Replicated Distribution

View replicated.chpl on GitHub

This primer demonstrates uses of replicatedDist, the Replicated Distribution.

Warning

The Replicated Distribution is currently unstable. Its functionality is likely to change in the future.

To use this distribution in a Chapel program, the following module must be used:

use ReplicatedDist;

Preface

For this example, we’ll create a distributed domain and array and then initialize it just to give a brief flavor of how the distribution maps across locales. Running this example on 6 locales does a nice job of illustrating the distribution characteristics.

Like other distributions, replicatedDist supports options to map to a different virtual locale grid than the one used by default (a multidimensional factoring of the built-in Locales array), as well as to control the amount of parallelism used in data parallel loops. For more details, see the documentation on Standard Distributions.

Make the program size configurable from the command line.

config const n = 8;

Declare a 2-dimensional domain Space that we will later use to initialize the distributed domains.

const Space = {1..n, 1..n};

Replicated Distribution

The replicatedDist distribution is different from other distributions: each of the original domain’s indices is replicated onto each locale, as are the corresponding array elements. For example, a domain {1..3} distributed using replicatedDist will store three indices per locale that the distribution is targeting (by default, all locales). Similarly, an array declared over that domain will store three elements per locale. Each locale’s copy of the domain or array is known as its replicand.

Consistency among these array replicands is NOT maintained automatically; users who want a replicated array to store the same values on every target locale will have to manage that consistency themselves.

In general, operations on replicated domains and arrays only refer the local replicand. The primary exception to this rule is the re-assignment of a replicated domain’s indices. In this case, the copy of the domain on each locale will be updated (and any arrays over the domain will be reallocated on each locale).

Here’s a declaration of a replicated domain and array:

const ReplicatedSpace = Space dmapped replicatedDist();
var RA: [ReplicatedSpace] int;

Queries about the size of a replicated domain or array will return the size per locale:

writeln("Replicated Array has ", RA.size, " elements per locale");

The following loop-based assignment to RA only affects the copy of ‘RA’ on the locale on which it’s running (the last locale, in this example). All other copies will remain in their default-initialized form.

on Locales[numLocales-1] do
  forall ra in RA do
    ra = here.id;

Similarly, when reading the array, only the local copy will be accessed. Thus, when running on more than one locale, the following writeln() will not see the modification performed by the loop above since the two statements are executed on distinct locales:

writeln("Locale 0's copy of RA is:\n", RA);

To access the replicands owned by other locales, we can use the replicand method, which takes a locale as an argument and returns the array local to that locale:

writeln("Locale ", numLocales-1, "'s copy of RA is:\n",
         RA.replicand(Locales[numLocales-1]));

Alternatively, we can print the replicated array from that locale:

on Locales[numLocales-1] do
  writeln("Locale ", numLocales-1, "'s copy of RA is:\n", RA);

In order to see the replicands owned by all the locales, let’s define and call a little utility function:

proc writeReplicands(X) {
  for loc in Locales {
    writeln(loc, ":");
    writeln(X.replicand(loc));
  }
}

writeln("Replicated Array Index Map");
writeReplicands(RA);
writeln();

Whole-array assignment is similarly local only to the current locale’s copy of the array:

var A: [Space] int = [(i,j) in Space] i*100 + j;
RA = A;
writeln("Replicated Array after whole-array assignment:");
writeReplicands(RA);
writeln();

Here, we have each locale update its own copy of RA to store its locale ID, which results in a modification to each replicand:

coforall loc in Locales do on loc do
  RA = loc.id;

writeln("Replicated Array after assigning on each locale:");
writeReplicands(RA);
writeln();

The following examples simply demonstrate that only the local replicand is accessed when an individual element is read…

on Locales[0] do
  writeln("on ", here, ": ", RA(Space.low));
on Locales[LocaleSpace.high] do
  writeln("on ", here, ": ", RA(Space.low));
writeln();

…or written:

on Locales[LocaleSpace.high] do
  RA(Space.low) = 7777;

writeln("Replicated Array after being indexed into");
writeReplicands(RA);
writeln();

…or the whole array is read:

on Locales[LocaleSpace.high] do
  A = RA + 4;
writeln("Non-Replicated Array after assignment from Replicated Array + 4");
writeln(A);
writeln();