// Distributed memory task parallel hello world /* This program shows how to use Chapel’s task parallelism and locality features to create a concurrent “Hello, world” program that spans distributed memory `locales` (compute nodes) as well as (optionally) the processing units (cores) within each locale. The number of locales on which to run is specified on the executable’s command line using the `-nl` or `–numLocales` flag (e.g., “./hello -nl 64“). */ // // The following `config const` specifies whether or not the locale // name should be printed (for ease-of-testing purposes); the default // can be overridden on the execution command line (e.g., // “–printLocaleName=false“). // config const printLocaleName = true; // // This one specifies the number of tasks to use per locale: // config const tasksPerLocale = 1; // // Use a `coforall-loop` to create a distinct task per locale on which // the program is executing. Here, we’re iterating over the built-in // `Locales` array which stores an array of locale values // corresponding 1:1 with the system’s compute nodes on which the // program is executing. Thus, each iteration corresponds to one of // the locales, represented by the loop index variable `loc`. // coforall loc in Locales { // // Migrate each task to its corresponding locale, `loc`. This is done // using an `’on’-clause`, which moves execution of the current task // to the locale corresponding to the expression following it. // on loc { // // Now use a second coforall-loop to create a number of tasks // corresponding to the `tasksPerLocale` configuration constant. // Since this loop body doesn’t contain any on-clauses, all tasks will // remain local to the current locale. // coforall tid in 0..#tasksPerLocale { // // Start building up the message to print using a string // variable, `message`. // var message = “Hello, world! (from “; // // If we’re running more than one task per locale, specialize the // message based on the task ID. // if (tasksPerLocale > 1) then message += “task ” + tid:string + ” of ” + tasksPerLocale:string + ” on “; // // Specialize the message based on the locale on which this task is // running: // // * `here` refers to the locale on which this task is running (same as `loc`) // * `.id` queries its unique ID in 0..`numLocales`-1 // * `.name` queries its name (similar to UNIX “hostname“) // * `numLocales` refers to the number of locales (as specified by -nl) // message += “locale ” + here.id:string + ” of ” + numLocales:string; if printLocaleName then message += ” named ” + loc.name; // // Terminate the message // message += “)”; // // Print out the message. Since each message is being printed by a // distinct task, they may appear in an arbitrary order. // writeln(message); } } } // // For further examples of using task parallelism, refer to // :ref:`examples/primers/taskParallel.chpl `. // // For further examples of using locales and on-clauses, refer to // :ref:`examples/primers/locales.chpl `. //

// Distributed memory task parallel hello world

/* This program shows how to use Chapel’s task parallelism and
locality features to create a concurrent “Hello, world” program
that spans distributed memory `locales` (compute nodes) as well as
(optionally) the processing units (cores) within each locale. The
number of locales on which to run is specified on the executable’s
command line using the `-nl` or `–numLocales` flag (e.g.,
“./hello -nl 64“).
*/

//
// The following `config const` specifies whether or not the locale
// name should be printed (for ease-of-testing purposes); the default
// can be overridden on the execution command line (e.g.,
// “–printLocaleName=false“).
//
config const printLocaleName = true;

//
// This one specifies the number of tasks to use per locale:
//
config const tasksPerLocale = 1;

//
// Use a `coforall-loop` to create a distinct task per locale on which
// the program is executing. Here, we’re iterating over the built-in
// `Locales` array which stores an array of locale values
// corresponding 1:1 with the system’s compute nodes on which the
// program is executing. Thus, each iteration corresponds to one of
// the locales, represented by the loop index variable `loc`.
//
coforall loc in Locales {

//
// Migrate each task to its corresponding locale, `loc`. This is done
// using an `’on’-clause`, which moves execution of the current task
// to the locale corresponding to the expression following it.
//
on loc {

//
// Now use a second coforall-loop to create a number of tasks
// corresponding to the `tasksPerLocale` configuration constant.
// Since this loop body doesn’t contain any on-clauses, all tasks will
// remain local to the current locale.
//
coforall tid in 0..#tasksPerLocale {

//
// Start building up the message to print using a string
// variable, `message`.
//
var message = “Hello, world! (from “;

//
// If we’re running more than one task per locale, specialize the
// message based on the task ID.
//
if (tasksPerLocale > 1) then
message += “task ” + tid:string + ” of ” + tasksPerLocale:string + ” on “;

//
// Specialize the message based on the locale on which this task is
// running:
//
// * `here` refers to the locale on which this task is running (same as `loc`)
// * `.id` queries its unique ID in 0..`numLocales`-1
// * `.name` queries its name (similar to UNIX “hostname“)
// * `numLocales` refers to the number of locales (as specified by -nl)
//
message += “locale ” + here.id:string + ” of ” + numLocales:string;

if printLocaleName then message += ” named ” + loc.name;

//
// Terminate the message
//
message += “)”;

//
// Print out the message. Since each message is being printed by a
// distinct task, they may appear in an arbitrary order.
//
writeln(message);
}
}
}

//
// For further examples of using task parallelism, refer to
// :ref:`examples/primers/taskParallel.chpl `.
//
// For further examples of using locales and on-clauses, refer to
// :ref:`examples/primers/locales.chpl `.
//

Leave a comment

Your email address will not be published. Required fields are marked *


eight + = 17

Leave a Reply