void matrix_inplace_matmul_mt2(matrix_type * A, const matrix_type * B , thread_pool_type * thread_pool){
int num_threads = thread_pool_get_max_running( thread_pool );
arg_pack_type ** arglist = util_malloc( num_threads * sizeof * arglist );
int it;
thread_pool_restart( thread_pool );
{
int rows = matrix_get_rows( A ) / num_threads;
int rows_mod = matrix_get_rows( A ) % num_threads;
int row_offset = 0;
for (it = 0; it < num_threads; it++) {
int row_size;
arglist[it] = arg_pack_alloc();
row_size = rows;
if (it < rows_mod)
row_size += 1;
arg_pack_append_int(arglist[it] , row_offset );
arg_pack_append_int(arglist[it] , row_size );
arg_pack_append_ptr(arglist[it] , A );
arg_pack_append_const_ptr(arglist[it] , B );
thread_pool_add_job( thread_pool , matrix_inplace_matmul_mt__ , arglist[it]);
row_offset += row_size;
}
}
thread_pool_join( thread_pool );
for (it = 0; it < num_threads; it++)
arg_pack_free( arglist[it] );
free( arglist );
}
thread_pool_type * thread_pool_alloc(int max_running , bool start_queue) {
thread_pool_type * pool = util_malloc( sizeof *pool );
pool->job_slots = util_calloc( max_running , sizeof * pool->job_slots );
pool->max_running = max_running;
pool->queue = NULL;
pool->accepting_jobs = false;
pthread_rwlock_init( &pool->queue_lock , NULL);
thread_pool_resize_queue( pool , 32 );
if (start_queue)
thread_pool_restart( pool );
return pool;
}