* Fork/Join Framework
* Test that tasks are actually processed concurrently when more than 2 worker threads
* are forked. This test would fail if the framework inadvertently held a global
* lock during task processing.
Copyright By PowCoder代写 加微信 powcoder
* This is part of the minimum requirements because such a flaw would be considered
* disastrous, completely negating the point of a thread pool execution framework.
* Written by G. Back for CS3214 Spring 2016.
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include “threadpool.h”
#include “threadpool_lib.h”
#define DEFAULT_THREADS 2
#define DEFAULT_TASKS 40
struct time_range {
uintptr_t taskno;
struct timeval start;
struct timeval end;
static void *
test_task(struct thread_pool *pool, struct time_range * data)
struct timespec sleep_time = {
.tv_sec = 0,
.tv_nsec = 50*1000*1000 // 50 ms should be plenty to allow for overlap
gettimeofday(&data->start, NULL);
nanosleep(&sleep_time, NULL);
gettimeofday(&data->end, NULL);
return (void *)data->taskno;
static int
run_test(int nthreads, int ntasks)
struct benchmark_data * bdata = start_benchmark();
struct thread_pool * threadpool = thread_pool_new(nthreads);
struct time_range start_end[ntasks];
struct future * futures[ntasks];
printf(“starting %d tasks…\n”, ntasks);
for (int i = 0; i < ntasks; i++) {
start_end[i].taskno = i;
futures[i] = thread_pool_submit(threadpool, (fork_join_task_t) test_task, start_end + i);
for (int i = 0; i < ntasks; i++) {
uintptr_t r = (uintptr_t) future_get(futures[i]);
// consistency check
if (r != i) {
fprintf(stderr, "Wrong result, expected %d, got %lu\n", i, r);
future_free(futures[i]);
thread_pool_shutdown_and_destroy(threadpool);
stop_benchmark(bdata);
// now analyze the data. We expect that at least nthreads different
// tasks were running concurrently. In other words, there exists a
// task t that was started while nthreads-1 other tasks had been started,
// but before each of these tasks finished.
// Try to find it.
int best = 0;
for (int i = 0; i < ntasks; i++) {
int conc_with_i = 0;
for (int j = 0; j < ntasks; j++) if (i != j) {
if (timercmp(&start_end[j].start, &start_end[i].start, <) &&
timercmp(&start_end[i].start, &start_end[j].end, <)) {
// j was running while when i was started.
conc_with_i++;
if (conc_with_i > best)
best = conc_with_i;
if (best < nthreads - 1) {
fprintf(stderr, "Wrong result, expected that %d tasks execute concurrently when "
"forked from the main thread, but I could only find a parallelism level of %d\n",
nthreads - 1, best);
report_benchmark_results(bdata);
printf("Test successful.\n");
free(bdata);
/**********************************************************************************/
static void
usage(char *av0, int exvalue)
fprintf(stderr, "Usage: %s [-n
” -n number of threads in pool, default %d\n”
” -t number of tasks, default %d\n”
, av0, DEFAULT_THREADS, DEFAULT_TASKS);
exit(exvalue);
main(int ac, char *av[])
int c, nthreads = DEFAULT_THREADS, ntasks = DEFAULT_TASKS;
while ((c = getopt(ac, av, “hn:t:”)) != EOF) {
switch (c) {
nthreads = atoi(optarg);
if (nthreads == 1) {
fprintf(stderr, “Cannot test for concurrency with only 1 thread\n”);
ntasks = atoi(optarg);
usage(av[0], EXIT_SUCCESS);
return run_test(nthreads, ntasks);
程序代写 CS代考 加微信: powcoder QQ: 1823890830 Email: powcoder@163.com