MATLAB PARALLEL COMPUTING TOOLBOX 4 Guide de l'utilisateur

Parallel Computing Toolbox™ Release Notes

x ContentsGeneric Scheduler Interface Enhancements . . . . . . . . . . . . 10-2Decode Functions Provided with Product . . . . . . . . . . . . . . 10-2

R2008bVersion: 4.0New FeaturesBug FixesCompatibility Considerations

R2008b14-2MATLAB Compiler Product Support for Parallel Computing ToolboxApplicationsThis release offers the ability to convert Parallel Computing Tool

14-3Compatibility ConsiderationsBecause spmd is a new keyword, it will conflict with any user-defined functions orvariables of the same name. If you

R2008b14-4• DestroyTaskFcn• CancelJobFcn• CancelTaskFcnNew toolbox functions to accommodate this ability are:• getJobSchedulerData• setJobSchedulerDat

14-5For more information about this option and others, see the matlabpool reference page.

R2008aVersion: 3.3New FeaturesBug FixesCompatibility Considerations

R2008a15-2Renamed Functions for Product Name ChangesAs of result of the product name changes, some function names are changing in thisrelease.Compatib

15-3memory allocation was approximately 50 MB. The new higher limits are approximately600MB on 32-bit systems and 2GB on 64-bit systems. See Object

xiR2009b New Distributed Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-2Renamed codistributor Functions . . . . . . . . . . . .

R2008a15-4findResource Now Sets Properties According to ConfigurationThe findResource function now sets the properties on the object it creates accord

15-5dfeval Now Destroys Its Job When FinishedWhen finished performing its distributed evaluation, the dfeval function now destroysthe job it created.

R2007bVersion: 3.2New FeaturesBug FixesCompatibility Considerations

R2007b16-2New Parallel for-Loops (parfor-Loops)New parallel for-loop (parfor-loop) functionality automatically executes a loop bodyin parallel on dyna

16-3Compatibility ConsiderationsThis new feature has no impact on how configurations are used in a program, only onhow configurations are created and

R2007aVersion: 3.1New FeaturesBug FixesCompatibility Considerations

R2007a17-2Local Scheduler and WorkersA local scheduler allows you to schedule jobs and run up to four workers or labs on asingle MATLAB client machine

17-3Vectorized Task CreationThe createTask function can now create a vector of tasks in a single call when youprovide a cell array of cell arrays for

xii ContentsR2008b MATLAB Compiler Product Support for Parallel ComputingToolbox Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . .

R2007a17-4function returns jobs sequenced by their ID, unless otherwise specified. This changemakes job manager behavior consistent with the behavior

17-5Compatibility ConsiderationsIn the previous version of the toolbox, the distributor function was used to define howan array was distributed. In m

R2006bVersion: 3.0New FeaturesBug FixesCompatibility Considerations

R2006b18-2Support for Windows Compute Cluster Server (CCS)Distributed Computing Toolbox software and MATLAB Distributed Computing Engine™software now

18-3efficient use of memory and faster processing, especially for large data sets. For moreinformation, see “Distributed Arrays and SPMD”.parfor: Par

R2006b18-4stopworkermdce_def.batUNIX and Macintosh Utilities RenamedIn previous versions of the distributed computing products, the MDCE utilities for

18-5Tasks property of the job contains four task objects. The first task in the job's Tasksproperty corresponds to the task run by the lab whose

xiiiparfor Syntax Has Single Usage . . . . . . . . . . . . . . . . . . . . . . 15-4Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xiv ContentsEnhanced MATLAB Functions . . . . . . . . . . . . . . . . . . . . . . . . 17-4darray Function Replaces distributor Function . . . . . . .

R2015aVersion: 6.6New FeaturesBug FixesCompatibility Considerations

R2015a1-2Support for mapreduce function on any cluster that supports parallelpoolsYou can now run parallel mapreduce on any cluster that supports a pa

1-3cdf2rdfgpuArray.freqspacehistcountsidivideinpolygonisdiagishermitianissymmetricistrilistriulegendrenonzerosnthrootpinvplanerotpolypolyareapolyderp

R2015a1-4information about cluster discovery, see “Discover Clusters”. For information aboutconfiguring and verifying the required DNS SRV record on y

1-5Improved profiler accuracy for GPU codeThe MATLAB profiler now reports more accurate timings for code running on a GPU. Forrelated information, se

How to Contact MathWorksLatest news:www.mathworks.comSales and services:www.mathworks.com/sales_and_servicesUser community:www.mathworks.com/matlabcen

R2015a1-6Compatibility ConsiderationsCalling matlabpool now generates an error. You should instead use parpool to create aparallel pool.Upgrade parall

R2014bVersion: 6.5New FeaturesBug FixesCompatibility Considerations

R2014b2-2Parallelization of mapreduce on local workersIf you have Parallel Computing Toolbox installed, and your default cluster profilespecifies a lo

2-3pagefun support for mldivide on GPUsFor gpuArray inputs, pagefun is enhanced to support @mldivide for square matrixdivisors of sizes up to 32-by-3

R2014b2-4Data Analysis on Hadoop clusters using mapreduceParallel Computing Toolbox and MATLAB Distributed Computing Server support theuse of Hadoop®

2-5Compatibility ConsiderationsGPU devices on 32-bit Windows machines are still supported in this release, but in afuture release support will be com

R2014aVersion: 6.4New FeaturesBug FixesCompatibility Considerations

R2014a3-2Number of local workers no longer limited to 12You can now run a local cluster of more than 12 workers on your client machine. Unlessyou adju

3-3Note the following enhancements for some of these functions:• filter now supports IIR filtering.• pagefun is enhanced to support most element-wise

iiiContentsR2015a Support for mapreduce function on any cluster that supportsparallel pools . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

R2014a3-4Note the following enhancements for some of these functions:• ifft and randi are new in support of distributed and codistributed arrays.• ran

3-5Compatibility ConsiderationsCalling matlabpool continues to work in this release, but now generates a warning. Youshould instead use parpool to cr

R2013bVersion: 6.3New FeaturesBug FixesCompatibility Considerations

R2013b4-2parpool: New command-line interface (replaces matlabpool),desktop indicator, and preferences for easier interaction with a parallelpool of MA

4-3Asynchronous Function Evaluation on Parallel PoolYou can evaluate functions asynchronously on one or all workers of a parallel pool. Useparfeval t

R2013b4-4Icon color and tool tips let you know if the pool is busy or ready, how large it is, andwhen it might time out. You can click the icon to sta

4-5Compatibility ConsiderationsThe default preference setting is to automatically start a pool when a parallel languageconstruct requires it. If you

R2013b4-6interp2pagefunNote the following for some of these functions:• pagefun allows you to iterate over the pages of a gpuArray, applying @mtimes t

4-7More MATLAB functions enabled for distributed arrays: permute,ipermute, and sortrowsThe following functions now support distributed arrays with al

iv ContentsUpgrade parallel computing products together . . . . . . . . . . 1-6R2014b Parallelization of mapreduce on local workers . . . . . . . . .

R2013b4-8Enhancements to MATLAB functions enabled for GPUs, including ones,zerosThe following functions are enhanced in their support for gpuArray dat

4-9This new option is the default 'NormalTransform' setting when using thePhilox4x32-10 or Threefry4x64-20 generator. The following command

R2013b4-10Compatibility ConsiderationsIn R2013b, any use of gpuDevice to select a GPU with compute capability 1.3, generatesa warning. The device is s

R2013aVersion: 6.2New FeaturesBug Fixes

R2013a5-2GPU-enabled functions in Image Processing Toolbox and Phased ArraySystem ToolboxMore toolboxes offer enhanced functionality for some of their

5-3• arrayfun and bsxfun support indexing and accessing variables of outer functionsfrom within nested functions.• arrayfun supports singleton expans

R2013a5-4code files are necessary for their execution, then automatically attaches those files to theMATLAB pool job so that the code is available to

R2012bVersion: 6.1New FeaturesBug FixesCompatibility Considerations

R2012b6-2More MATLAB functions enabled for GPUs, including convn, cov, andnormest• “gpuArray Support” on page 6-2• “MATLAB Code on the GPU” on page 6-

6-3toolboxes, see their respective release notes. Parallel Computing Toolbox is required toaccess this functionality.Performance improvements to GPU-

vAdditional GPU enabled Image Processing Toolbox functions:bwdist, imreconstruct, iradon, radon . . . . . . . . . . . . . . . . 3-2Enhancements to GPU

R2012b6-4• mrdivide is now fully supported, and is no longer limited to accepting only scalarsfor its second argument.• sparse is now fully supported

6-5Command Window. Now this text is appended to the task’s Diary property as the textis generated, rather than waiting until the task is complete. Yo

R2012aVersion: 6.0New FeaturesBug FixesCompatibility Considerations

R2012a7-2New Programming InterfaceThis release provides a new programming interface for accessing clusters, jobs, andtasks.General Concepts and Phrase

7-3Previous Scheduler Objects New Cluster Objectslocalscheduler parallel.cluster.Locallsfscheduler parallel.cluster.LSFmpiexec parallel.cluster.Mpiex

R2012a7-4Functions and MethodsThis table compares some functions and methods of the old interface to those of the new.Many functions do not have a nam

7-5Previous Scheduler Properties New Cluster PropertiesIsUsingSecureCommunication HasSecureCommunicationSchedulerHostname,MasterName, HostnameHostPar

R2012a7-6help parallel.job.CJSCommunicatingJobhelp parallel.task.CJSTaskhelp parallel.job.MJSIndependentJobhelp parallel.job.MJSCommunicatingJobhelp p

7-7Creating and Finding Jobs and TasksIn the old interface, to create or find jobs or tasks, you could specify property nameand value pairs in struct

vi ContentsMore MATLAB functions enabled for distributed arrays:permute, ipermute, and sortrows . . . . . . . . . . . . . . . . . . . . 4-7Enhancement

R2012a7-8properties, from which the toolbox calculates which wrapper to use. For more informationabout these properties and their possible values, in

7-9The Cluster Profile Manager lets you create, edit, validate, import, and export profiles,among other actions. To open the Cluster Profile Manager,

R2012a7-10Function Descriptionparallel.exportProfile Export profiles to specified .settings fileProfiles in Compiled ApplicationsBecause compiled appl

7-11Note the following enhancements and restrictions to some of these functions:• GPUArray usage now supports all data types supported by MATLAB, exc

R2012a7-12g = gpuDevice(idx)...g = gpuDevice(idx) % Resets GPU device, clears dataTo deselect the current device, use gpuDevice([ ]) with an empty ar

7-13 0.8632 0.7481 0.9901 0.0420 0.2307 0.7008 0.7516 0.5059existsOnGPU(R) 1reset(g); % Resets GPU deviceexistsOnGPU(R)

R2012a7-14Set CUDA Kernel Constant MemoryThe new setConstantMemory method on the CUDAKernel object lets you set kernelconstant memory from MATLAB. For

R2011bVersion: 5.2New FeaturesBug FixesCompatibility Considerations

R2011b8-2New Job MonitorThe Job Monitor is a tool that lets you track the jobs you have submitted to a cluster.It displays the jobs for the scheduler

8-3Number of Local Workers Increased to TwelveYou can now run up to 12 local workers on your MATLAB client machine. If you do notspecify the number o

viiR2012b More MATLAB functions enabled for GPUs, including convn,cov, and normest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

R2011b8-4The following functions set the GPU random number generator seed and stream:parallel.gpu.rngparallel.gpu.RandStreamAlso, arrayfun called with

8-5 3 5 7 4 9 2To see that M is a GPUArray, use the whos or class function.MATLAB Code on the GPUGPU support is extended to i

R2011b8-6Compatibility ConsiderationsIf you have scripts or functions that use message identifiers that changed, you mustupdate the code to use the ne

8-7Compatibility ConsiderationsIn past releases, when there was no error, the Error property contained an MExceptionobject with empty data fields, ge

R2011aVersion: 5.1New FeaturesBug FixesCompatibility Considerations

R2011a9-2Deployment of Local WorkersMATLAB Compiler generated standalone executables and libraries from parallelapplications can now launch up to eigh

9-3subsasgnsubsindexsubsrefvertcatand all the plotting related functions.GPUArray IndexingBecause GPUArray now supports subsasgn and subsref, you can

R2011a9-4Distributed Array SupportNewly Supported FunctionsThe following functions are enhanced to support distributed arrays, supporting all formsof

9-5Search for Cluster Head Nodes Using Active DirectoryThe findResource function can search Active Directory to identify your cluster headnode. For m

viii ContentsCluster Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-8New Cluster Profile Manager . . . . . . . .

R2010bVersion: 5.0New FeaturesBug FixesCompatibility Considerations

R2010b10-2GPU ComputingThis release provides the ability to perform calculations on a graphics processing unit(GPU). Features include the ability to:•

10-3These functions are included on the workers' path. If your submit functionsmake use of the definitions in these decode functions, you do not

R2010b10-4Enhanced Example ScriptsThis release provides new sets of example scripts for using the generic schedulerinterface. As in previous releases,

10-5Enhanced Functions for Distributed Arraysqr Supports Distributed ArraysThe qr function now supports distributed arrays. For restrictions on this

R2010b10-6transpose and ctranspose Support 2dbcIn addition to their original support for 1-D distribution schemes, thefunctionsctranspose and transpos

R2010aVersion: 4.3New FeaturesBug FixesCompatibility Considerations

R2010a11-2New Save and Load Abilities for Distributed ArraysYou now have the ability to save distributed arrays from the client to a single MAT-file.S

11-3New Remote Startup of mdce ProcessNew command-line functionality allows you to remotely start up MATLAB DistributedComputing Server processes on

ixTask Error Properties Updated . . . . . . . . . . . . . . . . . . . . . . . . 8-6R2011a Deployment of Local Workers . . . . . . . . . . . . . . . .

R2009bVersion: 4.2New FeaturesBug FixesCompatibility Considerations

R2009b12-2New Distributed ArraysA new form of distributed arrays provides direct access from the client to data stored onthe workers in a MATLAB pool.

12-3Old Function Name New Function Namecodcolon codistributed.coloncodistributed(..., 'convert') codistributed(...)codistributed(...) witho

R2009b12-4Enhancements to Admin CenterAdmin Center has several small enhancements, including more conveniently locatedmenu choices, modified dialog bo

12-5• JobDescriptionFile — A string set to the name of the XML file defining a base statefor job creationCompatibility ConsiderationsCCS is now just

R2009aVersion: 4.1New FeaturesBug FixesCompatibility Considerations

R2009a13-2Number of Local Workers Increased to EightYou can now run up to eight local workers on your MATLAB client machine. If you do notspecify the

13-3the MATLAB Distributed Computing Server, see the online installation instructions athttp://www.mathworks.com/distconfig.New Benchmarking DemosNew