>>help gpu

There are several options available for using your computer's graphics

processing unit (gpu) for matrix operations.

- Transfer data between the MATLAB workspace and the gpu

- Evaluate individual MATLAB functions that have been

overloaded for execution on the gpu

- Execute MATLAB code containing multiple functions using ARRAYFUN.

(Not all MATLAB functions are supported.)

- Create kernels from CU files for execution on the gpu

The gpu Computing section of the Parallel Computing Toolbox User's Guide

provides more information on these use cases and lists supported devices

and device drivers.

Data Transfer Operations

gpuArray - Transfer an array from the MATLAB workspace to the gpu

gather - Transfer an array from the gpu to the MATLAB workspace

MATLAB Overloads

MATLAB functions that have been made available for

execution on the gpu can be viewed using the command

methods('gpuArray')

Certain other MATLAB functions that support gpuArray inputs are

listed in the Parallel Computing Toolbox documentation.

Examples:

% FFT

A = gpuArray( rand( 2^16, 100 ) );

F = fft(A)

% MLDIVIDE

A = gpuArray( rand(1024) ); B = gpuArray( rand(1024,1) );

X = A\B

% MTIMES

A = gpuArray( rand(1024) ); B = gpuArray( rand(1024) );

C = A*B

Execute MATLAB code on the gpu

gpuArray/arrayfun - Apply a function to each element of an array on

the gpu

The function to evaluate on the gpu must exist on the path. Only a

subset of the MATLAB language is supported by ARRAYFUN on the gpu.

The restrictions are listed in the User's Guide.

Example:

% The file xycrull.m is one example of an existing MATLAB file

% that can be automatically executed on the gpu.

%

% Execute 'type xycrull' at the MATLAB prompt to view the contents

% of the file.

%

gt = gpuArray(rand(400));

[o1, o2] = arrayfun(@xycrull, gt)

CUDA Kernel Operations

parallel.gpu.CUDAKernel - Create a kernel object that corresponds

to a particular kernel in a CU file

parallel.gpu.CUDAKernel/feval - Evaluate a kernel on the gpu

Device Information

gpuDeviceCount - Return the number of gpu devices available

gpuDevice - Query or select a gpu device

Timing gpu Operations

gputimeit - Measure time required to run function on gpu