GPU CUDA and MEX Programming

Further accelerate your code using advanced GPU CUDA^® and MEX programming

If running MATLAB^® functions on the GPU does not sufficiently speed up your code, or if you need to use advanced GPU CUDA features, you can write your own CUDA code and run it in MATLAB by generating an executable MEX file using mexcuda or an executable kernel using parallel.gpu.CUDAKernel.

Generally, using MEX files is more flexible than using CUDAKernel objects. For more information, see Run MEX Functions Containing CUDA Code.

Functions

expand all

Key Functions and Classes

`mexcuda`	Compile MEX functions or PTX files for GPU computation
`CUDAKernel`	Kernel executable on GPU
`feval`	Evaluate kernel on GPU
`setConstantMemory`	Set some constant memory on GPU

C Functions and Classes

`mxGPUArray`	Type for MATLAB gpuArray
`mxGPUCopyFromMxArray`	Copy mxArray to mxGPUArray
`mxGPUCopyGPUArray`	Duplicate (deep copy) mxGPUArray object
`mxGPUCopyImag`	Copy imaginary part of mxGPUArray
`mxGPUCopyReal`	Copy real part of mxGPUArray
`mxGPUCreateComplexGPUArray`	Create complex GPU array from two real gpuArrays
`mxGPUCreateFromMxArray`	Create read-only mxGPUArray object from input mxArray
`mxGPUCreateGPUArray`	Create mxGPUArray object, allocating memory on GPU
`mxGPUCreateMxArrayOnCPU`	Create mxArray for returning CPU data to MATLAB with data from GPU
`mxGPUCreateMxArrayOnGPU`	Create mxArray for returning GPU data to MATLAB
`mxGPUDestroyGPUArray`	Delete mxGPUArray object
`mxGPUGetClassID`	mxClassID associated with data on GPU
`mxGPUGetComplexity`	Complexity of data on GPU
`mxGPUGetData`	Raw pointer to underlying data
`mxGPUGetDataReadOnly`	Read-only raw pointer to underlying data
`mxGPUGetDimensions`	mxGPUArray dimensions
`mxGPUGetNumberOfDimensions`	Size of dimension array for mxGPUArray
`mxGPUGetNumberOfElements`	Number of elements on GPU for array
`mxGPUIsSame`	Determine if two mxGPUArrays refer to same GPU data
`mxGPUIsSparse`	Determine if mxGPUArray contains sparse GPU data
`mxGPUIsValidGPUData`	Determine if mxArray is pointer to valid GPU data
`mxGPUSetDimensions`	Modify number of dimensions and size of each dimension
`mxIsGPUArray`	Determine if mxArray contains GPU data
`mxInitGPU`	Initialize MATLAB GPU library on currently selected device

Topics

Run MEX Functions Containing CUDA Code
All MEX files, including those containing CUDA code, have a single entry point known as mexFunction.
Run CUDA or PTX Code on GPU
This page explains how to create an executable kernel from a CUDA C++ source file (CU) file and run that kernel on a GPU in MATLAB.

Featured Examples

Accessing Advanced CUDA Features Using MEX

How advanced features of the GPU can be accessed using MEX files. It builds on the example Stencil Operations on a GPU. The previous example uses Conway's "Game of Life" to demonstrate how stencil operations can be performed using MATLAB® code that runs on a GPU. The present example demonstrates how you can further improve the performance of stencil operations using an advanced feature of the GPU: shared memory. You do this by writing your own CUDA® code in a MEX file and calling the MEX file from MATLAB. You can find an introduction to the use of the GPU in MEX files in Run MEX-Functions Containing CUDA Code.

Open Live Script

Illustrating Three Approaches to GPU Computing: The Mandelbrot Set

Adapt your MATLAB® code to compute the Mandelbrot Set using a GPU.

Open Live Script