dlmwrite — Write numeric matrices to delimiter-separated text files with MATLAB-compatible precision and offset controls.
dlmwrite(filename, M) writes the numeric contents of M to a plain text file using a configurable delimiter. It is MATLAB's legacy companion to csvwrite, adding support for custom delimiters, per-call precision, line endings, and zero-based row/column offsets. RunMat mirrors these semantics so that existing MATLAB scripts continue to behave identically.
How dlmwrite works in RunMat
- Accepts real numeric or logical arrays up to two dimensions (trailing singleton dimensions are ignored). Complex inputs raise an error. Logical values are converted to
0/1. - The default delimiter is a comma. Supply a character vector, string scalar, or control sequence (
"\t","\n","\r") to change it. Delimiters must not be empty. - Positional syntax matches MATLAB:
dlmwrite(filename, M, delimiter)anddlmwrite(filename, M, delimiter, row, col)provide quick overrides. Row/column offsets are zero-based and insert blank rows/columns filled with the delimiter before the numeric data. - Name-value pairs extend any syntax:
'-append'appends to an existing file,'delimiter'overrides the separator,'precision'accepts either a scalar (significant digits) or a C-style printf format such as'%.6f','newline'chooses between'pc'(CRLF),'unix'(LF), or'mac'(CR), and'roffset'/'coffset'mirror the positional offsets. - Empty fields created by offsets are emitted as empty tokens separated by the delimiter, matching MATLAB's behaviour.
- Files are opened in text mode. When appending, RunMat preserves the existing newline convention and honours offsets relative to the end of the file.
- Paths beginning with
~expand to the user's home directory before writing, matching other RunMat I/O builtins.
How dlmwrite runs on the GPU
dlmwrite performs all formatting on the CPU. When the input array lives on a GPU, RunMat gathers it through the active acceleration provider before serialisation. No provider hooks are required, and the builtin returns once the host-side write completes. If no provider is registered, the same gather error surface as other residency sinks appears.
GPU memory and residency
No. dlmwrite gathers GPU-resident tensors automatically before writing. This mirrors MATLAB's behaviour, where dlmwrite always operates on host-resident values.
Examples
Writing a numeric matrix with the default comma delimiter
A = [1 2 3; 4 5 6];
dlmwrite("scores.csv", A)Using a semicolon delimiter and positional offsets
B = magic(3);
dlmwrite("offset.txt", B, ";", 1, 2)Appending data with explicit row and column offsets
dlmwrite("log.txt", 1:3);
dlmwrite("log.txt", 10:12, "-append", "roffset", 1, "coffset", 1)Controlling precision with significant digits
vals = [12.34567, pi];
dlmwrite("precision3.csv", vals, "precision", 3)Emitting fixed-point values with a printf-style format string
dlmwrite("fixed.txt", rand(2, 2), "precision", "%.6f", "delimiter", "\t")Selecting the Windows line ending while writing GPU data
G = gpuArray(single([1.5; 2.5; 3.5]));
dlmwrite("gpu_pc.txt", G, "newline", "pc")Expected output:
% Data is gathered from the GPU and written using CR/LF line endings.FAQ
Which data types does dlmwrite support?
Real numeric and logical arrays up to two dimensions. Use writematrix/writecell for text or heterogeneous data.
How does the '-append' flag interact with offsets?
Offsets are applied relative to the end of the existing file. '-append','roffset',2 inserts two blank delimiter-filled rows before writing the new data.
What happens if I request both positional and name-value offsets?
RunMat follows MATLAB by allowing both; the last specification wins. For clarity, prefer one style per call.
Can I combine '-append' with positional delimiter arguments?
Yes. For example dlmwrite(fname, M, ';', '-append') uses the semicolon delimiter and appends to the file.
Does 'precision', N control decimal places or significant digits?
It controls significant digits (like MATLAB's %.*g). For fixed decimal places, pass a printf-style format such as '%.6f'.
Which line endings are supported?
'newline','pc' emits \r\n (Windows). 'newline','unix' emits \n, and 'newline','mac' emits \r. The default matches the current platform.
How are NaN and Inf values written?
They are emitted verbatim as NaN, Inf, or -Inf, independent of the precision setting, matching MATLAB.
How are offsets represented in the file?
roffset produces blank delimiter-separated rows, while coffset inserts empty fields before each data row. Downstream tools will see empty strings for the offset cells.
Does dlmwrite normalise path separators?
No. Paths are passed directly to the OS after optional ~ expansion, exactly like MATLAB.
Should new code use dlmwrite?
Prefer writematrix for new code. Use dlmwrite only when maintaining legacy scripts that rely on its exact output.
Related functions to explore
These functions work well alongside dlmwrite. Each page has runnable examples you can try in the browser.
dlmread, csvwrite, writematrix, fprintf, gpuArray, gather, csvread, readmatrix
Open-source implementation
Unlike proprietary runtimes, every RunMat function is open-source. Read exactly how dlmwrite works, line by line, in Rust.
- View dlmwrite.rs on GitHub
- Learn how the 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 — faster, on any GPU, with no license required.
- Simulations that took hours now take minutes. RunMat automatically optimizes your math for GPU execution on Apple, Nvidia, and AMD hardware. No code changes needed.
- Start running code in seconds. Open the browser sandbox or download a single binary. No license server, no IT ticket, no setup.
- A full development environment. GPU-accelerated 2D and 3D plotting, automatic versioning on every save, and a browser IDE you can share with a link.