eye - Create codistributed identity matrix

eye - Create codistributed identity matrix - MATLAB

Create codistributed identity matrix

Syntax

Description

cI = eye(n,codist) creates an n-by-n codistributed identity matrix and uses codist to specify the distribution of the array values across the workers.

Specify codist as "codistributed" to use the default codistributor1d distribution scheme or the distribution scheme defined by a codistributor1d or codistributor2dbc object.

When you create the codistributed array in a communicating job or spmd block, the function creates an array on each worker. If you create a codistributed array outside of a communicating job or spmd block, the array is stored only on the worker or client that creates the codistributed array.

By default, the codistributed array has the underlying type double.

example

cI = eye(n,m,codist) creates an n-by-m codistributed identity matrix with ones on the main diagonal and zeros elsewhere. For example, eye(2,3,"codistributed") creates a 2-by-3 codistributed array.

example

cI = eye(sz,codist) creates a codistributed identity matrix where the size vector sz defines the size of cI. For example, eye([2 3],"codistributed") also creates a 2-by-3 codistributed identity matrix.

example

cI = eye(___,datatype,codist) creates a codistributed identity matrix with the underlying data type datatype. For example, eye(1,"int8","codistributed") creates a codistributed 8-bit scalar integer 1. You can use this syntax with any of the size arguments in the previous syntaxes. You must specify codist after the array size and data type arguments.

example

cI = eye(___,"noCommunication") creates a codistributed identity matrix without using communication between workers.

When you create very large arrays or your communicating job or spmd block uses many workers, worker-worker communication can slow down array creation. Use this syntax to improve the performance of your code by removing the time required for worker-worker communication.

Tip

When you use this syntax, some error checking steps are skipped. Use this syntax to improve the performance of your code after you prototype your code without specifying "noCommunication".

You must specify "noCommunication" after the size, data type and codist arguments.

cI = eye(___,like=p) uses the array p to create a codistributed identity matrix. You can also specify "noCommunication" as part of the function call.

The returned array cI has the same underlying type, sparsity, and complexity (real or complex) as p.

Examples

collapse all

Create a 1000-by-1000 codistributed identity matrix, distributed using the default distribution scheme.

spmd(4)
    C = eye(1000,"codistributed");
end

With four workers, each worker contains a 1000-by-250 local piece of C.

Create a 1000-by-1000 codistributed uint16 identity matrix, distributed by its columns. Use a codistributor1d object to define the distribution scheme.

spmd(4)
    codist = codistributor1d(2,100*[1:spmdSize]);
    C = eye(1000,1000,"uint16",codist)
end

Worker 1: 
  This worker stores C(:,1:100).
          LocalPart: [1000x100 uint16]
      Codistributor: [1x1 codistributor1d]
Worker 2: 
  This worker stores C(:,101:300).
          LocalPart: [1000x200 uint16]
      Codistributor: [1x1 codistributor1d]
Worker 3: 
  This worker stores C(:,301:600).
          LocalPart: [1000x300 uint16]
      Codistributor: [1x1 codistributor1d]
Worker 4: 
  This worker stores C(:,601:1000).
          LocalPart: [1000x400 uint16]
      Codistributor: [1x1 codistributor1d]

Each worker contains a 1000-by-100*spmdIndex local piece of C.

You can also get the same codistributed array specifying the size vector.

spmd(4)
    C = eye([1000 1000],"uint16",codist);
end

First, create a codistributed array. Specify the underlying data type as single and use the default distribution scheme.

spmd(4)
p = eye(1000,"single","codistributed");
end

Create a new codistributed array that is the same size and data type as p. Use the 2-D block-cyclic codistributor2dbc function to define the distribution scheme.

spmd
C = eye(size(p),codistributor2dbc,"noCommunication",like=p);
end

Input Arguments

collapse all

Size of the first dimension of the identity matrix, specified as an integer.

If n is the only integer input argument, then cI is a square n-by-n codistributed identity matrix.
If n is 0, then cI is an empty matrix.
If n is negative, then the function treats it as 0.

Distribution scheme for codistributed array, specified as one of these options:

"codistributed" — Uses the default distribution scheme defined by the default codistributor1d object.
codistributor1d object — Uses the one-dimensional distribution scheme defined in a codistributor1d object. To use the default 1-D distribution scheme, you can specify the codistributor1d function without arguments.
codistributor2dbc object — Uses the two-dimensional block-cyclic distribution scheme defined in a codistributor2dbc object. To use the default 2-D block-cyclic distribution scheme, you can specify the codistributor2dbc function without arguments.

Size of the second dimension of the codistributed identity matrix, specified as an integer.

If m is 0, then cI is an empty matrix.
If m is negative, then the function treats it as 0.

Size of each dimension, specified as an integer row vector. Each element of this vector indicates the size of the corresponding dimension:

If the size of any dimension is 0, then cI is an empty array.
If the size of any dimension is negative, then the function treats it as 0.
Beyond the second dimension, eye ignores trailing dimensions with a size of 1. For example, eye([3 1 1 1],"codistributed") produces a 3-by-1 codistributed identity matrix.

Example: sz = [2 3 4] creates a 2-by-3-by-4 codistributed array.

Underlying data type of the returned array, specified as one of these options:

Prototype of array to create, specified as a codistributed array.

If you do not specify the datatype, the resulting codistributed array has the same underlying type, sparsity, and complexity (real or complex) as p.

Version History

Introduced in R2006b