acl/lib_fiber/samples/event_mutex/main.cpp
2022-10-27 06:14:15 -04:00

183 lines
3.8 KiB
C++

#include "stdafx.h"
#include <stdio.h>
#include <stdlib.h>
#include <vector>
static bool __show = false;
static int __nthreads = 2;
static int __nfibers = 2;
static int __nloop = 2;
static int __delay = 0;
static acl::fiber_event_t __event_type = acl::FIBER_EVENT_T_KERNEL;
static acl::atomic_long __counter = 0;
class myfiber : public acl::fiber {
public:
myfiber(acl::event_mutex& lock, int& nfibers)
: lock_(lock)
, nfibers_(nfibers)
{
}
protected:
// @override
void run(void) {
for (int i = 0; i < __nloop; i++) {
if (__show) {
printf("thread-%lu, fiber-%u begin lock\r\n",
acl::thread::thread_self(),
acl::fiber::self());
}
if (lock_.lock() == false) {
printf("lock error %s\r\n", acl::last_serror());
break;
}
if (__show) {
printf("thread-%lu, fiber-%u lock ok\r\n",
acl::thread::thread_self(),
acl::fiber::self());
}
if (__delay > 0) {
acl_doze(__delay);
}
if (__show) {
printf("thread-%lu, fiber-%u begin unlock\r\n",
acl::thread::thread_self(),
acl::fiber::self());
}
lock_.unlock();
__counter++;
if (__show) {
printf("thread-%lu, fiber-%u unlock ok, counter-%lld\r\n",
acl::thread::thread_self(), acl::fiber::self(),
__counter.value());
}
// acl_doze(10);
// acl::fiber::yield();
}
--nfibers_;
printf("thread-%lu, left nfibers: %d\r\n", acl::thread::self(), nfibers_);
if (nfibers_ == 0) {
acl::fiber::schedule_stop();
}
delete this;
}
private:
acl::event_mutex& lock_;
int& nfibers_;
~myfiber(void) {}
};
class mythread : public acl::thread {
public:
mythread(acl::event_mutex& lock) : lock_(lock) {
this->set_detachable(false);
}
~mythread(void) {}
private:
acl::event_mutex& lock_;
// @override
void* run(void) {
int nfibers = __nfibers, n = nfibers;
for (int i = 0; i < n; i++) {
acl::fiber* fb = new myfiber(lock_, nfibers);
fb->start();
}
acl::fiber::schedule_with(__event_type);
return NULL;
}
};
//////////////////////////////////////////////////////////////////////////////
static void usage(const char* procname) {
printf("usage: %s -h [help]\r\n"
" -e event_type[select|poll|kernel|io_uring]\r\n"
" -t nthreads[default: 2]\r\n"
" -c nfibers[default: 2]\r\n"
" -n nloop[default: 2]\r\n"
" -S [show info]\r\n"
" -d delay[default: 100 ms]\r\n"
, procname);
}
int main(int argc, char *argv[]) {
int ch;
acl::acl_cpp_init();
acl::log::stdout_open(true);
while ((ch = getopt(argc, argv, "he:c:t:n:d:S")) > 0) {
switch (ch) {
case 'h':
usage(argv[0]);
return 0;
case 'e':
if (strcasecmp(optarg, "kernel") == 0) {
__event_type = acl::FIBER_EVENT_T_KERNEL;
} else if (strcasecmp(optarg, "poll") == 0) {
__event_type = acl::FIBER_EVENT_T_POLL;
} else if (strcasecmp(optarg, "select") == 0) {
__event_type = acl::FIBER_EVENT_T_SELECT;
} else if (strcasecmp(optarg, "io_uring") == 0) {
__event_type = acl::FIBER_EVENT_T_IO_URING;
}
break;
case 'c':
__nfibers = atoi(optarg);
break;
case 't':
__nthreads = atoi(optarg);
break;
case 'n':
__nloop = atoi(optarg);
break;
case 'd':
__delay = atoi(optarg);
break;
case 'S':
__show = true;
break;
default:
break;
}
}
acl::event_mutex lock(false);
std::vector<acl::thread*> threads;
for (int i = 0; i < __nthreads; i++) {
acl::thread* thr = new mythread(lock);
threads.push_back(thr);
thr->start();
}
for (std::vector<acl::thread*>::iterator it = threads.begin();
it != threads.end(); ++it) {
(*it)->wait();
delete *it;
}
printf("All over, thread=%d, fibers=%d, nloop=%d, counter=%lld\r\n",
__nthreads, __nfibers, __nloop, __counter.value());
return 0;
}