscMPIInt tagi,tagj,*len_si,*len_s,*len_ri,icompleted=0,nrecv; 538 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 539 PetscInt len,proc,*dnz,*onz,*owners,nzi,nspacedouble; 540 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextci; 541 MPI_Request *swaits,*rwaits; 542 MPI_Status *sstatus,rstatus; 543 PetscLayout rowmap; 544 PetscInt *owners_co,*coi,*coj; /* i and j array of (p->B)^T*A*P - used in the communication */ 545 PetscMPIInt *len_r,*id_r; /* array of length of comm->size, store send/recv matrix values */ 546 PetscInt *api,*apj,*Jptr,apnz,*prmap=p->garray,con,j,ap_rmax=0,Crmax,*aj,*ai,*pi; 547 Mat_SeqAIJ *p_loc,*p_oth,*ad=(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data,*c_loc,*c_oth; 548 PetscScalar *apv; 549 PetscTable ta; 550 #if defined(PETSC_HAVE_HYPRE) 551 const char *algTypes[3] = {"scalable","nonscalable","hypre"}; 552 PetscInt nalg = 3; 553 #else 554 const char *algTypes[2] = {"scalable","nonscalable"}; 555 PetscInt nalg = 2; 556 #endif 557 PetscInt alg = 1; /* set default algorithm */ 558 #if defined(PETSC_USE_INFO) 559 PetscReal apfill; 560 #endif 561 #if defined(PTAP_PROFILE) 562 PetscLogDouble t0,t1,t11,t12,t2,t3,t4; 563 #endif 564 PetscBool flg; 565 566 PetscFunctionBegin; 567 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 568 569 /* pick an algorithm */ 570 ierr = PetscObjectOptionsBegin((PetscObject)A);CHKERRQ(ierr); 571 ierr = PetscOptionsEList("-matptap_via","Algorithmic approach","MatPtAP",algTypes,nalg,algTypes[1],&alg,&flg);CHKERRQ(ierr); 572 ierr = PetscOptionsEnd();CHKERRQ(ierr); 573 574 if (!flg && pN > 100000) { /* may switch to scalable algorithm as default */ 575 MatInfo Ainfo,Pinfo; 576 PetscInt nz_local; 577 PetscBool alg_scalable_loc=PETSC_FALSE,alg_scalable; 578 579 ierr = MatGetInfo(A,MAT_LOCAL,&Ainfo);CHKERRQ(ierr); 580 ierr = MatGetInfo(P,MAT_LOCAL,&Pinfo);CHKERRQ(ierr); 581 nz_local = (PetscInt)(Ainfo.nz_allocated + Pinfo.nz_allocated); 582 583 if (pN > fill*nz_local) alg_scalable_loc = PETSC_TRUE; 584 ierr = MPIU_Allreduce(&alg_scalable_loc,&alg_scalable,1,MPIU_BOOL,MPI_LOR,comm);CHKERRQ(ierr); 585 586 if (alg_scalable) { 587 alg = 0; /* scalable algorithm would 50% slower than nonscalable algorithm */ 588 ierr = PetscInfo2(P,"Use scalable algorithm, pN %D, fill*nz_allocated %g\n",pN,fill*nz_local);CHKERRQ(ierr); 589 } 590 } 591 /* ierr = PetscInfo2(P,"Use algorithm %D, pN %D\n",alg,pN);CHKERRQ(ierr); */ 592 593 if (alg == 0) { 594 ierr = MatPtAPSymbolic_MPIAIJ_MPIAIJ_scalable(A,P,fill,C);CHKERRQ(ierr); 595 (*C)->ops->ptapnumeric = MatPtAPNumeric_MPIAIJ_MPIAIJ_scalable; 596 PetscFunctionReturn(0); 597 598 #if defined(PETSC_HAVE_HYPRE) 599 } else if (alg == 2) { 600 /* Use boomerAMGBuildCoarseOperator */ 601 ierr = MatPtAPSymbolic_AIJ_AIJ_wHYPRE(A,P,fill,C);CHKERRQ(ierr); 602 PetscFunctionReturn(0); 603 #endif 604 } 605 606 #if defined(PTAP_PROFILE) 607 ierr = PetscTime(&t0);CHKERRQ(ierr); 608 #endif 609 610 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 611 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 612 613 /* create symbolic parallel matrix Cmpi */ 614 ierr = MatCreate(comm,&Cmpi);CHKERRQ(ierr); 615 ierr = MatSetType(Cmpi,MATMPIAIJ);CHKERRQ(ierr); 616 617 /* Do dense axpy in MatPtAPNumeric_MPIAIJ_MPIAIJ() */ 618 Cmpi->ops->ptapnumeric = MatPtAPNumeric_MPIAIJ_MPIAIJ; 619 620 /* create struct Mat_PtAPMPI and attached it to C later */ 621 ierr = PetscNew(&ptap);CHKERRQ(ierr); 622 ptap->reuse = MAT_INITIAL_MATRIX; 623 624 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 625 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 626 /* get P_loc by taking all local rows of P */ 627 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 628 629 /* (0) compute Rd = Pd^T, Ro = Po^T */ 630 /* --------------------------------- */ 631 ierr = MatTranspose_SeqAIJ(p->A,MAT_INITIAL_MATRIX,&ptap->Rd);CHKERRQ(ierr); 632 ierr = MatTranspose_SeqAIJ(p->B,MAT_INITIAL_MATRIX,&ptap->Ro);CHKERRQ(ierr); 633 #if defined(PTAP_PROFILE) 634 ierr = PetscTime(&t11);CHKERRQ(ierr); 635 #endif 636 637 /* (1) compute symbolic AP = A_loc*P = Ad*P_loc + Ao*P_oth (api,apj) */ 638 /* ----------------------------------------------------------------- */ 639 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 640 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 641 642 /* create and initialize a linked list */ 643 ierr = PetscTableCreate(pn,pN,&ta);CHKERRQ(ierr); /* for compute AP_loc and Cmpi */ 644 MatRowMergeMax_SeqAIJ(p_loc,ptap->P_loc->rmap->N,ta); 645 MatRowMergeMax_SeqAIJ(p_oth,ptap->P_oth->rmap->N,ta); 646 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); /* Crmax = nnz(sum of Prows) */ 647 /* printf("[%d] est %d, Crmax %d; pN %d\n",rank,5*(p_loc->rmax+p_oth->rmax + (PetscInt)(1.e-2*pN)),Crmax,pN); */ 648 649 ierr = PetscLLCondensedCreate(Crmax,pN,&lnk,&lnkbt);CHKERRQ(ierr); 650 651 /* Initial FreeSpace size is fill*(nnz(A) + nnz(P)) */ 652 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(ad->i[am],PetscIntSumTruncate(ao->i[am],p_loc->i[pm]))),&free_space);CHKERRQ(ierr); 653 current_space = free_space; 654 nspacedouble = 0; 655 656 ierr = PetscMalloc1(am+1,&api);CHKERRQ(ierr); 657 api[0] = 0; 658 for (i=0; i<am; i++) { 659 /* diagonal portion: Ad[i,:]*P */ 660 ai = ad->i; pi = p_loc->i; 661 nzi = ai[i+1] - ai[i]; 662 aj = ad->j + ai[i]; 663 for (j=0; j<nzi; j++) { 664 row = aj[j]; 665 pnz = pi[row+1] - pi[row]; 666 Jptr = p_loc->j + pi[row]; 667 /* add non-zero cols of P into the sorted linked list lnk */ 668 ierr = PetscLLCondensedAddSorted(pnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 669 } 670 /* off-diagonal portion: Ao[i,:]*P */ 671 ai = ao->i; pi = p_oth->i; 672 nzi = ai[i+1] - ai[i]; 673 aj = ao->j + ai[i]; 674 for (j=0; j<nzi; j++) { 675 row = aj[j]; 676 pnz = pi[row+1] - pi[row]; 677 Jptr = p_oth->j + pi[row]; 678 ierr = PetscLLCondensedAddSorted(pnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 679 } 680 apnz = lnk[0]; 681 api[i+1] = api[i] + apnz; 682 if (ap_rmax < apnz) ap_rmax = apnz; 683 684 /* if free space is not available, double the total space in the list */ 685 if (current_space->local_remaining<apnz) { 686 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(apnz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 687 nspacedouble++; 688 } 689 690 /* Copy data into free space, then initialize lnk */ 691 ierr = PetscLLCondensedClean(pN,apnz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 692 693 current_space->array += apnz; 694 current_space->local_used += apnz; 695 current_space->local_remaining -= apnz; 696 } 697 /* Allocate space for apj and apv, initialize apj, and */ 698 /* destroy list of free space and other temporary array(s) */ 699 ierr = PetscMalloc2(api[am],&apj,api[am],&apv);CHKERRQ(ierr); 700 ierr = PetscFreeSpaceContiguous(&free_space,apj);CHKERRQ(ierr); 701 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 702 703 /* Create AP_loc for reuse */ 704 ierr = MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,am,pN,api,apj,apv,&ptap->AP_loc);CHKERRQ(ierr); 705 706 #if defined(PETSC_USE_INFO) 707 apfill = (PetscReal)api[am]/(ad->i[am]+ao->i[am]+p_loc->i[pm]+1); 708 ptap->AP_loc->info.mallocs = nspacedouble; 709 ptap->AP_loc->info.fill_ratio_given = fill; 710 ptap->AP_loc->info.fill_ratio_needed = apfill; 711 712 if (api[am]) { 713 ierr = PetscInfo3(ptap->AP_loc,"AP_loc reallocs %D; Fill ratio: given %g needed %g.\n",nspacedouble,(double)fill,(double)apfill);CHKERRQ(ierr); 714 ierr = PetscInfo1(ptap->AP_loc,"Use MatPtAP(A,B,MatReuse,%g,&C) for best AP_loc performance.;\n",(double)apfill);CHKERRQ(ierr); 715 } else { 716 ierr = PetscInfo(ptap->AP_loc,"AP_loc is empty \n");CHKERRQ(ierr); 717 } 718 #endif 719 720 #if defined(PTAP_PROFILE) 721 ierr = PetscTime(&t12);CHKERRQ(ierr); 722 #endif 723 724 /* (2-1) compute symbolic Co = Ro*AP_loc */ 725 /* ------------------------------------ */ 726 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(ptap->Ro,ptap->AP_loc,fill,&ptap->C_oth);CHKERRQ(ierr); 727 #if defined(PTAP_PROFILE) 728 ierr = PetscTime(&t1);CHKERRQ(ierr); 729 #endif 730 731 /* (3) send coj of C_oth to other processors */ 732 /* ------------------------------------------ */ 733 /* determine row ownership */ 734 ierr = PetscLayoutCreate(comm,&rowmap);CHKERRQ(ierr); 735 rowmap->n = pn; 736 rowmap->bs = 1; 737 ierr = PetscLayoutSetUp(rowmap);CHKERRQ(ierr); 738 owners = rowmap->range; 739 740 /* determine the number of messages to send, their lengths */ 741 ierr = PetscMalloc4(size,&len_s,size,&len_si,size,&sstatus,size+2,&owners_co);CHKERRQ(ierr); 742 ierr = PetscMemzero(len_s,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 743 ierr = PetscMemzero(len_si,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 744 745 c_oth = (Mat_SeqAIJ*)ptap->C_oth->data; 746 coi = c_oth->i; coj = c_oth->j; 747 con = ptap->C_oth->rmap->n; 748 proc = 0; 749 for (i=0; i<con; i++) { 750 while (prmap[i] >= owners[proc+1]) proc++; 751 len_si[proc]++; /* num of rows in Co(=Pt*AP) to be sent to [proc] */ 752 len_s[proc] += coi[i+1] - coi[i]; /* num of nonzeros in Co to be sent to [proc] */ 753 } 754 755 len = 0; /* max length of buf_si[], see (4) */ 756 owners_co[0] = 0; 757 nsend = 0; 758 for (proc=0; proc<size; proc++) { 759 owners_co[proc+1] = owners_co[proc] + len_si[proc]; 760 if (len_s[proc]) { 761 nsend++; 762 len_si[proc] = 2*(len_si[proc] + 1); /* length of buf_si to be sent to [proc] */ 763 len += len_si[proc]; 764 } 765 } 766 767 /* determine the number and length of messages to receive for coi and coj */ 768 ierr = PetscGatherNumberOfMessages(comm,NULL,len_s,&nrecv);CHKERRQ(ierr); 769 ierr = PetscGatherMessageLengths2(comm,nsend,nrecv,len_s,len_si,&id_r,&len_r,&len_ri);CHKERRQ(ierr); 770 771 /* post the Irecv and Isend of coj */ 772 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 773 ierr = PetscPostIrecvInt(comm,tagj,nrecv,id_r,len_r,&buf_rj,&rwaits);CHKERRQ(ierr); 774 ierr = PetscMalloc1(nsend+1,&swaits);CHKERRQ(ierr); 775 for (proc=0, k=0; proc<size; proc++) { 776 if (!len_s[proc]) continue; 777 i = owners_co[proc]; 778 ierr = MPI_Isend(coj+coi[i],len_s[proc],MPIU_INT,proc,tagj,comm,swaits+k);CHKERRQ(ierr); 779 k++; 780 } 781 782 /* (2-2) compute symbolic C_loc = Rd*AP_loc */ 783 /* ---------------------------------------- */ 784 ierr = MatMatMultSymbolic_SeqAIJ_SeqAIJ(ptap->Rd,ptap->AP_loc,fill,&ptap->C_loc);CHKERRQ(ierr); 785 c_loc = (Mat_SeqAIJ*)ptap->C_loc->data; 786 787 /* receives coj are complete */ 788 for (i=0; i<nrecv; i++) { 789 ierr = MPI_Waitany(nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 790 } 791 ierr = PetscFree(rwaits);CHKERRQ(ierr); 792 if (nsend) {ierr = MPI_Waitall(nsend,swaits,sstatus);CHKERRQ(ierr);} 793 794 /* add received column indices into ta to update Crmax */ 795 for (k=0; k<nrecv; k++) {/* k-th received message */ 796 Jptr = buf_rj[k]; 797 for (j=0; j<len_r[k]; j++) { 798 ierr = PetscTableAdd(ta,*(Jptr+j)+1,1,INSERT_VALUES);CHKERRQ(ierr); 799 } 800 } 801 ierr = PetscTableGetCount(ta,&Crmax);CHKERRQ(ierr); 802 ierr = PetscTableDestroy(&ta);CHKERRQ(ierr); 803 804 /* (4) send and recv coi */ 805 /*-----------------------*/ 806 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 807 ierr = PetscPostIrecvInt(comm,tagi,nrecv,id_r,len_ri,&buf_ri,&rwaits);CHKERRQ(ierr); 808 ierr = PetscMalloc1(len+1,&buf_s);CHKERRQ(ierr); 809 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 810 for (proc=0,k=0; proc<size; proc++) { 811 if (!len_s[proc]) continue; 812 /* form outgoing message for i-structure: 813 buf_si[0]: nrows to be sent 814 [1:nrows]: row index (global) 815 [nrows+1:2*nrows+1]: i-structure index 816 */ 817 /*-------------------------------------------*/ 818 nrows = len_si[proc]/2 - 1; /* num of rows in Co to be sent to [proc] */ 819 buf_si_i = buf_si + nrows+1; 820 buf_si[0] = nrows; 821 buf_si_i[0] = 0; 822 nrows = 0; 823 for (i=owners_co[proc]; i<owners_co[proc+1]; i++) { 824 nzi = coi[i+1] - coi[i]; 825 buf_si_i[nrows+1] = buf_si_i[nrows] + nzi; /* i-structure */ 826 buf_si[nrows+1] = prmap[i] -owners[proc]; /* local row index */ 827 nrows++; 828 } 829 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,swaits+k);CHKERRQ(ierr); 830 k++; 831 buf_si += len_si[proc]; 832 } 833 for (i=0; i<nrecv; i++) { 834 ierr = MPI_Waitany(nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 835 } 836 ierr = PetscFree(rwaits);CHKERRQ(ierr); 837 if (nsend) {ierr = MPI_Waitall(nsend,swaits,sstatus);CHKERRQ(ierr);} 838 839 ierr = PetscFree4(len_s,len_si,sstatus,owners_co);CHKERRQ(ierr); 840 ierr = PetscFree(len_ri);CHKERRQ(ierr); 841 ierr = PetscFree(swaits);CHKERRQ(ierr); 842 ierr = PetscFree(buf_s);CHKERRQ(ierr); 843 #if defined(PTAP_PROFILE) 844 ierr = PetscTime(&t2);CHKERRQ(ierr); 845 #endif 846 /* (5) compute the local portion of Cmpi */ 847 /* ------------------------------------------ */ 848 /* set initial free space to be Crmax, sufficient for holding nozeros in each row of Cmpi */ 849 ierr = PetscFreeSpaceGet(Crmax,&free_space);CHKERRQ(ierr); 850 current_space = free_space; 851 852 ierr = PetscMalloc3(nrecv,&buf_ri_k,nrecv,&nextrow,nrecv,&nextci);CHKERRQ(ierr); 853 for (k=0; k<nrecv; k++) { 854 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 855 nrows = *buf_ri_k[k]; 856 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 857 nextci[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 858 } 859 860 ierr = MatPreallocateInitialize(comm,pn,pn,dnz,onz);CHKERRQ(ierr); 861 ierr = PetscLLCondensedCreate(Crmax,pN,&lnk,&lnkbt);CHKERRQ(ierr); 862 for (i=0; i<pn; i++) { 863 /* add C_loc into Cmpi */ 864 nzi = c_loc->i[i+1] - c_loc->i[i]; 865 Jptr = c_loc->j + c_loc->i[i]; 866 ierr = PetscLLCondensedAddSorted(nzi,Jptr,lnk,lnkbt);CHKERRQ(ierr); 867 868 /* add received col data into lnk */ 869 for (k=0; k<nrecv; k++) { /* k-th received message */ 870 if (i == *nextrow[k]) { /* i-th row */ 871 nzi = *(nextci[k]+1) - *nextci[k]; 872 Jptr = buf_rj[k] + *nextci[k]; 873 ierr = PetscLLCondensedAddSorted(nzi,Jptr,lnk,lnkbt);CHKERRQ(ierr); 874 nextrow[k]++; nextci[k]++; 875 } 876 } 877 nzi = lnk[0]; 878 879 /* copy data into free space, then initialize lnk */ 880 ierr = PetscLLCondensedClean(pN,nzi,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 881 ierr = MatPreallocateSet(i+owners[rank],nzi,current_space->array,dnz,onz);CHKERRQ(ierr); 882 } 883 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 884 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 885 ierr = PetscFreeSpaceDestroy(free_space);CHKERRQ(ierr); 886 #if defined(PTAP_PROFILE) 887 ierr = PetscTime(&t3);CHKERRQ(ierr); 888 #endif 889 890 /* local sizes and preallocation */ 891 ierr = MatSetSizes(Cmpi,pn,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 892 ierr = MatSetBlockSizes(Cmpi,PetscAbs(P->cmap->bs),PetscAbs(P->cmap->bs));CHKERRQ(ierr); 893 ierr = MatMPIAIJSetPreallocation(Cmpi,0,dnz,0,onz);CHKERRQ(ierr); 894 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 895 896 /* members in merge */ 897 ierr = PetscFree(id_r);CHKERRQ(ierr); 898 ierr = PetscFree(len_r);CHKERRQ(ierr); 899 ierr = PetscFree(buf_ri[0]);CHKERRQ(ierr); 900 ierr = PetscFree(buf_ri);CHKERRQ(ierr); 901 ierr = PetscFree(buf_rj[0]);CHKERRQ(ierr); 902 ierr = PetscFree(buf_rj);CHKERRQ(ierr); 903 ierr = PetscLayoutDestroy(&rowmap);CHKERRQ(ierr); 904 905 /* attach the supporting struct to Cmpi for reuse */ 906 c = (Mat_MPIAIJ*)Cmpi->data; 907 c->ptap = ptap; 908 ptap->duplicate = Cmpi->ops->duplicate; 909 ptap->destroy = Cmpi->ops->destroy; 910 911 if (alg == 1) { 912 ierr = PetscCalloc1(pN,&ptap->apa);CHKERRQ(ierr); 913 } 914 915 /* Cmpi is not ready for use - assembly will be done by MatPtAPNumeric() */ 916 Cmpi->assembled = PETSC_FALSE; 917 Cmpi->ops->destroy = MatDestroy_MPIAIJ_PtAP; 918 Cmpi->ops->duplicate = MatDuplicate_MPIAIJ_MatPtAP; 919 *C = Cmpi; 920 921 #if defined(PTAP_PROFILE) 922 ierr = PetscTime(&t4);CHKERRQ(ierr); 923 if (rank == 1) { 924 printf("PtAPSym: %g + %g + %g + %g + %g + %g = %g\n",t11-t0,t1-t11,t12-t11,t2-t2,t3-t2,t4-t3,t4-t0); 925 } 926 #endif 927 PetscFunctionReturn(0); 928 } 929 930 PetscErrorCode MatPtAPNumeric_MPIAIJ_MPIAIJ(Mat A,Mat P,Mat C) 931 { 932 PetscErrorCode ierr; 933 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c=(Mat_MPIAIJ*)C->data; 934 Mat_SeqAIJ *ad=(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 935 Mat_SeqAIJ *ap,*p_loc,*p_oth,*c_seq; 936 Mat_PtAPMPI *ptap = c->ptap; 937 Mat AP_loc,C_loc,C_oth; 938 PetscInt i,rstart,rend,cm,ncols,row; 939 PetscInt *api,*apj,am = A->rmap->n,j,col,apnz; 940 PetscScalar *apa; 941 const PetscInt *cols; 942 const PetscScalar *vals; 943 #if defined(PTAP_PROFILE) 944 PetscMPIInt rank; 945 MPI_Comm comm; 946 PetscLogDouble t0,t1,t2,t3,t4,eR,eAP,eCseq,eCmpi; 947 #endif 948 949 PetscFunctionBegin; 950 ierr = MatZeroEntries(C);CHKERRQ(ierr); 951 952 /* 1) get R = Pd^T,Ro = Po^T */ 953 #if defined(PTAP_PROFILE) 954 ierr = PetscTime(&t0);CHKERRQ(ierr); 955 #endif 956 if (ptap->reuse == MAT_REUSE_MATRIX) { 957 ierr = MatTranspose_SeqAIJ(p->A,MAT_REUSE_MATRIX,&ptap->Rd);CHKERRQ(ierr); 958 ierr = MatTranspose_SeqAIJ(p->B,MAT_REUSE_MATRIX,&ptap->Ro);CHKERRQ(ierr); 959 } 960 #if defined(PTAP_PROFILE) 961 ierr = PetscTime(&t1);CHKERRQ(ierr); 962 eR = t1 - t0; 963 #endif 964 965 /* 2) get AP_loc */ 966 AP_loc = ptap->AP_loc; 967 ap = (Mat_SeqAIJ*)AP_loc->data; 968 969 /* 2-1) get P_oth = ptap->P_oth and P_loc = ptap->P_loc */ 970 /*-----------------------------------------------------*/ 971 if (ptap->reuse == MAT_REUSE_MATRIX) { 972 /* P_oth and P_loc are obtained in MatPtASymbolic() when reuse == MAT_INITIAL_MATRIX */ 973 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_REUSE_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 974 ierr = MatMPIAIJGetLocalMat(P,MAT_REUSE_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 975 } 976 977 /* 2-2) compute numeric A_loc*P - dominating part */ 978 /* ---------------------------------------------- */ 979 /* get data from symbolic products */ 980 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 981 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 982 apa = ptap->apa; 983 api = ap->i; 984 apj = ap->j; 985 for (i=0; i<am; i++) { 986 /* AP[i,:] = A[i,:]*P = Ad*P_loc Ao*P_oth */ 987 AProw_nonscalable(i,ad,ao,p_loc,p_oth,apa); 988 apnz = api[i+1] - api[i]; 989 for (j=0; j<apnz; j++) { 990 col = apj[j+api[i]]; 991 ap->a[j+ap->i[i]] = apa[col]; 992 apa[col] = 0.0; 993 } 994 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 995 } 996 #if defined(PTAP_PROFILE) 997 ierr = PetscTime(&t2);CHKERRQ(ierr); 998 eAP = t2 - t1; 999 #endif 1000 1001 /* 3) C_loc = Rd*AP_loc, C_oth = Ro*AP_loc */ 1002 ierr = ((ptap->C_loc)->ops->matmultnumeric)(ptap->Rd,AP_loc,ptap->C_loc);CHKERRQ(ierr); 1003 ierr = ((ptap->C_oth)->ops->matmultnumeric)(ptap->Ro,AP_loc,ptap->C_oth);CHKERRQ(ierr); 1004 C_loc = ptap->C_loc; 1005 C_oth = ptap->C_oth; 1006 #if defined(PTAP_PROFILE) 1007 ierr = PetscTime(&t3);CHKERRQ(ierr); 1008 eCseq = t3 - t2; 1009 #endif 1010 1011 /* add C_loc and Co to to C */ 1012 ierr = MatGetOwnershipRange(C,&rstart,&rend);CHKERRQ(ierr); 1013 1014 /* C_loc -> C */ 1015 cm = C_loc->rmap->N; 1016 c_seq = (Mat_SeqAIJ*)C_loc->data; 1017 cols = c_seq->j; 1018 vals = c_seq->a; 1019 for (i=0; i<cm; i++) { 1020 ncols = c_seq->i[i+1] - c_seq->i[i]; 1021 row = rstart + i; 1022 ierr = MatSetValues(C,1,&row,ncols,cols,vals,ADD_VALUES);CHKERRQ(ierr); 1023 cols += ncols; vals += ncols; 1024 } 1025 1026 /* Co -> C, off-processor part */ 1027 cm = C_oth->rmap->N; 1028 c_seq = (Mat_SeqAIJ*)C_oth->data; 1029 cols = c_seq->j; 1030 vals = c_seq->a; 1031 for (i=0; i<cm; i++) { 1032 ncols = c_seq->i[i+1] - c_seq->i[i]; 1033 row = p->garray[i]; 1034 ierr = MatSetValues(C,1,&row,ncols,cols,vals,ADD_VALUES);CHKERRQ(ierr); 1035 cols += ncols; vals += ncols; 1036 } 1037 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1038 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1039 #if defined(PTAP_PROFILE) 1040 ierr = PetscTime(&t4);CHKERRQ(ierr); 1041 eCmpi = t4 - t3; 1042 1043 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 1044 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 1045 if (rank==1) { 1046 ierr = PetscPrintf(MPI_COMM_SELF," R %g, AP %g, Cseq %g, Cmpi %g = %g\n", eR,eAP,eCseq,eCmpi,eR+eAP+eCseq+eCmpi);CHKERRQ(ierr); 1047 } 1048 #endif 1049 ptap->reuse = MAT_REUSE_MATRIX; 1050 PetscFunctionReturn(0); 1051 } 1052 1053 PetscErrorCode MatPtAPSymbolic_MPIAIJ_MPIAIJ_ptap(Mat A,Mat P,PetscReal fill,Mat *C) 1054 { 1055 PetscErrorCode ierr; 1056 Mat Cmpi; 1057 Mat_PtAPMPI *ptap; 1058 PetscFreeSpaceList free_space=NULL,current_space=NULL; 1059 Mat_MPIAIJ *a =(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c; 1060 Mat_SeqAIJ *ad =(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 1061 Mat_SeqAIJ *p_loc,*p_oth; 1062 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*pdti,*pdtj,*poti,*potj,*ptJ; 1063 PetscInt *adi=ad->i,*aj,*aoi=ao->i,nnz; 1064 PetscInt *lnk,*owners_co,*coi,*coj,i,k,pnz,row; 1065 PetscInt am=A->rmap->n,pN=P->cmap->N,pm=P->rmap->n,pn=P->cmap->n; 1066 PetscBT lnkbt; 1067 MPI_Comm comm; 1068 PetscMPIInt size,rank,tagi,tagj,*len_si,*len_s,*len_ri,icompleted=0; 1069 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 1070 PetscInt len,proc,*dnz,*onz,*owners; 1071 PetscInt nzi,*pti,*ptj; 1072 PetscInt nrows,*buf_s,*buf_si,*buf_si_i,**nextrow,**nextci; 1073 MPI_Request *swaits,*rwaits; 1074 MPI_Status *sstatus,rstatus; 1075 Mat_Merge_SeqsToMPI *merge; 1076 PetscInt *api,*apj,*Jptr,apnz,*prmap=p->garray,pon,nspacedouble=0,j,ap_rmax=0; 1077 PetscReal afill=1.0,afill_tmp; 1078 1079 PetscFunctionBegin; 1080 ierr = PetscObjectGetComm((PetscObject)A,&comm);CHKERRQ(ierr); 1081 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 1082 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 1083 1084 /* create struct Mat_PtAPMPI and attached it to C later */ 1085 ierr = PetscNew(&ptap);CHKERRQ(ierr); 1086 ierr = PetscNew(&merge);CHKERRQ(ierr); 1087 ptap->merge = merge; 1088 ptap->reuse = MAT_INITIAL_MATRIX; 1089 1090 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 1091 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 1092 1093 /* get P_loc by taking all local rows of P */ 1094 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 1095 1096 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 1097 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 1098 pi_loc = p_loc->i; pj_loc = p_loc->j; 1099 pi_oth = p_oth->i; pj_oth = p_oth->j; 1100 1101 /* (1) compute symbolic AP = A_loc*P = A_diag*P_loc + A_off*P_oth (api,apj) */ 1102 /*--------------------------------------------------------------------------*/ 1103 ierr = PetscMalloc1(am+1,&api);CHKERRQ(ierr); 1104 api[0] = 0; 1105 1106 /* create and initialize a linked list */ 1107 ierr = PetscLLCondensedCreate(pN,pN,&lnk,&lnkbt);CHKERRQ(ierr); 1108 1109 /* Initial FreeSpace size is fill*(nnz(A) + nnz(P)) -OOM for ex56, np=8k on Intrepid! */ 1110 ierr = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,PetscIntSumTruncate(adi[am],PetscIntSumTruncate(aoi[am],pi_loc[pm]))),&free_space);CHKERRQ(ierr); 1111 current_space = free_space; 1112 1113 for (i=0; i<am; i++) { 1114 /* diagonal portion of A */ 1115 nzi = adi[i+1] - adi[i]; 1116 aj = ad->j + adi[i]; 1117 for (j=0; j<nzi; j++) { 1118 row = aj[j]; 1119 pnz = pi_loc[row+1] - pi_loc[row]; 1120 Jptr = pj_loc + pi_loc[row]; 1121 /* add non-zero cols of P into the sorted linked list lnk */ 1122 ierr = PetscLLCondensedAddSorted(pnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 1123 } 1124 /* off-diagonal portion of A */ 1125 nzi = aoi[i+1] - aoi[i]; 1126 aj = ao->j + aoi[i]; 1127 for (j=0; j<nzi; j++) { 1128 row = aj[j]; 1129 pnz = pi_oth[row+1] - pi_oth[row]; 1130 Jptr = pj_oth + pi_oth[row]; 1131 ierr = PetscLLCondensedAddSorted(pnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 1132 } 1133 apnz = lnk[0]; 1134 api[i+1] = api[i] + apnz; 1135 if (ap_rmax < apnz) ap_rmax = apnz; 1136 1137 /* if free space is not available, double the total space in the list */ 1138 if (current_space->local_remaining<apnz) { 1139 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(apnz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 1140 nspacedouble++; 1141 } 1142 1143 /* Copy data into free space, then initialize lnk */ 1144 ierr = PetscLLCondensedClean(pN,apnz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 1145 1146 current_space->array += apnz; 1147 current_space->local_used += apnz; 1148 current_space->local_remaining -= apnz; 1149 } 1150 1151 /* Allocate space for apj, initialize apj, and */ 1152 /* destroy list of free space and other temporary array(s) */ 1153 ierr = PetscMalloc1(api[am]+1,&apj);CHKERRQ(ierr); 1154 ierr = PetscFreeSpaceContiguous(&free_space,apj);CHKERRQ(ierr); 1155 afill_tmp = (PetscReal)api[am]/(adi[am]+aoi[am]+pi_loc[pm]+1); 1156 if (afill_tmp > afill) afill = afill_tmp; 1157 1158 /* (2) determine symbolic Co=(p->B)^T*AP - send to others (coi,coj)*/ 1159 /*-----------------------------------------------------------------*/ 1160 ierr = MatGetSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 1161 1162 /* then, compute symbolic Co = (p->B)^T*AP */ 1163 pon = (p->B)->cmap->n; /* total num of rows to be sent to other processors 1164 >= (num of nonzero rows of C_seq) - pn */ 1165 ierr = PetscMalloc1(pon+1,&coi);CHKERRQ(ierr); 1166 coi[0] = 0; 1167 1168 /* set initial free space to be fill*(nnz(p->B) + nnz(AP)) */ 1169 nnz = PetscRealIntMultTruncate(fill,PetscIntSumTruncate(poti[pon],api[am])); 1170 ierr = PetscFreeSpaceGet(nnz,&free_space);CHKERRQ(ierr); 1171 current_space = free_space; 1172 1173 for (i=0; i<pon; i++) { 1174 pnz = poti[i+1] - poti[i]; 1175 ptJ = potj + poti[i]; 1176 for (j=0; j<pnz; j++) { 1177 row = ptJ[j]; /* row of AP == col of Pot */ 1178 apnz = api[row+1] - api[row]; 1179 Jptr = apj + api[row]; 1180 /* add non-zero cols of AP into the sorted linked list lnk */ 1181 ierr = PetscLLCondensedAddSorted(apnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 1182 } 1183 nnz = lnk[0]; 1184 1185 /* If free space is not available, double the total space in the list */ 1186 if (current_space->local_remaining<nnz) { 1187 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(nnz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 1188 nspacedouble++; 1189 } 1190 1191 /* Copy data into free space, and zero out denserows */ 1192 ierr = PetscLLCondensedClean(pN,nnz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 1193 1194 current_space->array += nnz; 1195 current_space->local_used += nnz; 1196 current_space->local_remaining -= nnz; 1197 1198 coi[i+1] = coi[i] + nnz; 1199 } 1200 1201 ierr = PetscMalloc1(coi[pon],&coj);CHKERRQ(ierr); 1202 ierr = PetscFreeSpaceContiguous(&free_space,coj);CHKERRQ(ierr); 1203 afill_tmp = (PetscReal)coi[pon]/(poti[pon] + api[am]+1); 1204 if (afill_tmp > afill) afill = afill_tmp; 1205 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 1206 1207 /* (3) send j-array (coj) of Co to other processors */ 1208 /*--------------------------------------------------*/ 1209 ierr = PetscCalloc1(size,&merge->len_s);CHKERRQ(ierr); 1210 len_s = merge->len_s; 1211 merge->nsend = 0; 1212 1213 1214 /* determine row ownership */ 1215 ierr = PetscLayoutCreate(comm,&merge->rowmap);CHKERRQ(ierr); 1216 merge->rowmap->n = pn; 1217 merge->rowmap->bs = 1; 1218 1219 ierr = PetscLayoutSetUp(merge->rowmap);CHKERRQ(ierr); 1220 owners = merge->rowmap->range; 1221 1222 /* determine the number of messages to send, their lengths */ 1223 ierr = PetscMalloc2(size,&len_si,size,&sstatus);CHKERRQ(ierr); 1224 ierr = PetscMemzero(len_si,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 1225 ierr = PetscMalloc1(size+2,&owners_co);CHKERRQ(ierr); 1226 1227 proc = 0; 1228 for (i=0; i<pon; i++) { 1229 while (prmap[i] >= owners[proc+1]) proc++; 1230 len_si[proc]++; /* num of rows in Co(=Pt*AP) to be sent to [proc] */ 1231 len_s[proc] += coi[i+1] - coi[i]; /* num of nonzeros in Co to be sent to [proc] */ 1232 } 1233 1234 len = 0; /* max length of buf_si[], see (4) */ 1235 owners_co[0] = 0; 1236 for (proc=0; proc<size; proc++) { 1237 owners_co[proc+1] = owners_co[proc] + len_si[proc]; 1238 if (len_s[proc]) { 1239 merge->nsend++; 1240 len_si[proc] = 2*(len_si[proc] + 1); /* length of buf_si to be sent to [proc] */ 1241 len += len_si[proc]; 1242 } 1243 } 1244 1245 /* determine the number and length of messages to receive for coi and coj */ 1246 ierr = PetscGatherNumberOfMessages(comm,NULL,len_s,&merge->nrecv);CHKERRQ(ierr); 1247 ierr = PetscGatherMessageLengths2(comm,merge->nsend,merge->nrecv,len_s,len_si,&merge->id_r,&merge->len_r,&len_ri);CHKERRQ(ierr); 1248 1249 /* post the Irecv and Isend of coj */ 1250 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 1251 ierr = PetscPostIrecvInt(comm,tagj,merge->nrecv,merge->id_r,merge->len_r,&buf_rj,&rwaits);CHKERRQ(ierr); 1252 ierr = PetscMalloc1(merge->nsend+1,&swaits);CHKERRQ(ierr); 1253 for (proc=0, k=0; proc<size; proc++) { 1254 if (!len_s[proc]) continue; 1255 i = owners_co[proc]; 1256 ierr = MPI_Isend(coj+coi[i],len_s[proc],MPIU_INT,proc,tagj,comm,swaits+k);CHKERRQ(ierr); 1257 k++; 1258 } 1259 1260 /* receives and sends of coj are complete */ 1261 for (i=0; i<merge->nrecv; i++) { 1262 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 1263 } 1264 ierr = PetscFree(rwaits);CHKERRQ(ierr); 1265 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 1266 1267 /* (4) send and recv coi */ 1268 /*-----------------------*/ 1269 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 1270 ierr = PetscPostIrecvInt(comm,tagi,merge->nrecv,merge->id_r,len_ri,&buf_ri,&rwaits);CHKERRQ(ierr); 1271 ierr = PetscMalloc1(len+1,&buf_s);CHKERRQ(ierr); 1272 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 1273 for (proc=0,k=0; proc<size; proc++) { 1274 if (!len_s[proc]) continue; 1275 /* form outgoing message for i-structure: 1276 buf_si[0]: nrows to be sent 1277 [1:nrows]: row index (global) 1278 [nrows+1:2*nrows+1]: i-structure index 1279 */ 1280 /*-------------------------------------------*/ 1281 nrows = len_si[proc]/2 - 1; /* num of rows in Co to be sent to [proc] */ 1282 buf_si_i = buf_si + nrows+1; 1283 buf_si[0] = nrows; 1284 buf_si_i[0] = 0; 1285 nrows = 0; 1286 for (i=owners_co[proc]; i<owners_co[proc+1]; i++) { 1287 nzi = coi[i+1] - coi[i]; 1288 buf_si_i[nrows+1] = buf_si_i[nrows] + nzi; /* i-structure */ 1289 buf_si[nrows+1] = prmap[i] -owners[proc]; /* local row index */ 1290 nrows++; 1291 } 1292 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,swaits+k);CHKERRQ(ierr); 1293 k++; 1294 buf_si += len_si[proc]; 1295 } 1296 i = merge->nrecv; 1297 while (i--) { 1298 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 1299 } 1300 ierr = PetscFree(rwaits);CHKERRQ(ierr); 1301 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 1302 1303 ierr = PetscFree2(len_si,sstatus);CHKERRQ(ierr); 1304 ierr = PetscFree(len_ri);CHKERRQ(ierr); 1305 ierr = PetscFree(swaits);CHKERRQ(ierr); 1306 ierr = PetscFree(buf_s);CHKERRQ(ierr); 1307 1308 /* (5) compute the local portion of C (mpi mat) */ 1309 /*----------------------------------------------*/ 1310 ierr = MatGetSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 1311 1312 /* allocate pti array and free space for accumulating nonzero column info */ 1313 ierr = PetscMalloc1(pn+1,&pti);CHKERRQ(ierr); 1314 pti[0] = 0; 1315 1316 /* set initial free space to be fill*(nnz(P) + nnz(AP)) */ 1317 nnz = PetscRealIntMultTruncate(fill,PetscIntSumTruncate(pi_loc[pm],api[am])); 1318 ierr = PetscFreeSpaceGet(nnz,&free_space);CHKERRQ(ierr); 1319 current_space = free_space; 1320 1321 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextci);CHKERRQ(ierr); 1322 for (k=0; k<merge->nrecv; k++) { 1323 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 1324 nrows = *buf_ri_k[k]; 1325 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 1326 nextci[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 1327 } 1328 ierr = MatPreallocateInitialize(comm,pn,pn,dnz,onz);CHKERRQ(ierr); 1329 1330 for (i=0; i<pn; i++) { 1331 /* add pdt[i,:]*AP into lnk */ 1332 pnz = pdti[i+1] - pdti[i]; 1333 ptJ = pdtj + pdti[i]; 1334 for (j=0; j<pnz; j++) { 1335 row = ptJ[j]; /* row of AP == col of Pt */ 1336 apnz = api[row+1] - api[row]; 1337 Jptr = apj + api[row]; 1338 /* add non-zero cols of AP into the sorted linked list lnk */ 1339 ierr = PetscLLCondensedAddSorted(apnz,Jptr,lnk,lnkbt);CHKERRQ(ierr); 1340 } 1341 1342 /* add received col data into lnk */ 1343 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 1344 if (i == *nextrow[k]) { /* i-th row */ 1345 nzi = *(nextci[k]+1) - *nextci[k]; 1346 Jptr = buf_rj[k] + *nextci[k]; 1347 ierr = PetscLLCondensedAddSorted(nzi,Jptr,lnk,lnkbt);CHKERRQ(ierr); 1348 nextrow[k]++; nextci[k]++; 1349 } 1350 } 1351 nnz = lnk[0]; 1352 1353 /* if free space is not available, make more free space */ 1354 if (current_space->local_remaining<nnz) { 1355 ierr = PetscFreeSpaceGet(PetscIntSumTruncate(nnz,current_space->total_array_size),¤t_space);CHKERRQ(ierr); 1356 nspacedouble++; 1357 } 1358 /* copy data into free space, then initialize lnk */ 1359 ierr = PetscLLCondensedClean(pN,nnz,current_space->array,lnk,lnkbt);CHKERRQ(ierr); 1360 ierr = MatPreallocateSet(i+owners[rank],nnz,current_space->array,dnz,onz);CHKERRQ(ierr); 1361 1362 current_space->array += nnz; 1363 current_space->local_used += nnz; 1364 current_space->local_remaining -= nnz; 1365 1366 pti[i+1] = pti[i] + nnz; 1367 } 1368 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 1369 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 1370 1371 ierr = PetscMalloc1(pti[pn]+1,&ptj);CHKERRQ(ierr); 1372 ierr = PetscFreeSpaceContiguous(&free_space,ptj);CHKERRQ(ierr); 1373 afill_tmp = (PetscReal)pti[pn]/(pi_loc[pm] + api[am]+1); 1374 if (afill_tmp > afill) afill = afill_tmp; 1375 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 1376 1377 /* (6) create symbolic parallel matrix Cmpi */ 1378 /*------------------------------------------*/ 1379 ierr = MatCreate(comm,&Cmpi);CHKERRQ(ierr); 1380 ierr = MatSetSizes(Cmpi,pn,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 1381 ierr = MatSetBlockSizes(Cmpi,PetscAbs(P->cmap->bs),PetscAbs(P->cmap->bs));CHKERRQ(ierr); 1382 ierr = MatSetType(Cmpi,MATMPIAIJ);CHKERRQ(ierr); 1383 ierr = MatMPIAIJSetPreallocation(Cmpi,0,dnz,0,onz);CHKERRQ(ierr); 1384 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 1385 1386 merge->bi = pti; /* Cseq->i */ 1387 merge->bj = ptj; /* Cseq->j */ 1388 merge->coi = coi; /* Co->i */ 1389 merge->coj = coj; /* Co->j */ 1390 merge->buf_ri = buf_ri; 1391 merge->buf_rj = buf_rj; 1392 merge->owners_co = owners_co; 1393 merge->destroy = Cmpi->ops->destroy; 1394 1395 /* attach the supporting struct to Cmpi for reuse */ 1396 c = (Mat_MPIAIJ*)Cmpi->data; 1397 c->ptap = ptap; 1398 ptap->api = api; 1399 ptap->apj = apj; 1400 ptap->duplicate = Cmpi->ops->duplicate; 1401 ptap->destroy = Cmpi->ops->destroy; 1402 1403 /* Cmpi is not ready for use - assembly will be done by MatPtAPNumeric() */ 1404 Cmpi->assembled = PETSC_FALSE; 1405 Cmpi->ops->destroy = MatDestroy_MPIAIJ_PtAP; 1406 Cmpi->ops->duplicate = MatDuplicate_MPIAIJ_MatPtAP; 1407 Cmpi->ops->ptapnumeric = MatPtAPNumeric_MPIAIJ_MPIAIJ_ptap; 1408 *C = Cmpi; 1409 1410 /* flag 'scalable' determines which implementations to be used: 1411 0: do dense axpy in MatPtAPNumeric() - fast, but requires storage of a nonscalable dense array apa; 1412 1: do sparse axpy in MatPtAPNumeric() - might slow, uses a sparse array apa */ 1413 /* set default scalable */ 1414 ptap->scalable = PETSC_FALSE; /* PETSC_TRUE; */ 1415 1416 ierr = PetscOptionsGetBool(((PetscObject)Cmpi)->options,((PetscObject)Cmpi)->prefix,"-matptap_scalable",&ptap->scalable,NULL);CHKERRQ(ierr); 1417 if (!ptap->scalable) { /* Do dense axpy */ 1418 ierr = PetscCalloc1(pN,&ptap->apa);CHKERRQ(ierr); 1419 } else { 1420 ierr = PetscCalloc1(ap_rmax+1,&ptap->apa);CHKERRQ(ierr); 1421 } 1422 1423 #if defined(PETSC_USE_INFO) 1424 if (pti[pn] != 0) { 1425 ierr = PetscInfo3(Cmpi,"Reallocs %D; Fill ratio: given %g needed %g.\n",nspacedouble,(double)fill,(double)afill);CHKERRQ(ierr); 1426 ierr = PetscInfo1(Cmpi,"Use MatPtAP(A,P,MatReuse,%g,&C) for best performance.\n",(double)afill);CHKERRQ(ierr); 1427 } else { 1428 ierr = PetscInfo(Cmpi,"Empty matrix product\n");CHKERRQ(ierr); 1429 } 1430 #endif 1431 PetscFunctionReturn(0); 1432 } 1433 1434 PetscErrorCode MatPtAPNumeric_MPIAIJ_MPIAIJ_ptap(Mat A,Mat P,Mat C) 1435 { 1436 PetscErrorCode ierr; 1437 Mat_MPIAIJ *a =(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c=(Mat_MPIAIJ*)C->data; 1438 Mat_SeqAIJ *ad=(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 1439 Mat_SeqAIJ *pd=(Mat_SeqAIJ*)(p->A)->data,*po=(Mat_SeqAIJ*)(p->B)->data; 1440 Mat_SeqAIJ *p_loc,*p_oth; 1441 Mat_PtAPMPI *ptap; 1442 Mat_Merge_SeqsToMPI *merge; 1443 PetscInt *adi=ad->i,*aoi=ao->i,*adj,*aoj,*apJ,nextp; 1444 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*pJ,*pj; 1445 PetscInt i,j,k,anz,pnz,apnz,nextap,row,*cj; 1446 MatScalar *ada,*aoa,*apa,*pa,*ca,*pa_loc,*pa_oth,valtmp; 1447 PetscInt am =A->rmap->n,cm=C->rmap->n,pon=(p->B)->cmap->n; 1448 MPI_Comm comm; 1449 PetscMPIInt size,rank,taga,*len_s; 1450 PetscInt *owners,proc,nrows,**buf_ri_k,**nextrow,**nextci; 1451 PetscInt **buf_ri,**buf_rj; 1452 PetscInt cnz=0,*bj_i,*bi,*bj,bnz,nextcj; /* bi,bj,ba: local array of C(mpi mat) */ 1453 MPI_Request *s_waits,*r_waits; 1454 MPI_Status *status; 1455 MatScalar **abuf_r,*ba_i,*pA,*coa,*ba; 1456 PetscInt *api,*apj,*coi,*coj; 1457 PetscInt *poJ=po->j,*pdJ=pd->j,pcstart=P->cmap->rstart,pcend=P->cmap->rend; 1458 PetscBool scalable; 1459 #if defined(PTAP_PROFILE) 1460 PetscLogDouble t0,t1,t2,eP,t3,t4,et2_AP=0.0,ePtAP=0.0,t2_0,t2_1,t2_2; 1461 #endif 1462 1463 PetscFunctionBegin; 1464 ierr = PetscObjectGetComm((PetscObject)C,&comm);CHKERRQ(ierr); 1465 #if defined(PTAP_PROFILE) 1466 ierr = PetscTime(&t0);CHKERRQ(ierr); 1467 #endif 1468 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 1469 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 1470 1471 ptap = c->ptap; 1472 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"); 1473 merge = ptap->merge; 1474 apa = ptap->apa; 1475 scalable = ptap->scalable; 1476 1477 /* 1) get P_oth = ptap->P_oth and P_loc = ptap->P_loc */ 1478 /*-----------------------------------------------------*/ 1479 if (ptap->reuse == MAT_INITIAL_MATRIX) { 1480 /* P_oth and P_loc are obtained in MatPtASymbolic(), skip calling MatGetBrowsOfAoCols() and MatMPIAIJGetLocalMat() */ 1481 ptap->reuse = MAT_REUSE_MATRIX; 1482 } else { /* update numerical values of P_oth and P_loc */ 1483 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_REUSE_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 1484 ierr = MatMPIAIJGetLocalMat(P,MAT_REUSE_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 1485 } 1486 #if defined(PTAP_PROFILE) 1487 ierr = PetscTime(&t1);CHKERRQ(ierr); 1488 eP = t1-t0; 1489 #endif 1490 /* 1491 printf("[%d] Ad: %d, %d; Ao: %d, %d; P_loc: %d, %d; P_oth %d, %d;\n",rank, 1492 a->A->rmap->N,a->A->cmap->N,a->B->rmap->N,a->B->cmap->N, 1493 ptap->P_loc->rmap->N,ptap->P_loc->cmap->N, 1494 ptap->P_oth->rmap->N,ptap->P_oth->cmap->N); 1495 */ 1496 1497 /* 2) compute numeric C_seq = P_loc^T*A_loc*P - dominating part */ 1498 /*--------------------------------------------------------------*/ 1499 /* get data from symbolic products */ 1500 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 1501 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 1502 pi_loc=p_loc->i; pj_loc=p_loc->j; pa_loc=p_loc->a; 1503 pi_oth=p_oth->i; pj_oth=p_oth->j; pa_oth=p_oth->a; 1504 1505 coi = merge->coi; coj = merge->coj; 1506 ierr = PetscCalloc1(coi[pon]+1,&coa);CHKERRQ(ierr); 1507 1508 bi = merge->bi; bj = merge->bj; 1509 owners = merge->rowmap->range; 1510 ierr = PetscCalloc1(bi[cm]+1,&ba);CHKERRQ(ierr); /* ba: Cseq->a */ 1511 1512 api = ptap->api; apj = ptap->apj; 1513 1514 if (!scalable) { /* Do dense axpy on apa (length of pN, stores A[i,:]*P) - nonscalable, but faster (could take 1/3 scalable time) */ 1515 ierr = PetscInfo(C,"Using non-scalable dense axpy\n");CHKERRQ(ierr); 1516 #if defined(PTAP_PROFILE) 1517 ierr = PetscTime(&t1);CHKERRQ(ierr); 1518 #endif 1519 for (i=0; i<am; i++) { 1520 #if defined(PTAP_PROFILE) 1521 ierr = PetscTime(&t2_0);CHKERRQ(ierr); 1522 #endif 1523 /* 2-a) form i-th sparse row of A_loc*P = Ad*P_loc + Ao*P_oth */ 1524 /*------------------------------------------------------------*/ 1525 apJ = apj + api[i]; 1526 1527 /* diagonal portion of A */ 1528 anz = adi[i+1] - adi[i]; 1529 adj = ad->j + adi[i]; 1530 ada = ad->a + adi[i]; 1531 for (j=0; j<anz; j++) { 1532 row = adj[j]; 1533 pnz = pi_loc[row+1] - pi_loc[row]; 1534 pj = pj_loc + pi_loc[row]; 1535 pa = pa_loc + pi_loc[row]; 1536 1537 /* perform dense axpy */ 1538 valtmp = ada[j]; 1539 for (k=0; k<pnz; k++) { 1540 apa[pj[k]] += valtmp*pa[k]; 1541 } 1542 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 1543 } 1544 1545 /* off-diagonal portion of A */ 1546 anz = aoi[i+1] - aoi[i]; 1547 aoj = ao->j + aoi[i]; 1548 aoa = ao->a + aoi[i]; 1549 for (j=0; j<anz; j++) { 1550 row = aoj[j]; 1551 pnz = pi_oth[row+1] - pi_oth[row]; 1552 pj = pj_oth + pi_oth[row]; 1553 pa = pa_oth + pi_oth[row]; 1554 1555 /* perform dense axpy */ 1556 valtmp = aoa[j]; 1557 for (k=0; k<pnz; k++) { 1558 apa[pj[k]] += valtmp*pa[k]; 1559 } 1560 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 1561 } 1562 #if defined(PTAP_PROFILE) 1563 ierr = PetscTime(&t2_1);CHKERRQ(ierr); 1564 et2_AP += t2_1 - t2_0; 1565 #endif 1566 1567 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 1568 /*--------------------------------------------------------------*/ 1569 apnz = api[i+1] - api[i]; 1570 /* put the value into Co=(p->B)^T*AP (off-diagonal part, send to others) */ 1571 pnz = po->i[i+1] - po->i[i]; 1572 poJ = po->j + po->i[i]; 1573 pA = po->a + po->i[i]; 1574 for (j=0; j<pnz; j++) { 1575 row = poJ[j]; 1576 cnz = coi[row+1] - coi[row]; 1577 cj = coj + coi[row]; 1578 ca = coa + coi[row]; 1579 /* perform dense axpy */ 1580 valtmp = pA[j]; 1581 for (k=0; k<cnz; k++) { 1582 ca[k] += valtmp*apa[cj[k]]; 1583 } 1584 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 1585 } 1586 /* put the value into Cd (diagonal part) */ 1587 pnz = pd->i[i+1] - pd->i[i]; 1588 pdJ = pd->j + pd->i[i]; 1589 pA = pd->a + pd->i[i]; 1590 for (j=0; j<pnz; j++) { 1591 row = pdJ[j]; 1592 cnz = bi[row+1] - bi[row]; 1593 cj = bj + bi[row]; 1594 ca = ba + bi[row]; 1595 /* perform dense axpy */ 1596 valtmp = pA[j]; 1597 for (k=0; k<cnz; k++) { 1598 ca[k] += valtmp*apa[cj[k]]; 1599 } 1600 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 1601 } 1602 1603 /* zero the current row of A*P */ 1604 for (k=0; k<apnz; k++) apa[apJ[k]] = 0.0; 1605 #if defined(PTAP_PROFILE) 1606 ierr = PetscTime(&t2_2);CHKERRQ(ierr); 1607 ePtAP += t2_2 - t2_1; 1608 #endif 1609 } 1610 } else { /* Do sparse axpy on apa (length of ap_rmax, stores A[i,:]*P) - scalable, but slower */ 1611 ierr = PetscInfo(C,"Using scalable sparse axpy\n");CHKERRQ(ierr); 1612 /*-----------------------------------------------------------------------------------------*/ 1613 pA=pa_loc; 1614 for (i=0; i<am; i++) { 1615 #if defined(PTAP_PROFILE) 1616 ierr = PetscTime(&t2_0);CHKERRQ(ierr); 1617 #endif 1618 /* form i-th sparse row of A*P */ 1619 apnz = api[i+1] - api[i]; 1620 apJ = apj + api[i]; 1621 /* diagonal portion of A */ 1622 anz = adi[i+1] - adi[i]; 1623 adj = ad->j + adi[i]; 1624 ada = ad->a + adi[i]; 1625 for (j=0; j<anz; j++) { 1626 row = adj[j]; 1627 pnz = pi_loc[row+1] - pi_loc[row]; 1628 pj = pj_loc + pi_loc[row]; 1629 pa = pa_loc + pi_loc[row]; 1630 valtmp = ada[j]; 1631 nextp = 0; 1632 for (k=0; nextp<pnz; k++) { 1633 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 1634 apa[k] += valtmp*pa[nextp++]; 1635 } 1636 } 1637 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 1638 } 1639 /* off-diagonal portion of A */ 1640 anz = aoi[i+1] - aoi[i]; 1641 aoj = ao->j + aoi[i]; 1642 aoa = ao->a + aoi[i]; 1643 for (j=0; j<anz; j++) { 1644 row = aoj[j]; 1645 pnz = pi_oth[row+1] - pi_oth[row]; 1646 pj = pj_oth + pi_oth[row]; 1647 pa = pa_oth + pi_oth[row]; 1648 valtmp = aoa[j]; 1649 nextp = 0; 1650 for (k=0; nextp<pnz; k++) { 1651 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 1652 apa[k] += valtmp*pa[nextp++]; 1653 } 1654 } 1655 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 1656 } 1657 #if defined(PTAP_PROFILE) 1658 ierr = PetscTime(&t2_1);CHKERRQ(ierr); 1659 et2_AP += t2_1 - t2_0; 1660 #endif 1661 1662 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 1663 /*--------------------------------------------------------------*/ 1664 pnz = pi_loc[i+1] - pi_loc[i]; 1665 pJ = pj_loc + pi_loc[i]; 1666 for (j=0; j<pnz; j++) { 1667 nextap = 0; 1668 row = pJ[j]; /* global index */ 1669 if (row < pcstart || row >=pcend) { /* put the value into Co */ 1670 row = *poJ; 1671 cj = coj + coi[row]; 1672 ca = coa + coi[row]; poJ++; 1673 } else { /* put the value into Cd */ 1674 row = *pdJ; 1675 cj = bj + bi[row]; 1676 ca = ba + bi[row]; pdJ++; 1677 } 1678 valtmp = pA[j]; 1679 for (k=0; nextap<apnz; k++) { 1680 if (cj[k]==apJ[nextap]) ca[k] += valtmp*apa[nextap++]; 1681 } 1682 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 1683 } 1684 pA += pnz; 1685 /* zero the current row info for A*P */ 1686 ierr = PetscMemzero(apa,apnz*sizeof(MatScalar));CHKERRQ(ierr); 1687 #if defined(PTAP_PROFILE) 1688 ierr = PetscTime(&t2_2);CHKERRQ(ierr); 1689 ePtAP += t2_2 - t2_1; 1690 #endif 1691 } 1692 } 1693 #if defined(PTAP_PROFILE) 1694 ierr = PetscTime(&t2);CHKERRQ(ierr); 1695 #endif 1696 1697 /* 3) send and recv matrix values coa */ 1698 /*------------------------------------*/ 1699 buf_ri = merge->buf_ri; 1700 buf_rj = merge->buf_rj; 1701 len_s = merge->len_s; 1702 ierr = PetscCommGetNewTag(comm,&taga);CHKERRQ(ierr); 1703 ierr = PetscPostIrecvScalar(comm,taga,merge->nrecv,merge->id_r,merge->len_r,&abuf_r,&r_waits);CHKERRQ(ierr); 1704 1705 ierr = PetscMalloc2(merge->nsend+1,&s_waits,size,&status);CHKERRQ(ierr); 1706 for (proc=0,k=0; proc<size; proc++) { 1707 if (!len_s[proc]) continue; 1708 i = merge->owners_co[proc]; 1709 ierr = MPI_Isend(coa+coi[i],len_s[proc],MPIU_MATSCALAR,proc,taga,comm,s_waits+k);CHKERRQ(ierr); 1710 k++; 1711 } 1712 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,r_waits,status);CHKERRQ(ierr);} 1713 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,s_waits,status);CHKERRQ(ierr);} 1714 1715 ierr = PetscFree2(s_waits,status);CHKERRQ(ierr); 1716 ierr = PetscFree(r_waits);CHKERRQ(ierr); 1717 ierr = PetscFree(coa);CHKERRQ(ierr); 1718 #if defined(PTAP_PROFILE) 1719 ierr = PetscTime(&t3);CHKERRQ(ierr); 1720 #endif 1721 1722 /* 4) insert local Cseq and received values into Cmpi */ 1723 /*------------------------------------------------------*/ 1724 ierr = PetscMalloc3(merge->nrecv,&buf_ri_k,merge->nrecv,&nextrow,merge->nrecv,&nextci);CHKERRQ(ierr); 1725 for (k=0; k<merge->nrecv; k++) { 1726 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 1727 nrows = *(buf_ri_k[k]); 1728 nextrow[k] = buf_ri_k[k]+1; /* next row number of k-th recved i-structure */ 1729 nextci[k] = buf_ri_k[k] + (nrows + 1); /* poins to the next i-structure of k-th recved i-structure */ 1730 } 1731 1732 for (i=0; i<cm; i++) { 1733 row = owners[rank] + i; /* global row index of C_seq */ 1734 bj_i = bj + bi[i]; /* col indices of the i-th row of C */ 1735 ba_i = ba + bi[i]; 1736 bnz = bi[i+1] - bi[i]; 1737 /* add received vals into ba */ 1738 for (k=0; k<merge->nrecv; k++) { /* k-th received message */ 1739 /* i-th row */ 1740 if (i == *nextrow[k]) { 1741 cnz = *(nextci[k]+1) - *nextci[k]; 1742 cj = buf_rj[k] + *(nextci[k]); 1743 ca = abuf_r[k] + *(nextci[k]); 1744 nextcj = 0; 1745 for (j=0; nextcj<cnz; j++) { 1746 if (bj_i[j] == cj[nextcj]) { /* bcol == ccol */ 1747 ba_i[j] += ca[nextcj++]; 1748 } 1749 } 1750 nextrow[k]++; nextci[k]++; 1751 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 1752 } 1753 } 1754 ierr = MatSetValues(C,1,&row,bnz,bj_i,ba_i,INSERT_VALUES);CHKERRQ(ierr); 1755 } 1756 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1757 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1758 1759 ierr = PetscFree(ba);CHKERRQ(ierr); 1760 ierr = PetscFree(abuf_r[0]);CHKERRQ(ierr); 1761 ierr = PetscFree(abuf_r);CHKERRQ(ierr); 1762 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 1763 #if defined(PTAP_PROFILE) 1764 ierr = PetscTime(&t4);CHKERRQ(ierr); 1765 if (rank==1) { 1766 ierr = PetscPrintf(MPI_COMM_SELF," [%d] PtAPNum %g/P + %g/PtAP( %g/A*P + %g/Pt*AP ) + %g/comm + %g/Cloc = %g\n\n",rank,eP,t2-t1,et2_AP,ePtAP,t3-t2,t4-t3,t4-t0);CHKERRQ(ierr); 1767 } 1768 #endif 1769 PetscFunctionReturn(0); 1770 } 1771