RunMat
GitHub

write — Write numeric or text payloads to TCP client connections with MATLAB-compatible encoding behavior.

write(t, data, ...) sends binary or text payloads through a tcpclient connection while honoring MATLAB-compatible datatype conversion, byte-order handling, and timeout rules.

Syntax

count = write(client, data)
count = write(client, data, datatype)

Inputs

NameTypeRequiredDefaultDescription
clientAnyYestcpclient handle struct.
dataAnyYesPayload to send.
datatypeStringScalarNo"uint8"Data type label (for example "uint8", "double", "char", "string").

Returns

NameTypeDescription
countNumericScalarNumber of elements written to the socket.

Errors

IdentifierWhenMessage
RunMat:write:InvalidTcpClientClient handle is missing, malformed, invalid, or disconnected.write: invalid tcpclient handle
RunMat:write:InvalidInputArgument list shape is unsupported for write.write: invalid argument list
RunMat:write:InvalidDataPayload cannot be converted to the requested datatype.write: invalid data payload

How write works

  • write(t, data) converts data to unsigned 8-bit integers (the MATLAB default) and sends the bytes to the peer. The return value is the number of elements written when the caller requests an output argument.
  • write(t, data, datatype) encodes the payload using the supplied MATLAB datatype token. Supported values mirror MATLAB: "uint8" (default), "int8", "uint16", "int16", "uint32", "int32", "uint64", "int64", "single", "double", "char", and "string". Numeric conversions saturate to the destination range just like MATLAB cast operations. "char" treats values as single-byte character codes and "string" encodes UTF-8 text.
  • The client’s ByteOrder property controls how multi-byte numeric values are serialised. "little-endian" is the default, while "big-endian" matches the traditional network byte order.
  • When the socket cannot send the entire payload before the timeout expires, write raises RunMat:write:Timeout. If the peer closes the connection before or during the transfer the builtin raises RunMat:write:ConnectionClosed and marks the client as disconnected.
  • Inputs that originate on the GPU are gathered back to the host automatically before any bytes are written.

Does RunMat run write on the GPU?

Networking occurs on the host CPU. If data or the tcpclient struct resides on the GPU, RunMat gathers the values to host memory before converting them to bytes. Acceleration providers are not involved and the resulting payload remains on the CPU. Providers that support residency tracking automatically mark any gathered tensors as released.

Examples

Sending an array of bytes to an echo service

client = tcpclient("127.0.0.1", 50000);
count = write(client, uint8(1:4))

Expected output:

count =
     4

Writing doubles with explicit byte order

client = tcpclient("localhost", 50001, "ByteOrder", "big-endian");
values = [1.5 2.5 3.5];
write(client, values, "double")

Transmitting ASCII text

client = tcpclient("127.0.0.1", 50002);
write(client, "RunMat TCP", "char")

Sending UTF-8 encoded strings

client = tcpclient("127.0.0.1", 50003);
write(client, "Διακριτό", "string")

Handling connection closures

client = tcpclient("example.com", 12345, "Timeout", 0.25);
try
    write(client, uint8([1 2 3 4]));
catch err
    disp(err.identifier)
end

Expected output:

RunMat:write:ConnectionClosed

Using write with coding agents

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

Run a small write example, explain the result, then change one input and compare the output.

FAQ

How many output values does write return?

When the caller requests an output, the builtin returns the number of elements written (after datatype conversion). This mirrors the behaviour of MATLAB’s numeric I/O routines. If no output is requested, the value is discarded.

Does write support complex numbers?

No. The input must be real. Pass separate real and imaginary parts or convert to a byte representation manually.

How are values rounded when converting to integer datatypes?

Floating-point inputs are rounded to the nearest integer and then saturated to the target range, matching MATLAB casts (for example uint8(255.7) becomes 256 → 255, int8(-128.2) becomes -128).

What happens to GPU-resident tensors?

They are gathered automatically before the write. Networking is a CPU-only subsystem, so the resulting data is sent from host memory and any temporary handles are released after the transfer.

Can I stream large payloads?

Yes. write loops until the entire payload has been sent or an error occurs. Large payloads honour the client’s timeout and byte-order settings.

Open-source implementation

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

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

Download RunMat

Download RunMat for full performance, or use RunMat in your browser for zero setup.