Reductions

View reductions.chpl on GitHub

This primer includes example reductions.

An array is filled with random values and the locations of the maximum and minimum are found. The Euclidean norm of the array’s columns is computed using + reductions over slices of A. Finally, an && reduction is used to compute whether all values in A are greater than 0.25 and the results of the computations are printed.

For an example usage of the standard module Random, see its primer.

use Random; // For random number generation

config const seed = 31415; // Random generation seed
config const size = 10;    // The size of each side of the array

var A: [1..size, 1..size] real; // The 2D work array

Fill an array with random values

Fill A with random real values between 0 and 1. Uses the NPB random number generator for historical reasons.

fillRandom(A, seed);
writeln("A is: "); writeln(A);
writeln();

Find the average value of an array

We find the average value of the array by summing over its elements and dividing by the number of elements it contains.

var eltAvg = (+ reduce A) / size**2;
writeln("The average element of A has the value ", eltAvg);
writeln();

Find the 1-norm of an array

We can find the 1-norm of A by summing over the absolute value of the elements.

The expression abs(A) creates a new matrix which contains in each of its elements the absolute value of the corresponding element in A. The + reduce clause just sums these up.

var oneNorm = + reduce abs(A);
writeln("The 1-norm of A is ", oneNorm);
writeln();

The Frobenius norm

The Frobenius norm is the square root of the sum over all elements of their respective squares.

The expression below can be broken down thus:

  1. The Frobenius norm is the square root of some quantity.

  2. The quantity is the sum over all elements of a matrix.

  3. That matrix is the promotion of A by **2. That is, each of its elements is the square of the corresponding element of A.

var frobNorm = sqrt(+ reduce A**2);
writeln("The Frobenius norm of A is ", frobNorm);
writeln();

maxloc and minloc reductions

Apply minloc and maxloc reductions. We capture the results into the maxVal, maxLoc, minVal, minLoc variables.

maxloc and minloc reductions expect a 2-tuple argument that can be iterated over using zippered iteration. They produce a 2-tuple result. The first component of the result is the maximum (or minimum) over the first component of the argument. The second component of the result indicates its location, i.e. the corresponding element of the second component of the argument.

var (maxVal, maxLoc) = maxloc reduce zip(A, A.domain);
var (minVal, minLoc) = minloc reduce zip(A, A.domain);
writeln("The maximum value in A is: A", maxLoc, " = ", maxVal);
writeln("The minimum value in A is: A", minLoc, " = ", minVal);
writeln("The difference is: ", maxVal - minVal);
writeln();

The Euclidean norm

Compute Euclidean norms for each column using + reductions.

Breaking down the statement below:

  1. vecNorms is a 1-D array containing size elements (indexed by 1..size).

  2. The elements of vecNorms are the square-root of some quantity.

  3. The quantity is the sum over all of the elements of some vector.

  4. The vector consists of the promotion of the j-th column of A by **2. That is, each element of that column vector is squared.

var vecNorms = [j in 1..size] sqrt(+ reduce A(1..size, j)**2);
writeln("The Euclidean norm of each column is: ", vecNorms);
writeln();

&& reduction

Use the && reduction to compute if all values in A are greater than 0.25.

The parenthesized value is the promotion of A by > 0.25. This yields an array of the same size as A, containing boolean values that are true if the corresponding element in A exceeds 0.25 and false otherwise.

The clause && reduce computes the result of applying the Boolean AND operator between all elements of the promoted array.

var onlyBigValues = && reduce (A > 0.25);
if onlyBigValues then
  writeln("The values in A are all greater than 0.25");
else
  writeln("Some values in A are less than 0.25");
writeln();