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 12 /* 13 #define DEBUG_MATPTAP 14 */ 15 16 extern PetscErrorCode MatDestroy_MPIAIJ(Mat); 17 #undef __FUNCT__ 18 #define __FUNCT__ "MatDestroy_MPIAIJ_PtAP" 19 PetscErrorCode MatDestroy_MPIAIJ_PtAP(Mat A) 20 { 21 PetscErrorCode ierr; 22 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 23 Mat_PtAPMPI *ptap=a->ptap; 24 25 PetscFunctionBegin; 26 if (ptap){ 27 Mat_Merge_SeqsToMPI *merge=ptap->merge; 28 ierr = PetscFree2(ptap->startsj_s,ptap->startsj_r);CHKERRQ(ierr); 29 ierr = PetscFree(ptap->bufa);CHKERRQ(ierr); 30 ierr = MatDestroy(&ptap->P_loc);CHKERRQ(ierr); 31 ierr = MatDestroy(&ptap->P_oth);CHKERRQ(ierr); 32 ierr = MatDestroy(&ptap->A_loc);CHKERRQ(ierr); /* used by MatTransposeMatMult() */ 33 if (ptap->api){ierr = PetscFree(ptap->api);CHKERRQ(ierr);} 34 if (ptap->apj){ierr = PetscFree(ptap->apj);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 55 PetscFunctionReturn(0); 56 } 57 58 #undef __FUNCT__ 59 #define __FUNCT__ "MatDuplicate_MPIAIJ_MatPtAP" 60 PetscErrorCode MatDuplicate_MPIAIJ_MatPtAP(Mat A, MatDuplicateOption op, Mat *M) 61 { 62 PetscErrorCode ierr; 63 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data; 64 Mat_PtAPMPI *ptap = a->ptap; 65 Mat_Merge_SeqsToMPI *merge = ptap->merge; 66 67 PetscFunctionBegin; 68 ierr = (*merge->duplicate)(A,op,M);CHKERRQ(ierr); 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=PETSC_NULL,current_space=PETSC_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 nlnk,*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=((PetscObject)A)->comm; 109 PetscMPIInt size,rank,tagi,tagj,*len_si,*len_s,*len_ri; 110 PetscInt **buf_rj,**buf_ri,**buf_ri_k; 111 PetscInt len,proc,*dnz,*onz,*owners; 112 PetscInt nzi,*bi,*bj; 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 rstart = P->cmap->rstart,rmax; 120 PetscScalar *vals; 121 122 PetscFunctionBegin; 123 /* check if matrix local sizes are compatible */ 124 if (A->rmap->rstart != P->rmap->rstart || A->rmap->rend != P->rmap->rend){ 125 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); 126 } 127 if (A->cmap->rstart != P->rmap->rstart || A->cmap->rend != P->rmap->rend){ 128 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); 129 } 130 131 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 132 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 133 134 /* create struct Mat_PtAPMPI and attached it to C later */ 135 ierr = PetscNew(Mat_PtAPMPI,&ptap);CHKERRQ(ierr); 136 ptap->reuse = MAT_INITIAL_MATRIX; 137 138 139 /* get P_oth by taking rows of P (= non-zero cols of local A) from other processors */ 140 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_INITIAL_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 141 #if defined(DEBUG_MATPTAP) 142 if (!rank) ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] P_oth is done...\n",rank); 143 #endif 144 145 /* get P_loc by taking all local rows of P */ 146 ierr = MatMPIAIJGetLocalMat(P,MAT_INITIAL_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 147 #if defined(DEBUG_MATPTAP) 148 ierr = MPI_Barrier(comm);CHKERRQ(ierr); 149 if (!rank) ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] P_loc is done...\n",rank); 150 #endif 151 152 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 153 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 154 pi_loc = p_loc->i; pj_loc = p_loc->j; 155 pi_oth = p_oth->i; pj_oth = p_oth->j; 156 157 /* first, compute symbolic AP = A_loc*P = A_diag*P_loc + A_off*P_oth */ 158 /*-------------------------------------------------------------------*/ 159 ierr = PetscMalloc((am+1)*sizeof(PetscInt),&api);CHKERRQ(ierr); 160 api[0] = 0; 161 162 /* create and initialize a linked list */ 163 nlnk = pN+1; 164 #if defined(DEBUG_MATPTAP) 165 ierr = MPI_Barrier(comm);CHKERRQ(ierr); 166 if (!rank) ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] call PetscLLCreate(), pN %D, P_rmax %D %D; Annz %D\n",rank,pN,p_loc->rmax,p_oth->rmax,adi[am]+aoi[am]); 167 #endif 168 ierr = PetscLLCreate(pN,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 169 #if defined(DEBUG_MATPTAP) 170 ierr = MPI_Barrier(comm);CHKERRQ(ierr); 171 if (!rank) ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] call PetscLLCreate() is done\n",rank); 172 #endif 173 174 /* Initial FreeSpace size is fill*(nnz(A) + nnz(P)) */ 175 ierr = PetscFreeSpaceGet((PetscInt)(fill*(adi[am]+aoi[am]+pi_loc[pm])),&free_space);CHKERRQ(ierr); 176 current_space = free_space; 177 178 for (i=0; i<am; i++) { 179 apnz = 0; 180 /* diagonal portion of A */ 181 nzi = adi[i+1] - adi[i]; 182 aj = ad->j + adi[i]; 183 for (j=0; j<nzi; j++){ 184 row = aj[j]; 185 pnz = pi_loc[row+1] - pi_loc[row]; 186 Jptr = pj_loc + pi_loc[row]; 187 /* add non-zero cols of P into the sorted linked list lnk */ 188 ierr = PetscLLAddSorted(pnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 189 apnz += nlnk; 190 } 191 /* off-diagonal portion of A */ 192 nzi = aoi[i+1] - aoi[i]; 193 aj = ao->j + aoi[i]; 194 for (j=0; j<nzi; j++){ 195 row = aj[j]; 196 pnz = pi_oth[row+1] - pi_oth[row]; 197 Jptr = pj_oth + pi_oth[row]; 198 ierr = PetscLLAddSorted(pnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 199 apnz += nlnk; 200 } 201 202 api[i+1] = api[i] + apnz; 203 if (ap_rmax < apnz) ap_rmax = apnz; 204 205 /* if free space is not available, double the total space in the list */ 206 if (current_space->local_remaining<apnz) { 207 ierr = PetscFreeSpaceGet(apnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 208 nspacedouble++; 209 } 210 211 /* Copy data into free space, then initialize lnk */ 212 ierr = PetscLLClean(pN,pN,apnz,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 213 current_space->array += apnz; 214 current_space->local_used += apnz; 215 current_space->local_remaining -= apnz; 216 } 217 /* Allocate space for apj, initialize apj, and */ 218 /* destroy list of free space and other temporary array(s) */ 219 ierr = PetscMalloc((api[am]+1)*sizeof(PetscInt),&apj);CHKERRQ(ierr); 220 ierr = PetscFreeSpaceContiguous(&free_space,apj);CHKERRQ(ierr); 221 afill_tmp = (PetscReal)api[am]/(adi[am]+aoi[am]+pi_loc[pm]+1); 222 if (afill_tmp > afill) afill = afill_tmp; 223 224 /* determine symbolic Co=(p->B)^T*AP - send to others */ 225 /*----------------------------------------------------*/ 226 ierr = MatGetSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 227 228 /* then, compute symbolic Co = (p->B)^T*AP */ 229 pon = (p->B)->cmap->n; /* total num of rows to be sent to other processors 230 >= (num of nonzero rows of C_seq) - pn */ 231 ierr = PetscMalloc((pon+1)*sizeof(PetscInt),&coi);CHKERRQ(ierr); 232 coi[0] = 0; 233 234 /* set initial free space to be fill*(nnz(p->B) + nnz(AP)) */ 235 nnz = fill*(poti[pon] + api[am]); 236 ierr = PetscFreeSpaceGet(nnz,&free_space); 237 current_space = free_space; 238 239 for (i=0; i<pon; i++) { 240 nnz = 0; 241 pnz = poti[i+1] - poti[i]; 242 ptJ = potj + poti[i]; 243 for (j=0; j<pnz; j++){ 244 row = ptJ[j]; /* row of AP == col of Pot */ 245 apnz = api[row+1] - api[row]; 246 Jptr = apj + api[row]; 247 /* add non-zero cols of AP into the sorted linked list lnk */ 248 ierr = PetscLLAddSorted(apnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 249 nnz += nlnk; 250 } 251 252 /* If free space is not available, double the total space in the list */ 253 if (current_space->local_remaining<nnz) { 254 ierr = PetscFreeSpaceGet(nnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 255 nspacedouble++; 256 } 257 258 /* Copy data into free space, and zero out denserows */ 259 ierr = PetscLLClean(pN,pN,nnz,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 260 current_space->array += nnz; 261 current_space->local_used += nnz; 262 current_space->local_remaining -= nnz; 263 coi[i+1] = coi[i] + nnz; 264 } 265 ierr = PetscMalloc((coi[pon]+1)*sizeof(PetscInt),&coj);CHKERRQ(ierr); 266 ierr = PetscFreeSpaceContiguous(&free_space,coj);CHKERRQ(ierr); 267 afill_tmp = (PetscReal)coi[pon]/(poti[pon] + api[am]+1); 268 if (afill_tmp > afill) afill = afill_tmp; 269 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->B,&poti,&potj);CHKERRQ(ierr); 270 271 #if defined(DEBUG_MATPTAP) 272 ierr = MPI_Barrier(comm);CHKERRQ(ierr); 273 if (!rank) ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] local Co is done\n",rank); 274 #endif 275 276 /* send j-array (coj) of Co to other processors */ 277 /*----------------------------------------------*/ 278 /* determine row ownership */ 279 ierr = PetscNew(Mat_Merge_SeqsToMPI,&merge);CHKERRQ(ierr); 280 ierr = PetscLayoutCreate(comm,&merge->rowmap);CHKERRQ(ierr); 281 merge->rowmap->n = pn; 282 merge->rowmap->bs = 1; 283 ierr = PetscLayoutSetUp(merge->rowmap);CHKERRQ(ierr); 284 owners = merge->rowmap->range; 285 286 /* determine the number of messages to send, their lengths */ 287 ierr = PetscMalloc(size*sizeof(PetscMPIInt),&len_si);CHKERRQ(ierr); 288 ierr = PetscMemzero(len_si,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 289 ierr = PetscMalloc(size*sizeof(PetscMPIInt),&merge->len_s);CHKERRQ(ierr); 290 len_s = merge->len_s; 291 merge->nsend = 0; 292 293 ierr = PetscMalloc((size+2)*sizeof(PetscInt),&owners_co);CHKERRQ(ierr); 294 ierr = PetscMemzero(len_s,size*sizeof(PetscMPIInt));CHKERRQ(ierr); 295 296 proc = 0; 297 for (i=0; i<pon; i++){ 298 while (prmap[i] >= owners[proc+1]) proc++; 299 len_si[proc]++; /* num of rows in Co to be sent to [proc] */ 300 len_s[proc] += coi[i+1] - coi[i]; 301 } 302 303 len = 0; /* max length of buf_si[] */ 304 owners_co[0] = 0; 305 for (proc=0; proc<size; proc++){ 306 owners_co[proc+1] = owners_co[proc] + len_si[proc]; 307 if (len_si[proc]){ 308 merge->nsend++; 309 len_si[proc] = 2*(len_si[proc] + 1); 310 len += len_si[proc]; 311 } 312 } 313 314 /* determine the number and length of messages to receive for coi and coj */ 315 ierr = PetscGatherNumberOfMessages(comm,PETSC_NULL,len_s,&merge->nrecv);CHKERRQ(ierr); 316 ierr = PetscGatherMessageLengths2(comm,merge->nsend,merge->nrecv,len_s,len_si,&merge->id_r,&merge->len_r,&len_ri);CHKERRQ(ierr); 317 318 /* post the Irecv and Isend of coj */ 319 ierr = PetscCommGetNewTag(comm,&tagj);CHKERRQ(ierr); 320 ierr = PetscPostIrecvInt(comm,tagj,merge->nrecv,merge->id_r,merge->len_r,&buf_rj,&rwaits);CHKERRQ(ierr); 321 ierr = PetscMalloc((merge->nsend+1)*sizeof(MPI_Request),&swaits);CHKERRQ(ierr); 322 for (proc=0, k=0; proc<size; proc++){ 323 if (!len_s[proc]) continue; 324 i = owners_co[proc]; 325 ierr = MPI_Isend(coj+coi[i],len_s[proc],MPIU_INT,proc,tagj,comm,swaits+k);CHKERRQ(ierr); 326 k++; 327 } 328 329 /* receives and sends of coj are complete */ 330 ierr = PetscMalloc(size*sizeof(MPI_Status),&sstatus);CHKERRQ(ierr); 331 for (i=0; i<merge->nrecv; i++){ 332 PetscMPIInt icompleted; 333 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 334 } 335 ierr = PetscFree(rwaits);CHKERRQ(ierr); 336 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 337 338 /* send and recv coi */ 339 /*-------------------*/ 340 ierr = PetscCommGetNewTag(comm,&tagi);CHKERRQ(ierr); 341 ierr = PetscPostIrecvInt(comm,tagi,merge->nrecv,merge->id_r,len_ri,&buf_ri,&rwaits);CHKERRQ(ierr); 342 ierr = PetscMalloc((len+1)*sizeof(PetscInt),&buf_s);CHKERRQ(ierr); 343 buf_si = buf_s; /* points to the beginning of k-th msg to be sent */ 344 for (proc=0,k=0; proc<size; proc++){ 345 if (!len_s[proc]) continue; 346 /* form outgoing message for i-structure: 347 buf_si[0]: nrows to be sent 348 [1:nrows]: row index (global) 349 [nrows+1:2*nrows+1]: i-structure index 350 */ 351 /*-------------------------------------------*/ 352 nrows = len_si[proc]/2 - 1; 353 buf_si_i = buf_si + nrows+1; 354 buf_si[0] = nrows; 355 buf_si_i[0] = 0; 356 nrows = 0; 357 for (i=owners_co[proc]; i<owners_co[proc+1]; i++){ 358 nzi = coi[i+1] - coi[i]; 359 buf_si_i[nrows+1] = buf_si_i[nrows] + nzi; /* i-structure */ 360 buf_si[nrows+1] =prmap[i] -owners[proc]; /* local row index */ 361 nrows++; 362 } 363 ierr = MPI_Isend(buf_si,len_si[proc],MPIU_INT,proc,tagi,comm,swaits+k);CHKERRQ(ierr); 364 k++; 365 buf_si += len_si[proc]; 366 } 367 i = merge->nrecv; 368 while (i--) { 369 PetscMPIInt icompleted; 370 ierr = MPI_Waitany(merge->nrecv,rwaits,&icompleted,&rstatus);CHKERRQ(ierr); 371 } 372 ierr = PetscFree(rwaits);CHKERRQ(ierr); 373 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,swaits,sstatus);CHKERRQ(ierr);} 374 /* 375 ierr = PetscInfo2(A,"nsend: %d, nrecv: %d\n",merge->nsend,merge->nrecv);CHKERRQ(ierr); 376 for (i=0; i<merge->nrecv; i++){ 377 ierr = PetscInfo3(A,"recv len_ri=%d, len_rj=%d from [%d]\n",len_ri[i],merge->len_r[i],merge->id_r[i]);CHKERRQ(ierr); 378 } 379 */ 380 ierr = PetscFree(len_si);CHKERRQ(ierr); 381 ierr = PetscFree(len_ri);CHKERRQ(ierr); 382 ierr = PetscFree(swaits);CHKERRQ(ierr); 383 ierr = PetscFree(sstatus);CHKERRQ(ierr); 384 ierr = PetscFree(buf_s);CHKERRQ(ierr); 385 386 #if defined(DEBUG_MATPTAP) 387 ierr = MPI_Barrier(comm);CHKERRQ(ierr); 388 if (!rank) ierr = PetscPrintf(PETSC_COMM_SELF,"[%d] Co is done\n",rank); 389 #endif 390 391 /* compute the local portion of C (mpi mat) */ 392 /*------------------------------------------*/ 393 ierr = MatGetSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 394 395 /* allocate bi array and free space for accumulating nonzero column info */ 396 ierr = PetscMalloc((pn+1)*sizeof(PetscInt),&bi);CHKERRQ(ierr); 397 bi[0] = 0; 398 399 /* set initial free space to be fill*(nnz(P) + nnz(AP)) */ 400 nnz = fill*(pi_loc[pm] + api[am]); 401 ierr = PetscFreeSpaceGet(nnz,&free_space); 402 current_space = free_space; 403 404 ierr = PetscMalloc3(merge->nrecv,PetscInt**,&buf_ri_k,merge->nrecv,PetscInt*,&nextrow,merge->nrecv,PetscInt*,&nextci);CHKERRQ(ierr); 405 for (k=0; k<merge->nrecv; k++){ 406 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 407 nrows = *buf_ri_k[k]; 408 nextrow[k] = buf_ri_k[k] + 1; /* next row number of k-th recved i-structure */ 409 nextci[k] = buf_ri_k[k] + (nrows + 1);/* poins to the next i-structure of k-th recved i-structure */ 410 } 411 ierr = MatPreallocateInitialize(comm,pn,pn,dnz,onz);CHKERRQ(ierr); 412 rmax = 0; 413 for (i=0; i<pn; i++) { 414 /* add pdt[i,:]*AP into lnk */ 415 nnz = 0; 416 pnz = pdti[i+1] - pdti[i]; 417 ptJ = pdtj + pdti[i]; 418 for (j=0; j<pnz; j++){ 419 row = ptJ[j]; /* row of AP == col of Pt */ 420 apnz = api[row+1] - api[row]; 421 Jptr = apj + api[row]; 422 /* add non-zero cols of AP into the sorted linked list lnk */ 423 ierr = PetscLLAddSorted(apnz,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 424 nnz += nlnk; 425 } 426 /* add received col data into lnk */ 427 for (k=0; k<merge->nrecv; k++){ /* k-th received message */ 428 if (i == *nextrow[k]) { /* i-th row */ 429 nzi = *(nextci[k]+1) - *nextci[k]; 430 Jptr = buf_rj[k] + *nextci[k]; 431 ierr = PetscLLAddSorted(nzi,Jptr,pN,nlnk,lnk,lnkbt);CHKERRQ(ierr); 432 nnz += nlnk; 433 nextrow[k]++; nextci[k]++; 434 } 435 } 436 437 /* if free space is not available, make more free space */ 438 if (current_space->local_remaining<nnz) { 439 ierr = PetscFreeSpaceGet(nnz+current_space->total_array_size,¤t_space);CHKERRQ(ierr); 440 nspacedouble++; 441 } 442 /* copy data into free space, then initialize lnk */ 443 ierr = PetscLLClean(pN,pN,nnz,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 444 ierr = MatPreallocateSet(i+owners[rank],nnz,current_space->array,dnz,onz);CHKERRQ(ierr); 445 current_space->array += nnz; 446 current_space->local_used += nnz; 447 current_space->local_remaining -= nnz; 448 bi[i+1] = bi[i] + nnz; 449 if (nnz > rmax) rmax = nnz; 450 } 451 ierr = MatRestoreSymbolicTranspose_SeqAIJ(p->A,&pdti,&pdtj);CHKERRQ(ierr); 452 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 453 454 ierr = PetscMalloc((bi[pn]+1)*sizeof(PetscInt),&bj);CHKERRQ(ierr); 455 ierr = PetscFreeSpaceContiguous(&free_space,bj);CHKERRQ(ierr); 456 afill_tmp = (PetscReal)bi[pn]/(pi_loc[pm] + api[am]+1); 457 if (afill_tmp > afill) afill = afill_tmp; 458 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 459 460 /* create symbolic parallel matrix Cmpi - why cannot be assembled in Numeric part - check runex55_SA ? */ 461 /*------------------------------------------------------------------------------------------------------*/ 462 ierr = PetscMalloc((rmax+1)*sizeof(PetscScalar),&vals);CHKERRQ(ierr); 463 ierr = PetscMemzero(vals,rmax*sizeof(PetscScalar));CHKERRQ(ierr); 464 465 ierr = MatCreate(comm,&Cmpi);CHKERRQ(ierr); 466 ierr = MatSetSizes(Cmpi,pn,pn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 467 ierr = MatSetBlockSizes(Cmpi,P->cmap->bs,P->cmap->bs);CHKERRQ(ierr); 468 ierr = MatSetType(Cmpi,MATMPIAIJ);CHKERRQ(ierr); 469 ierr = MatMPIAIJSetPreallocation(Cmpi,0,dnz,0,onz);CHKERRQ(ierr); 470 ierr = MatPreallocateFinalize(dnz,onz);CHKERRQ(ierr); 471 472 for (i=0; i<pn; i++){ 473 row = i + rstart; 474 nnz = bi[i+1] - bi[i]; 475 Jptr = bj + bi[i]; 476 ierr = MatSetValues(Cmpi,1,&row,nnz,Jptr,vals,INSERT_VALUES);CHKERRQ(ierr); 477 } 478 ierr = MatAssemblyBegin(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 479 ierr = MatAssemblyEnd(Cmpi,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 480 ierr = PetscFree(vals);CHKERRQ(ierr); 481 482 merge->bi = bi; 483 merge->bj = bj; 484 merge->coi = coi; 485 merge->coj = coj; 486 merge->buf_ri = buf_ri; 487 merge->buf_rj = buf_rj; 488 merge->owners_co = owners_co; 489 merge->destroy = Cmpi->ops->destroy; 490 merge->duplicate = Cmpi->ops->duplicate; 491 492 /* Cmpi is not ready for use - assembly will be done by MatPtAPNumeric() */ 493 /* Cmpi->assembled = PETSC_FALSE; */ 494 Cmpi->ops->destroy = MatDestroy_MPIAIJ_PtAP; 495 Cmpi->ops->duplicate = MatDuplicate_MPIAIJ_MatPtAP; 496 497 /* attach the supporting struct to Cmpi for reuse */ 498 c = (Mat_MPIAIJ*)Cmpi->data; 499 c->ptap = ptap; 500 ptap->api = api; 501 ptap->apj = apj; 502 ptap->merge = merge; 503 ptap->rmax = ap_rmax; 504 505 *C = Cmpi; 506 #if defined(PETSC_USE_INFO) 507 if (bi[pn] != 0) { 508 ierr = PetscInfo3(Cmpi,"Reallocs %D; Fill ratio: given %G needed %G.\n",nspacedouble,fill,afill);CHKERRQ(ierr); 509 ierr = PetscInfo1(Cmpi,"Use MatPtAP(A,P,MatReuse,%G,&C) for best performance.\n",afill);CHKERRQ(ierr); 510 } else { 511 ierr = PetscInfo(Cmpi,"Empty matrix product\n");CHKERRQ(ierr); 512 } 513 #endif 514 PetscFunctionReturn(0); 515 } 516 517 #undef __FUNCT__ 518 #define __FUNCT__ "MatPtAPNumeric_MPIAIJ_MPIAIJ" 519 PetscErrorCode MatPtAPNumeric_MPIAIJ_MPIAIJ(Mat A,Mat P,Mat C) 520 { 521 PetscErrorCode ierr; 522 Mat_Merge_SeqsToMPI *merge; 523 Mat_MPIAIJ *a=(Mat_MPIAIJ*)A->data,*p=(Mat_MPIAIJ*)P->data,*c=(Mat_MPIAIJ*)C->data; 524 Mat_SeqAIJ *ad=(Mat_SeqAIJ*)(a->A)->data,*ao=(Mat_SeqAIJ*)(a->B)->data; 525 Mat_SeqAIJ *pd=(Mat_SeqAIJ*)(p->A)->data,*po=(Mat_SeqAIJ*)(p->B)->data; 526 Mat_SeqAIJ *p_loc,*p_oth; 527 Mat_PtAPMPI *ptap; 528 PetscInt *adi=ad->i,*aoi=ao->i,*adj,*aoj,*apJ,nextp; 529 PetscInt *pi_loc,*pj_loc,*pi_oth,*pj_oth,*pJ,*pj; 530 PetscInt i,j,k,anz,pnz,apnz,nextap,row,*cj; 531 MatScalar *ada,*aoa,*apa,*pa,*ca,*pa_loc,*pa_oth,valtmp; 532 PetscInt am=A->rmap->n,cm=C->rmap->n,pon=(p->B)->cmap->n; 533 MPI_Comm comm=((PetscObject)C)->comm; 534 PetscMPIInt size,rank,taga,*len_s; 535 PetscInt *owners,proc,nrows,**buf_ri_k,**nextrow,**nextci; 536 PetscInt **buf_ri,**buf_rj; 537 PetscInt cnz=0,*bj_i,*bi,*bj,bnz,nextcj; /* bi,bj,ba: local array of C(mpi mat) */ 538 MPI_Request *s_waits,*r_waits; 539 MPI_Status *status; 540 MatScalar **abuf_r,*ba_i,*pA,*coa,*ba; 541 PetscInt *api,*apj,*coi,*coj; 542 PetscInt *poJ=po->j,*pdJ=pd->j,pcstart=P->cmap->rstart,pcend=P->cmap->rend; 543 PetscBool scalable=PETSC_TRUE; 544 545 PetscFunctionBegin; 546 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 547 ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr); 548 549 ptap = c->ptap; 550 if (!ptap) SETERRQ(((PetscObject)C)->comm,PETSC_ERR_ARG_INCOMP,"MatPtAP() has not been called to create matrix C yet, cannot use MAT_REUSE_MATRIX"); 551 merge = ptap->merge; 552 553 /* 1) get P_oth = ptap->P_oth and P_loc = ptap->P_loc */ 554 /*--------------------------------------------------*/ 555 if (ptap->reuse == MAT_INITIAL_MATRIX){ 556 ptap->reuse = MAT_REUSE_MATRIX; 557 } else { /* update numerical values of P_oth and P_loc */ 558 ierr = MatGetBrowsOfAoCols_MPIAIJ(A,P,MAT_REUSE_MATRIX,&ptap->startsj_s,&ptap->startsj_r,&ptap->bufa,&ptap->P_oth);CHKERRQ(ierr); 559 ierr = MatMPIAIJGetLocalMat(P,MAT_REUSE_MATRIX,&ptap->P_loc);CHKERRQ(ierr); 560 } 561 562 /* 2) compute numeric C_seq = P_loc^T*A_loc*P - dominating part */ 563 /*--------------------------------------------------------------*/ 564 /* get data from symbolic products */ 565 p_loc = (Mat_SeqAIJ*)(ptap->P_loc)->data; 566 p_oth = (Mat_SeqAIJ*)(ptap->P_oth)->data; 567 pi_loc=p_loc->i; pj_loc=p_loc->j; pJ=pj_loc; pa_loc=p_loc->a; 568 pi_oth=p_oth->i; pj_oth=p_oth->j; pa_oth=p_oth->a; 569 570 coi = merge->coi; coj = merge->coj; 571 ierr = PetscMalloc((coi[pon]+1)*sizeof(MatScalar),&coa);CHKERRQ(ierr); 572 ierr = PetscMemzero(coa,coi[pon]*sizeof(MatScalar));CHKERRQ(ierr); 573 574 bi = merge->bi; bj = merge->bj; 575 owners = merge->rowmap->range; 576 ierr = PetscMalloc((bi[cm]+1)*sizeof(MatScalar),&ba);CHKERRQ(ierr); 577 ierr = PetscMemzero(ba,bi[cm]*sizeof(MatScalar));CHKERRQ(ierr); 578 579 api = ptap->api; apj = ptap->apj; 580 581 /* flag 'scalable' determines which implementations to be used: 582 0: do dense axpy in MatPtAPNumeric() - fast, but requires storage of a nonscalable dense array apa; 583 1: do sparse axpy in MatPtAPNumeric() - might slow, uses a sparse array apa */ 584 /* set default scalable */ 585 ierr = PetscOptionsGetBool(((PetscObject)C)->prefix,"-matptap_scalable",&scalable,PETSC_NULL);CHKERRQ(ierr); 586 if (!scalable){ /* Do dense axpy */ 587 /*--------------------------------------------------*/ 588 /* malloc apa to store dense row A[i,:]*P - nonscalable! */ 589 ierr = PetscMalloc((P->cmap->N)*sizeof(PetscScalar),&apa);CHKERRQ(ierr); 590 ierr = PetscMemzero(apa,P->cmap->N*sizeof(PetscScalar));CHKERRQ(ierr); 591 592 for (i=0; i<am; i++) { 593 /* 2-a) form i-th sparse row of A_loc*P = Ad*P_loc + Ao*P_oth */ 594 /*------------------------------------------------------------*/ 595 apJ = apj + api[i]; 596 597 /* diagonal portion of A */ 598 anz = adi[i+1] - adi[i]; 599 adj = ad->j + adi[i]; 600 ada = ad->a + adi[i]; 601 for (j=0; j<anz; j++) { 602 row = adj[j]; 603 pnz = pi_loc[row+1] - pi_loc[row]; 604 pj = pj_loc + pi_loc[row]; 605 pa = pa_loc + pi_loc[row]; 606 607 /* perform dense axpy */ 608 valtmp = ada[j]; 609 for (k=0; k<pnz; k++){ 610 apa[pj[k]] += valtmp*pa[k]; 611 } 612 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 613 } 614 615 /* off-diagonal portion of A */ 616 anz = aoi[i+1] - aoi[i]; 617 aoj = ao->j + aoi[i]; 618 aoa = ao->a + aoi[i]; 619 for (j=0; j<anz; j++) { 620 row = aoj[j]; 621 pnz = pi_oth[row+1] - pi_oth[row]; 622 pj = pj_oth + pi_oth[row]; 623 pa = pa_oth + pi_oth[row]; 624 625 /* perform dense axpy */ 626 valtmp = aoa[j]; 627 for (k=0; k<pnz; k++){ 628 apa[pj[k]] += valtmp*pa[k]; 629 } 630 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 631 } 632 633 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 634 /*--------------------------------------------------------------*/ 635 apnz = api[i+1] - api[i]; 636 /* put the value into Co=(p->B)^T*AP (off-diagonal part, send to others) */ 637 pnz = po->i[i+1] - po->i[i]; 638 poJ = po->j + po->i[i]; 639 pA = po->a + po->i[i]; 640 for (j=0; j<pnz; j++){ 641 row = poJ[j]; 642 cnz = coi[row+1] - coi[row]; 643 cj = coj + coi[row]; 644 ca = coa + coi[row]; 645 /* perform dense axpy */ 646 valtmp = pA[j]; 647 for (k=0; k<cnz; k++) { 648 ca[k] += valtmp*apa[cj[k]]; 649 } 650 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 651 } 652 653 /* put the value into Cd (diagonal part) */ 654 pnz = pd->i[i+1] - pd->i[i]; 655 pdJ = pd->j + pd->i[i]; 656 pA = pd->a + pd->i[i]; 657 for (j=0; j<pnz; j++){ 658 row = pdJ[j]; 659 cnz = bi[row+1] - bi[row]; 660 cj = bj + bi[row]; 661 ca = ba + bi[row]; 662 /* perform dense axpy */ 663 valtmp = pA[j]; 664 for (k=0; k<cnz; k++) { 665 ca[k] += valtmp*apa[cj[k]]; 666 } 667 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 668 } 669 670 /* zero the current row of A*P */ 671 for (k=0; k<apnz; k++) apa[apJ[k]] = 0.0; 672 } 673 } else {/* Perform sparse axpy */ 674 /*----------------------------------------------------*/ 675 /* malloc apa to store sparse row A[i,:]*P */ 676 ierr = PetscMalloc((ptap->rmax+1)*sizeof(MatScalar),&apa);CHKERRQ(ierr); 677 ierr = PetscMemzero(apa,ptap->rmax*sizeof(MatScalar));CHKERRQ(ierr); 678 679 pA=pa_loc; 680 for (i=0; i<am; i++) { 681 /* form i-th sparse row of A*P */ 682 apnz = api[i+1] - api[i]; 683 apJ = apj + api[i]; 684 /* diagonal portion of A */ 685 anz = adi[i+1] - adi[i]; 686 adj = ad->j + adi[i]; 687 ada = ad->a + adi[i]; 688 for (j=0; j<anz; j++) { 689 row = adj[j]; 690 pnz = pi_loc[row+1] - pi_loc[row]; 691 pj = pj_loc + pi_loc[row]; 692 pa = pa_loc + pi_loc[row]; 693 valtmp = ada[j]; 694 nextp = 0; 695 for (k=0; nextp<pnz; k++) { 696 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 697 apa[k] += valtmp*pa[nextp++]; 698 } 699 } 700 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 701 } 702 /* off-diagonal portion of A */ 703 anz = aoi[i+1] - aoi[i]; 704 aoj = ao->j + aoi[i]; 705 aoa = ao->a + aoi[i]; 706 for (j=0; j<anz; j++) { 707 row = aoj[j]; 708 pnz = pi_oth[row+1] - pi_oth[row]; 709 pj = pj_oth + pi_oth[row]; 710 pa = pa_oth + pi_oth[row]; 711 valtmp = aoa[j]; 712 nextp = 0; 713 for (k=0; nextp<pnz; k++) { 714 if (apJ[k] == pj[nextp]) { /* col of AP == col of P */ 715 apa[k] += valtmp*pa[nextp++]; 716 } 717 } 718 ierr = PetscLogFlops(2.0*pnz);CHKERRQ(ierr); 719 } 720 721 /* 2-b) Compute Cseq = P_loc[i,:]^T*AP[i,:] using outer product */ 722 /*--------------------------------------------------------------*/ 723 pnz = pi_loc[i+1] - pi_loc[i]; 724 pJ = pj_loc + pi_loc[i]; 725 for (j=0; j<pnz; j++) { 726 nextap = 0; 727 row = pJ[j]; /* global index */ 728 if (row < pcstart || row >=pcend) { /* put the value into Co */ 729 row = *poJ; 730 cj = coj + coi[row]; 731 ca = coa + coi[row]; poJ++; 732 } else { /* put the value into Cd */ 733 row = *pdJ; 734 cj = bj + bi[row]; 735 ca = ba + bi[row]; pdJ++; 736 } 737 valtmp = pA[j]; 738 for (k=0; nextap<apnz; k++) { 739 if (cj[k]==apJ[nextap]) ca[k] += valtmp*apa[nextap++]; 740 } 741 ierr = PetscLogFlops(2.0*apnz);CHKERRQ(ierr); 742 } 743 pA += pnz; 744 /* zero the current row info for A*P */ 745 ierr = PetscMemzero(apa,apnz*sizeof(MatScalar));CHKERRQ(ierr); 746 } 747 } 748 ierr = PetscFree(apa);CHKERRQ(ierr); 749 750 /* 3) send and recv matrix values coa */ 751 /*------------------------------------*/ 752 buf_ri = merge->buf_ri; 753 buf_rj = merge->buf_rj; 754 len_s = merge->len_s; 755 ierr = PetscCommGetNewTag(comm,&taga);CHKERRQ(ierr); 756 ierr = PetscPostIrecvScalar(comm,taga,merge->nrecv,merge->id_r,merge->len_r,&abuf_r,&r_waits);CHKERRQ(ierr); 757 758 ierr = PetscMalloc2(merge->nsend+1,MPI_Request,&s_waits,size,MPI_Status,&status);CHKERRQ(ierr); 759 for (proc=0,k=0; proc<size; proc++){ 760 if (!len_s[proc]) continue; 761 i = merge->owners_co[proc]; 762 ierr = MPI_Isend(coa+coi[i],len_s[proc],MPIU_MATSCALAR,proc,taga,comm,s_waits+k);CHKERRQ(ierr); 763 k++; 764 } 765 if (merge->nrecv) {ierr = MPI_Waitall(merge->nrecv,r_waits,status);CHKERRQ(ierr);} 766 if (merge->nsend) {ierr = MPI_Waitall(merge->nsend,s_waits,status);CHKERRQ(ierr);} 767 768 ierr = PetscFree2(s_waits,status);CHKERRQ(ierr); 769 ierr = PetscFree(r_waits);CHKERRQ(ierr); 770 ierr = PetscFree(coa);CHKERRQ(ierr); 771 772 /* 4) insert local Cseq and received values into Cmpi */ 773 /*------------------------------------------------------*/ 774 ierr = PetscMalloc3(merge->nrecv,PetscInt**,&buf_ri_k,merge->nrecv,PetscInt*,&nextrow,merge->nrecv,PetscInt*,&nextci);CHKERRQ(ierr); 775 for (k=0; k<merge->nrecv; k++){ 776 buf_ri_k[k] = buf_ri[k]; /* beginning of k-th recved i-structure */ 777 nrows = *(buf_ri_k[k]); 778 nextrow[k] = buf_ri_k[k]+1; /* next row number of k-th recved i-structure */ 779 nextci[k] = buf_ri_k[k] + (nrows + 1);/* poins to the next i-structure of k-th recved i-structure */ 780 } 781 782 for (i=0; i<cm; i++) { 783 row = owners[rank] + i; /* global row index of C_seq */ 784 bj_i = bj + bi[i]; /* col indices of the i-th row of C */ 785 ba_i = ba + bi[i]; 786 bnz = bi[i+1] - bi[i]; 787 /* add received vals into ba */ 788 for (k=0; k<merge->nrecv; k++){ /* k-th received message */ 789 /* i-th row */ 790 if (i == *nextrow[k]) { 791 cnz = *(nextci[k]+1) - *nextci[k]; 792 cj = buf_rj[k] + *(nextci[k]); 793 ca = abuf_r[k] + *(nextci[k]); 794 nextcj = 0; 795 for (j=0; nextcj<cnz; j++){ 796 if (bj_i[j] == cj[nextcj]){ /* bcol == ccol */ 797 ba_i[j] += ca[nextcj++]; 798 } 799 } 800 nextrow[k]++; nextci[k]++; 801 ierr = PetscLogFlops(2.0*cnz);CHKERRQ(ierr); 802 } 803 } 804 ierr = MatSetValues(C,1,&row,bnz,bj_i,ba_i,INSERT_VALUES);CHKERRQ(ierr); 805 } 806 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 807 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 808 809 ierr = PetscFree(ba);CHKERRQ(ierr); 810 ierr = PetscFree(abuf_r[0]);CHKERRQ(ierr); 811 ierr = PetscFree(abuf_r);CHKERRQ(ierr); 812 ierr = PetscFree3(buf_ri_k,nextrow,nextci);CHKERRQ(ierr); 813 #if defined(PETSC_USE_INFO) 814 if (scalable){ 815 ierr = PetscInfo(C,"Use scalable sparse axpy\n");CHKERRQ(ierr); 816 } else { 817 ierr = PetscInfo(C,"Use non-scalable dense axpy\n");CHKERRQ(ierr); 818 } 819 #endif 820 PetscFunctionReturn(0); 821 } 822