int main(int argc , char* argv[])
{
svar.ParseMain(argc,argv);
mode=svar.GetInt("Mode",0);
if(mode<0||mode>=9) return -1;
circulate=svar.GetInt("Circle",circulate);
THREAD_COUNT=svar.GetInt("Count",THREAD_COUNT);
usec=svar.GetInt("Delay",usec);
modeStr=modes[mode];
std::vector<Thread> thread_array(THREAD_COUNT);
boost::progress_timer t;
for(int i=0;i<10;i++)
queue.write(i);
std::cout<<"-----------thread create-----------"<<std::endl;
for(int i=0;i<THREAD_COUNT;++i)
{
thread_array[i].reset(new boost::thread(testdata));
}
while(1)
{
if(count1>=THREAD_COUNT)
return 0;
}
}
void amuxCRSParallelPThreads (double a,
unsigned n, unsigned const * const iA, unsigned const * const jA,
double const * const A, double const * const x, double* y,
unsigned num_of_pthreads)
{
#ifdef HAVE_PTHREADS
// fill thread data objects
MatMultThreadParam** thread_param_array (new MatMultThreadParam*[num_of_pthreads]);
double step_size ((double)(n)/(double)(num_of_pthreads));
for (unsigned k(0); k<num_of_pthreads; k++) {
const unsigned beg (static_cast<unsigned>(k*step_size));
const unsigned end (static_cast<unsigned>((k+1)*step_size));
thread_param_array[k] = new MatMultThreadParam (a, beg, end, iA, jA, A, x, y);
}
// allocate thread_array and return value array
pthread_t *thread_array (new pthread_t[num_of_pthreads]);
int *ret_vals (new int[num_of_pthreads]);
// create threads
for (unsigned k(0); k<num_of_pthreads; k++) {
ret_vals[k] = pthread_create( &(thread_array[k]), NULL, amuxCRSpthread, thread_param_array[k]);
}
// join threads
for (unsigned k(0); k<num_of_pthreads; k++) {
pthread_join (thread_array[k], NULL);
}
delete [] ret_vals;
for (unsigned k(0); k<num_of_pthreads; k++)
delete thread_param_array[k];
delete [] thread_param_array;
delete [] thread_array;
#else
(void)num_of_pthreads;
amuxCRS (a, n, iA, jA, A, x, y);
#endif
}
void amuxCRSParallelPThreads (double a,
unsigned n, unsigned const * const iA, unsigned const * const jA,
double const * const A, double const * const x, double* y,
unsigned num_of_pthreads, unsigned const*const workload_intervals)
{
(void) n; // Unused if HAVE_PTHREADS is not defined.
#ifdef HAVE_PTHREADS
// fill thread data objects
MatMultThreadParam** thread_param_array (new MatMultThreadParam*[num_of_pthreads]);
for (unsigned k(0); k<num_of_pthreads; k++) {
thread_param_array[k] = new MatMultThreadParam (a, workload_intervals[k], workload_intervals[k+1], iA, jA, A, x, y);
}
// allocate thread_array and return value array
pthread_t *thread_array (new pthread_t[num_of_pthreads]);
int *ret_vals (new int[num_of_pthreads]);
// create threads
for (unsigned k(0); k<num_of_pthreads; k++) {
ret_vals[k] = pthread_create( &(thread_array[k]), NULL, amuxCRSpthread, thread_param_array[k]);
}
// join threads
for (unsigned k(0); k<num_of_pthreads; k++) {
pthread_join (thread_array[k], NULL);
}
delete [] ret_vals;
for (unsigned k(0); k<num_of_pthreads; k++)
delete thread_param_array[k];
delete [] thread_param_array;
delete [] thread_array;
#else
(void)num_of_pthreads;
amuxCRS (a, n, iA, jA, A, x, y);
#endif
}
BASKER_INLINE
int Basker<Int,Entry,Exe_Space>::nfactor_domain_error
(
INT_1DARRAY threads_start
)
{
Int nthread_remalloc = 0;
for(Int ti = 0; ti < num_threads; ti++)
{
//Note: jdb we can make this into a switch
if(thread_array(ti).error_type ==
BASKER_ERROR_NOERROR)
{
threads_start(ti) = BASKER_MAX_IDX;
continue;
}//end if NOERROR
if(thread_array(ti).error_type ==
BASKER_ERROR_SINGULAR)
{
printf("ERROR THREAD: %d DOMBLK SINGULAR: %d\n",
ti,
thread_array(ti).error_blk);
return BASKER_ERROR;
}//end if SINGULAR
if(thread_array(ti).error_type ==
BASKER_ERROR_NOMALLOC)
{
printf("ERROR THREADS: %d DOMBLK NOMALLOC: %d\n",
ti,
thread_array(ti).error_blk);
return BASKER_ERROR;
}//end if NOMALLOC
if(thread_array(ti).error_type ==
BASKER_ERROR_REMALLOC)
{
BASKER_ASSERT(thread_array(ti).error_blk > 0,
"nfactor_dom_error error_blk");
printf("ERROR THREADS: %d DOMBLK MALLOC: %d \n",
ti,
thread_array(ti).error_blk);
//Resize L
BASKER_MATRIX &L = LL(thread_array(ti).error_blk)(0);
REALLOC_INT_1DARRAY(L.row_idx,
L.nnz,
thread_array(ti).error_info);
REALLOC_ENTRY_1DARRAY(L.val,
L.nnz,
thread_array(ti).error_info);
L.nnz = thread_array(ti).error_info;
//clean up workspace
if(L.w_fill == BASKER_TRUE)
{
//Clear workspace
for(Int i = 0; i < L.iws_size*L.iws_mult; ++i)
{
L.iws(i) = (Int) 0;
}
for(Int i = 0; i < L.ews_size*L.ews_mult; ++i)
{
L.ews(i) = (Entry) 0;
}
//Clear perm
for(Int i = L.srow; i < L.srow+L.nrow; ++i)
{
gperm(i) = BASKER_MAX_IDX;
}
}
//Resize U
BASKER_MATRIX &U = LU(thread_array(ti).error_blk)(0);
REALLOC_INT_1DARRAY(U.row_idx,
U.nnz,
thread_array(ti).error_info);
REALLOC_ENTRY_1DARRAY(U.val,
U.nnz,
thread_array(ti).error_info);
U.nnz = thread_array(ti).error_info;
threads_start(ti) = thread_array(ti).error_blk;
//Reset
thread_array(ti).error_type = BASKER_ERROR_NOERROR;
thread_array(ti).error_blk = BASKER_MAX_IDX;
thread_array(ti).error_info = BASKER_MAX_IDX;
nthread_remalloc++;
}//if REMALLOC
}//for all threads
if(nthread_remalloc == 0)
{
return BASKER_SUCCESS;
}
else
{
return nthread_remalloc;
}
//Should never be here
BASKER_ASSERT(0==1, "nfactor_diag_error, should never");
return BASKER_SUCCESS;
}//end nfactor_domain_error
BASKER_INLINE
int Basker<Int,Entry,Exe_Space>::nfactor_diag_error
(
INT_1DARRAY threads_start
)
{
Int nthread_remalloc = 0;
for(Int ti = 0; ti < num_threads; ti++)
{
//Note: jdb we can make this into a switch
if(thread_array(ti).error_type ==
BASKER_ERROR_NOERROR)
{
threads_start(ti) = BASKER_MAX_IDX;
continue;
}//end if NOERROR
if(thread_array(ti).error_type ==
BASKER_ERROR_SINGULAR)
{
printf("ERROR THREAD: %d DIAGBLK SINGULAR: %d\n",
ti,
thread_array(ti).error_blk);
return BASKER_ERROR;
}//end if SINGULAR
if(thread_array(ti).error_type ==
BASKER_ERROR_NOMALLOC)
{
printf("ERROR THREADS: %d DIAGBLK NOMALLOC: %d\n",
ti,
thread_array(ti).error_blk);
return BASKER_ERROR;
}//end if NOMALLOC
if(thread_array(ti).error_type ==
BASKER_ERROR_REMALLOC)
{
BASKER_ASSERT(thread_array(ti).error_blk > 0,
"nfactor_diag_error error_blk");
printf("ERROR THREADS: %d DIAGBLK MALLOC: %d \n",
ti,
thread_array(ti).error_blk);
//Clean the workspace
printf("test: %d %d \n",
thread_array(ti).iws_size*thread_array(ti).iws_mult,
thread_array(ti).ews_size*thread_array(ti).ews_mult);
for(Int i = 0;
i < thread_array(ti).iws_size*thread_array(ti).iws_mult;
i++)
{
thread_array(ti).iws(i) = (Int) 0;
}
for(Int i = 0;
i < thread_array(ti).ews_size*thread_array(ti).ews_mult;
i++)
{
thread_array(ti).ews(i) = (Entry) 0;
}
//Resize L
BASKER_MATRIX &L = LBTF(thread_array(ti).error_blk);
REALLOC_INT_1DARRAY(L.row_idx,
L.nnz,
thread_array(ti).error_info);
REALLOC_ENTRY_1DARRAY(L.val,
L.nnz,
thread_array(ti).error_info);
L.nnz = thread_array(ti).error_info;
for(Int i = 0; i < L.ncol; i++)
{
L.col_ptr(i) = 0;
}
for(Int i = L.srow; i < (L.srow+L.nrow); i++)
{
gperm(i) = BASKER_MAX_IDX;
}
//Resize U
BASKER_MATRIX &U = UBTF(thread_array(ti).error_blk);
REALLOC_INT_1DARRAY(U.row_idx,
U.nnz,
thread_array(ti).error_info);
REALLOC_ENTRY_1DARRAY(U.val,
U.nnz,
thread_array(ti).error_info);
U.nnz = thread_array(ti).error_info;
for(Int i = 0; i < U.ncol; i++)
{
U.col_ptr(i) = 0;
}
printf("Setting thread start(%d) %d \n",
ti, thread_array(ti).error_blk);
threads_start(ti) = thread_array(ti).error_blk;
//Reset
thread_array(ti).error_type = BASKER_ERROR_NOERROR;
thread_array(ti).error_blk = BASKER_MAX_IDX;
thread_array(ti).error_info = BASKER_MAX_IDX;
nthread_remalloc++;
}//if REMALLOC
}//for all threads
if(nthread_remalloc == 0)
{
return BASKER_SUCCESS;
}
else
{
return nthread_remalloc;
}
//Should never be here
BASKER_ASSERT(0==1, "nfactor_diag_error, should never");
return BASKER_SUCCESS;
}//end nfactor_diag_error
