The parallel\Channel class

(0.9.0)

Unbuffered Channels

An unbuffered channel will block on calls to parallel\Channel::send() until there is a receiver, and block on calls to parallel\Channel::recv() until there is a sender. This means an unbuffered channel is not only a way to share data among tasks but also a simple method of synchronization.

An unbuffered channel is the fastest way to share data among tasks, requiring the least copying.

Buffered Channels

A buffered channel will not block on calls to parallel\Channel::send() until capacity is reached, calls to parallel\Channel::recv() will block until there is data in the buffer.

Closures over Channels

A powerful feature of parallel channels is that they allow the exchange of closures between tasks (and runtimes).

When a closure is sent over a channel the closure is buffered, it doesn't change the buffering of the channel transmitting the closure, but it does effect the static scope inside the closure: The same closure sent to different runtimes, or the same runtime, will not share their static scope.

This means that whenever a closure is executed that was transmitted by a channel, static state will be as it was when the closure was buffered.

Anonymous Channels

The anonymous channel constructor allows the programmer to avoid assigning names to every channel: parallel will generate a unique name for anonymous channels.

Class synopsis

final class parallel\Channel {
/* Anonymous Constructor */
public __construct()
public __construct(int $capacity)
/* Access */
public make(string $name): Channel
public make(string $name, int $capacity): Channel
public open(string $name): Channel
/* Sharing */
public recv(): mixed
public send(mixed $value): void
/* Closing */
public close(): void
/* Constant for Infinitely Buffered */
const Infinite;
}

Table of Contents

add a note

User Contributed Notes 4 notes

up
4
hdvianna
4 years ago
This is an example of using a channel to produce data for consumers. In this example, the producer Runtime instance will send the time in seconds in which the consumers shall sleep.

<?php

use parallel\{Runtime, Channel};

main($argv);

function
main(array $argv)
{
if (
count($argv) !== 3) {
echo
"Type: hello-parallel.php <number-of-tasks> <maximum-time-of-sleep (in seconds)>" . PHP_EOL;
echo
"Example: hello-parallel.php 5 3" . PHP_EOL;
die;
} else {
$numberOfTasks = intval($argv[1]);
$maximumTimeOfSleep = intval($argv[2]);
$t1 = microtime(true);
parallelize($numberOfTasks, $maximumTimeOfSleep);
$endTime = microtime(true) - $t1;
echo
PHP_EOL."Finished $numberOfTasks task(s) in {$endTime}s".PHP_EOL;
}
}

function
parallelize(int $numberOfTasks, int $maximumTimeOfSleep)
{
$channel = new Channel();

$taskIds = array_map(function () use ($maximumTimeOfSleep) {
return
$id = uniqid("task::");
},
range(0, $numberOfTasks - 1));

$timesToSleep = array_map(function () use ($maximumTimeOfSleep) {
return
rand(1, $maximumTimeOfSleep);
},
$taskIds);

$producer = new Runtime();
$producerFuture = $producer->run(function (Channel $channel, array $timesToSleep) {
foreach (
$timesToSleep as $timeToSleep) {
$channel->send($timeToSleep);
}
}, [
$channel, $timesToSleep]);

$consumerFutures = array_map(function (string $id) use ($channel) {
$runtime = new Runtime();
return
$runtime->run(function (string $id, Channel $channel) {
$timeToSleep = $channel->recv();
echo
"Hello from $id. I will sleep for $timeToSleep second(s).".PHP_EOL;
sleep($timeToSleep);
echo
"$id slept for $timeToSleep second(s).".PHP_EOL;
return
$timeToSleep;
}, [
$id, $channel]);
},
$taskIds);

wait($consumerFutures);
wait([$producerFuture]);
}

function
wait(array $futures)
{
return
array_map(function ($future) {
return
$future->value();
},
$futures);
}
up
2
rustysun
4 years ago
an example used unbuffered channel.
<?php

use parallel\{Channel,Runtime};

$sum=function(array $a, Channel $ch) {
$sum=0;
foreach (
$a as $v) {
$sum+=$v;
}
$ch->send($sum);
};
try {
$a=[7, 2, 8, 1, 4, 0, 9, 10];
//unbuffered channel
$runtime=new Runtime;
$ch2=new Channel;
$runtime->run($sum, [array_slice($a, 0, $num), $ch2]);
$runtime->run($sum, [array_slice($a, $num), $ch2]);
//receive from channel
$x=$ch2->recv();
$y=$ch2->recv();
$ch2->close();
echo
"\nch2:", $x, "\t", $y, "\t", $x + $y, "\n";
} catch(
Error $err) {
echo
"\nError:", $err->getMessage();
} catch(
Exception $e) {
echo
"\nException:", $e->getMessage();
}

//output
//ch2:18 23 41
up
2
gam6itko
2 years ago
<?php

// the very weird way to calculate factorial ^_^
// we create one thread and synching them with buffered channel
// at fact only one thread is executing at the time

use parallel\{Channel, Future, Runtime};

for (
$n = 0; $n <= 10; $n++) {
echo
"!$n = " . factorial($n) . PHP_EOL;
}

/**
* Creates $n threads.
*/
function factorial(int $n): int
{
// buffered channel - using for sync threads ^_^
$channel = new Channel(1);
$futureList = [];
for (
$i = 2; $i <= $n; $i++) {
$runtime = new Runtime();
$futureList[] = $runtime->run(
static function (
Channel $channel, $multiplier): void {
$f = $channel->recv();
$channel->send($f * $multiplier);
},
[
$channel, $i]
);
}

$channel->send(1);

// waiting until all threads are done
do {
$allDone = array_reduce(
$futureList,
function (
bool $c, Future $future): bool {

return
$c && $future->done();
},
true
);
} while (
false === $allDone);

return
$channel->recv();
}

// output:
// !0 = 1
// !1 = 1
// !2 = 2
// !3 = 6
// !4 = 24
// !5 = 120
// !6 = 720
// !7 = 5040
// !8 = 40320
// !9 = 362880
// !10 = 3628800
up
1
rustysun
4 years ago
<?php
use parallel\Channel;

function
sum(array $a, Channel $ch) {
$sum=0;
foreach (
$a as $v) {
$sum+=$v;
}
$ch->send($sum);
}

try {
$a=[7, 2, 8, 1, 4, 0, 9, 10];
$ch1=Channel::make('sum', 2);
$ch2=new Channel;
$num=count($a) / 2;
sum(array_slice($a, 0, $num), $ch1);
sum(array_slice($a, $num), $ch1);

//receive from channel
$x=$ch1->recv();
$y=$ch1->recv();
$ch1->close();
echo
"\nch1:", $x, "\t", $y, "\t", $x + $y, "\n";
} catch(
Error $err) {
echo
"\nError:", $err->getMessage();
} catch(
Exception $e) {
echo
"\nException:", $e->getMessage();
}
To Top