CyclicDist

Usage

use CyclicDist;

or

import CyclicDist;
record cyclicDist: writeSerializable

The cyclicDist distribution uses a round-robin partitioning to map d-dimensional indices to a d-dimensional array of locales, starting from a given index.

Warning

The cyclicDist distribution was, until recently, a class named Cyclic. Today, Cyclic is still supported in a deprecated form, yet is an alias to the cyclicDist record here. In our experience, most uses of Cyclic in distribution contexts should continue to work, but updating to cyclicDist is requested going forward due to the deprecation.

More precisely, for a cyclicDist distribution with:

rank

d

start index

(s_1, ...., s_d)

over locales

targetLocales: [0..<N_1, ...., 0..<N_d] locale

Each index (i_1, ...., i_d) is mapped to the locale targetLocales[j_1, ...., j_d] where, for each k in 1..d, we have:

j_k = (i_k - s_k) (mod N_k)

Example

The following code declares a domain D distributed using a cyclicDist distribution with a start index of (1,1), and declares an array A over that domain. The forall loop sets each array element to the ID of the locale to which it is mapped.

use CyclicDist;

const Dist = new cyclicDist(startIdx=(1,1));
const D = Dist.createDomain({1..8, 1..8});
var A: [D] int;

forall a in A do
  a = here.id;

writeln(A);

When run on 6 locales, the output is:

0 1 0 1 0 1 0 1
2 3 2 3 2 3 2 3
4 5 4 5 4 5 4 5
0 1 0 1 0 1 0 1
2 3 2 3 2 3 2 3
4 5 4 5 4 5 4 5
0 1 0 1 0 1 0 1
2 3 2 3 2 3 2 3

Data-Parallel Iteration

As demonstrated by the above example, a forall loop over a cyclicDist-distributed domain or array executes each iteration on the locale owning the index in question.

By default, parallelism within each locale is applied to that locale’s local, strided block of indices by creating a task for each available processor core (or the number of local indices if it is less than the number of cores). The local domain indices are then statically divided as evenly as possible between those tasks.

Initializer Arguments

The cyclicDist initializer is defined as follows:

proc cyclicDist.init(
  startIdx,
  targetLocales: [] locale = Locales)

The argument startIdx is a tuple of integers defining an index that will be distributed to the first locale in targetLocales. In the 1-dimensional case, startIdx can be an integer or a tuple with a single element.

The argument targetLocales is an array containing the target locales to which this distribution maps indices and data. The rank of targetLocales must match the rank of the distribution, or be 1. If the rank of targetLocales is 1, a greedy heuristic is used to reshape the array of target locales so that it matches the rank of the distribution and each dimension contains an approximately equal number of indices.

Convenience Factory Methods

It is common for a cyclicDist-distributed domain or array to use its first index as the start Index in a Cyclic distribution. It is also common not to override any of the other defaulted initializer arguments. In such cases, factory methods are provided for convenience.

These methods take a domain or series of ranges as arguments and return a cyclic-distributed domain or array. For example, the following declarations create new 5 x 5 cyclic-distributed domains and arrays using (1, 1) as the starting index:

use CyclicDist;

var CyclicDom1 = cyclicDist.createDomain({1..5, 1..5});
var CyclicArr1 = cyclicDist.createArray({1..5, 1..5}, real);
var CyclicDom2 = cyclicDist.createDomain(1..5, 1..5);
var CyclicArr2 = cyclicDist.createArray(1..5, 1..5, real);

The helper methods on Cyclic have the following signatures:

proc type createDomain(dom: domain(?), targetLocales = Locales)

Create a cyclic-distributed domain. The lower bounds of the domain are used as the starting indices.

proc type createDomain(rng: range(?)..., targetLocales = Locales)

Create a cyclic-distributed domain from a series of ranges. The lower bounds of the ranges are used as the starting indices.

proc type createArray(dom: domain(?), type eltType, targetLocales = Locales)

Create a default-initialized cyclic-distributed array whose indices match those of the given domain.

proc type createArray(rng: range(?)..., type eltType, targetLocales = Locales)

Create a default-initialized cyclic-distributed array using a domain constructed from the series of ranges.

proc type createArray(dom: domain(?), type eltType, initExpr, targetLocales = Locales)

Create a cyclic-distributed array whose indices match those of the given domain.

The array’s values are initialized using initExpr which can be any of the following:

  • a value coercible to eltType — all elements of the array will be assigned with this value

  • an iterator expression with compatible size and type — the array elements will be initialized with the values yielded by the iterator

  • an array of compatible size and type — the array will be assigned into the distributed array

proc type createArray(rng: range(?)..., type eltType, initExpr, targetLocales = Locales)

Create a cyclic-distributed array using a domain constructed from the series of ranges.

The array’s values are initialized using initExpr which can be any of the following:

  • a value coercible to eltType — all elements of the array will be assigned with this value

  • an iterator expression with compatible size and type — the array elements will be initialized with the values yielded by the iterator

  • an array of compatible size and type — the array will be assigned into the distributed array

proc createDomain(dom: domain(?))

Create a cyclic-distributed domain over an existing cyclicDist by copying the index space from the passed domain.

proc createDomain(rng: range(?)...)

Create a cyclic-distributed domain from a series of ranges over an existing cyclicDist.

Limitations

This distribution has not been tuned for performance.