cell — Create empty cell arrays in MATLAB and RunMat.

cell constructs heterogeneous containers where each element starts as empty ([] by default). It supports scalar, multi-dimension, size-vector, and size(A) forms, plus 'like' prototypes, with behavior aligned to MATLAB and RunMat.

Syntax

C = cell()
C = cell(n)
C = cell(sz)
C = cell(m, n, ...)
C = cell("like", prototype)
C = cell(sz, "like", prototype)
C = cell(m, n, ..., "like", prototype)

Inputs

Returns

Errors

How cell works

• cell (with no arguments) returns a 0×0 empty cell array.
• cell(n) returns an n×n cell array whose elements default to the empty value.
• cell(m, n, p, ...) accepts any number of size scalars and returns an N-D cell array whose dimensions match the arguments.
• cell(sz) or cell(size(A)) accepts a numeric size vector of any length. A single-element vector produces an m×1 cell array; longer vectors create higher-dimensional cell arrays matching the provided extents.
• cell(___, 'like', prototype) copies the prototype's size when no explicit dimensions are supplied and fills each element with the empty value that matches the prototype's class (for example logical empty arrays, complex empties, or empty strings). When the prototype is a gpuArray, RunMat keeps the host-empty double fallback until GPU-resident cell storage ships.
• Size arguments must be finite, non-negative integers that are representable in double precision.

Does RunMat run cell on the GPU?

Cell arrays live on the CPU heap because each element can contain arbitrarily typed values that may require garbage-collected storage. When cell is invoked inside a fused GPU expression RunMat terminates the fusion group, gathers any GPU-resident size inputs, and creates the cell array on the host. 'Like' prototypes that refer to gpuArray inputs are honoured for shape inference. Each cell currently receives an empty double array even if the prototype itself is a gpuArray; this fallback will disappear once device-side cell residency lands. Once the acceleration layer adds true GPU cells, providers can register hooks without changing user code.

GPU memory and residency

No. Cell arrays are created on the host regardless of the residency of the size arguments. If the inputs happen to live on the GPU (for example, a gpuArray size vector), RunMat gathers those values before allocating the cell array. The result is always a host cell array; downstream code can still place tensors or gpuArrays inside individual cells as needed.

Examples

Creating a square cell array of empty placeholders

C = cell(3);
size(C)

Expected output:

ans =
     3     3

Creating a rectangular 2-by-4 cell array

C = cell(2, 4);
class(C)

Expected output:

ans =
    'cell'

Creating a 3-D cell array

C = cell(2, 3, 4);
size(C)

Expected output:

ans =
     2     3     4

Matching the size of an existing matrix

A = ones(5, 2);
C = cell(size(A));
size(C)

Expected output:

ans =
     5     2

Creating a cell array using a 'like' prototype

proto = logical.empty(0, 0);
C = cell(2, 3, 'like', proto);
cellfun(@isempty, C)

Expected output:

ans =
     1     1     1
     1     1     1

Using a column size vector

sz = [4; 1];
C = cell(sz);
size(C)

Expected output:

ans =
     4     1

Verifying that every cell starts with []

C = cell(2, 2);
isequal(C{1,1}, [], C{2,2}, [])

Expected output:

ans = logical
     1

Using cell with coding agents

Open a RunMat example with live inputs, then ask the agent to explain how cell changes the result.

FAQ

Related Cells functions

Open-source implementation

Unlike proprietary runtimes, every RunMat function is open-source. Read exactly how cell is executed, line by line, in Rust.

• View the source for cell in Rust on GitHub
• Learn how the RunMat runtime works
• Found a bug? Open an issue with a minimal reproduction.

About RunMat

RunMat is an open-source runtime that executes MATLAB-syntax code blazing on any GPU. It is licensed under the Apache 2.0 license.

• RunMat automatically optimizes your math for GPU execution on Apple, Nvidia, and AMD hardware. No code changes needed. Simulations that took hours now take minutes.
• Start running code in seconds. RunMat runs in the browser, on the desktop, or from the CLI. No license server, no IT ticket.

Getting started · Benchmarks · Pricing