/*
--------------------------------------------------------
purpose -- assemble any aggregates in the aggregate list
created -- 98mar26, cca
--------------------------------------------------------
*/
static void
assembleAggregates (
int J,
SubMtx *BJ,
SubMtxList *aggList,
SubMtxManager *mtxmanager,
int msglvl,
FILE *msgFile
) {
SubMtx *BJhat, *BJhead ;
double *entBJ, *entBJhat ;
int inc1, inc1hat, inc2, inc2hat, ncol, ncolhat, nrow, nrowhat ;
if ( BJ == NULL || aggList == NULL ) {
fprintf(stderr,
"\n fatal error in assembleAggregates()"
"\n BJ = %p, aggList = %p", BJ, aggList) ;
exit(-1) ;
}
if ( SubMtxList_isListNonempty(aggList, BJ->rowid) ) {
if ( msglvl > 3 ) {
fprintf(msgFile, "\n\n aggregate list is not-empty") ;
fflush(msgFile) ;
}
SubMtx_denseInfo(BJ, &nrow, &ncol, &inc1, &inc2, &entBJ) ;
if ( msglvl > 2 ) {
fprintf(msgFile,
"\n\n BJ(%d,%d) : nrow %d, ncol %d, inc1 %d, inc2 %d, ent %p",
BJ->rowid, BJ->colid, nrow, ncol, inc1, inc2, entBJ) ;
fflush(msgFile) ;
}
BJhead = SubMtxList_getList(aggList, J) ;
for ( BJhat = BJhead ; BJhat != NULL ; BJhat = BJhat->next ) {
if ( BJhat == NULL ) {
fprintf(stderr,
"\n 3. fatal error in forwardVisit(%d)"
"\n BJhat = NULL", J) ;
exit(-1) ;
}
SubMtx_denseInfo(BJhat, &nrowhat, &ncolhat, &inc1hat, &inc2hat,
&entBJhat) ;
if ( msglvl > 2 ) {
fprintf(msgFile,
"\n BJhat(%d,%d) : nrow %d, ncol %d, inc1 %d, inc2 %d, ent %p",
BJhat->rowid, BJhat->colid,
nrowhat, ncolhat, inc1hat, inc2hat, entBJhat) ;
fflush(msgFile) ;
}
if ( nrow != nrowhat || ncol != ncolhat
|| inc1 != inc1hat || inc2 != inc2hat || entBJhat == NULL ) {
fprintf(stderr, "\n fatal error") ;
exit(-1) ;
}
if ( msglvl > 3 ) {
fprintf(msgFile, "\n\n BJ") ;
SubMtx_writeForHumanEye(BJ, msgFile) ;
fprintf(msgFile, "\n\n BJhat") ;
SubMtx_writeForHumanEye(BJhat, msgFile) ;
fflush(msgFile) ;
}
if ( SUBMTX_IS_REAL(BJhat) ) {
DVadd(nrow*ncol, entBJ, entBJhat) ;
} else if ( SUBMTX_IS_COMPLEX(BJhat) ) {
DVadd(2*nrow*ncol, entBJ, entBJhat) ;
}
}
SubMtxManager_releaseListOfObjects(mtxmanager, BJhead) ;
if ( msglvl > 3 ) {
fprintf(msgFile, "\n\n BJ after assembly") ;
SubMtx_writeForHumanEye(BJ, msgFile) ;
fflush(msgFile) ;
}
}
return ; }
/*--------------------------------------------------------------------*/
static void
InpMtx_MT_mmm (
int flag,
InpMtx *A,
DenseMtx *Y,
double alpha[],
DenseMtx *X,
int nthread,
int msglvl,
FILE *msgFile
) {
double t1, t2 ;
int myid, nent, rc ;
MTmvmObj *MTmvmObjs, *obj ;
/*
-------------------------------
set up the nthread data objects
-------------------------------
*/
MARKTIME(t1) ;
MTmvmObjs = setup(A, Y, alpha, X, nthread) ;
MARKTIME(t2) ;
if ( msglvl > 0 ) {
fprintf(msgFile, "\n %% CPU %8.3f : setup time", t2 - t1) ;
}
#if THREAD_TYPE == TT_POSIX
{
pthread_t *tids ;
pthread_attr_t attr ;
void *status ;
/*
##### NOTE: for SGI machines, this command must be present
##### for the thread scheduling to be efficient.
##### this is NOT a POSIX call, but SGI needs it anyway
pthread_setconcurrency(nthread) ;
*/
pthread_attr_init(&attr) ;
/*
pthread_attr_setscope(&attr, PTHREAD_SCOPE_SYSTEM) ;
*/
pthread_attr_setscope(&attr, PTHREAD_SCOPE_PROCESS) ;
ALLOCATE(tids, pthread_t, nthread) ;
MARKTIME(t1) ;
for ( myid = 0, obj = MTmvmObjs ; myid < nthread ; myid++, obj++ ) {
switch ( flag ) {
case NONSYM :
rc = pthread_create(&tids[myid], &attr, worker_nonsym_mmm, obj) ;
break ;
case SYM :
rc = pthread_create(&tids[myid], &attr, worker_sym_mmm, obj) ;
break ;
case HERM :
rc = pthread_create(&tids[myid], &attr, worker_herm_mmm, obj) ;
break ;
case NONSYM_T :
rc = pthread_create(&tids[myid], &attr, worker_nonsym_mmm_T, obj);
break ;
case NONSYM_H :
rc = pthread_create(&tids[myid], &attr, worker_nonsym_mmm_H, obj);
break ;
}
if ( rc != 0 ) {
fprintf(stderr,
"\n fatal error, myid = %d, rc = %d from pthread_create",
myid, rc) ;
exit(-1) ;
} else if ( msglvl > 2 ) {
fprintf(stderr, "\n %% thread %d created", myid) ;
}
}
MARKTIME(t2) ;
if ( msglvl > 0 ) {
fprintf(msgFile, "\n %% CPU %8.3f : thread creation time", t2 - t1) ;
}
MARKTIME(t1) ;
for ( myid = 0 ; myid < nthread ; myid++ ) {
pthread_join(tids[myid], &status) ;
}
MARKTIME(t2) ;
if ( msglvl > 0 ) {
fprintf(msgFile, "\n %% CPU %8.3f : thread join time", t2 - t1) ;
}
FREE(tids) ;
pthread_attr_destroy(&attr) ;
}
#endif
/*
-------------------------------------
accumulate the rhs hand side matrices
-------------------------------------
*/
MARKTIME(t1) ;
nent = Y->nrow * Y->ncol ;
for ( myid = 1, obj = MTmvmObjs + 1 ;
myid < nthread ;
myid++, obj++ ) {
if ( INPMTX_IS_REAL_ENTRIES(A) ) {
DVadd(nent, DenseMtx_entries(Y), DenseMtx_entries(obj->Y)) ;
} else if ( INPMTX_IS_COMPLEX_ENTRIES(A) ) {
DVadd(2*nent, DenseMtx_entries(Y), DenseMtx_entries(obj->Y)) ;
}
}
MARKTIME(t2) ;
if ( msglvl > 0 ) {
fprintf(msgFile,
"\n %% CPU %8.3f : time to accumulate rhs", t2 - t1) ;
}
/*
---------------------------
release the data structures
---------------------------
*/
MARKTIME(t1) ;
for ( myid = 0, obj = MTmvmObjs ; myid < nthread ; myid++, obj++ ) {
InpMtx_free(obj->A) ;
if ( myid > 0 ) {
DenseMtx_free(obj->Y) ;
}
}
FREE(MTmvmObjs) ;
MARKTIME(t2) ;
if ( msglvl > 0 ) {
fprintf(msgFile,
"\n %% CPU %8.3f : time to release and free data", t2 - t1) ;
}
return ; }
