codistributed - Access elements of arrays distributed among workers in parallel pool - MATLAB (original) (raw)
Access elements of arrays distributed among workers in parallel pool
Description
Arrays partitioned among the workers in a parallel pool are accessible from the workers as codistributed array objects. Codistributed array objects on workers that you create inside spmd statements or from within task functions of communicating jobs are accessed as distributed arrays on the client.
Creation
Use either the codistributed function, or the codistributed.build function to create a codistributed object.
Syntax
Description
[C](#mw%5F4d261fd2-cd09-4866-97bd-4db57d6ae163) = codistributed([X](#mw%5Fa86d67d7-6bcb-451b-a9cb-58bcfb6a1bbf)) distributes a replicated array X using the default distribution scheme and returns a codistributed array C. X must be a replicated array, that is, it must have the same value on all workers. The size of array C is the same as the size of X.
[C](#mw%5F4d261fd2-cd09-4866-97bd-4db57d6ae163) = codistributed([X](#mw%5Fa86d67d7-6bcb-451b-a9cb-58bcfb6a1bbf),[workerIndex](#mw%5F45faf75a-eafb-43a2-853b-c395fb3da15d)) distributes a local array X that resides on the worker identified byworkerIndex, using the default codistributor. The local arrayX must be defined on all workers, but the function uses only theWorkerIndex to construct C.size(C) is the same as the size of X.
[C](#mw%5F4d261fd2-cd09-4866-97bd-4db57d6ae163) = codistributed([X](#mw%5Fa86d67d7-6bcb-451b-a9cb-58bcfb6a1bbf),[codist](#mw%5F01a18da1-f827-47f2-981c-a3b612ef1480)) distributes a replicated array X using the distribution scheme defined by codistributor codist. X must be a replicated array, that is, it must have the same value on all workers. The size ofC is the same as the size of X. For information on constructing codistributor objects, see codistributor1d and codistributor2dbc.
[C](#mw%5F4d261fd2-cd09-4866-97bd-4db57d6ae163) = codistributed([X](#mw%5Fa86d67d7-6bcb-451b-a9cb-58bcfb6a1bbf),[workerIndex](#mw%5F45faf75a-eafb-43a2-853b-c395fb3da15d),[codist](#mw%5F01a18da1-f827-47f2-981c-a3b612ef1480)) distributes a local array X that resides on the worker identified byworkerIndex, using the codistributor codist. The local array X must be defined on all workers, but the function uses only the WorkerIndex to construct C.size(C) is the same as the size of X.
[C](#mw%5F4d261fd2-cd09-4866-97bd-4db57d6ae163) = codistributed([C1](#mw%5Fe7ba16b8-0ffb-4eeb-8b78-169b0180a328),[codist](#mw%5F01a18da1-f827-47f2-981c-a3b612ef1480)) accepts an array C1 that is already codistributed, and redistributes it into C according to the distribution scheme defined by the codistributor codist. This syntax is equivalent to C = redistribute(C1,codist). If the existing distribution scheme forC1 is the same as that specified in codist, then the result C is the same as the input C1.
Input Arguments
X — Array to distribute
array
Array to distribute, specified as an array.
codist — Codistributor
codistributor1d object | codistributor2dbc object
Codistributor, specified as a codistributor1d orcodistributor2dbc object. For information on creating codistributors, see codistributor1d and codistributor2dbc. To use the default distribution scheme, you can specify a codistributor constructor without arguments.
workerIndex — Index of worker
positive integer scalar
Index of the worker where the local array is stored, specified as a positive integer scalar.
Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64
C1 — Codistributed array to redistribute
codistributed array
Codistributed array to redistribute, specified as a codistributed array.
Output Arguments
C — Codistributed array
codistributed array
Codistributed array stored on parts of workers in a parallel pool.
Object Functions
| gather | Transfer distributed array, Composite object, orgpuArray object to local workspace |
|---|---|
| getCodistributor | Codistributor object for existing codistributed array |
| getLocalPart | Local portion of codistributed array |
| globalIndices | Global indices for local part of codistributed array |
| Inf | Create codistributed array of all Inf values |
| iscodistributed | True for codistributed array |
| redistribute | Redistribute codistributed array with another distribution scheme |
The object functions for codistributed arrays are too numerous to list here. Most resemble and behave the same as built-in MATLAB® functions. For a complete list of object functions, see Run MATLAB Functions with Distributed Arrays.
Among the object functions there are several for examining the characteristics of the array itself. Most behave like the MATLAB functions of the same name.
Examples
Create Codistributed Arrays
Create a 1000-by-1000 codistributed array C1 using the default distribution scheme.
parpool('Processes',4)
spmd
N = 1000;
X = magic(N);
C1 = codistributed(X);
end
Array X is replicated on every worker, and arrayC1 is partitioned among the workers.
Create Codistributed Arrays with Distribution Scheme
Create a 1000-by-1000 codistributed array C2, distributed by rows (over the first dimension) with the codistributor1d distribution scheme.
spmd N = 1000; X = magic(N); C2 = codistributed(X,codistributor1d(1)); end
Tips
The gather function performs the inverse ofcodistributed. Use the gather function to retrieve codistributed arrays into the client work space.
Version History
Introduced in R2008b