poka.sh

 Docs

MLIR

Generated by scripts/gen-format-pages.py from docs/data/format-metadata/, FILE_FORMATS.csv and docs/formats-master.csv. Do not edit by hand: change the source, then run the script.

MLIR Preview.

At a glance
Purpose MLIR (Multi-Level Intermediate Representation) in text form: operations from compiler dialects such as func, arith, linalg and tensor.
Extensions .mlir
Poka parser poka-mlir
Viewer poka-metadata-card-viewer
Depth tier Metadata card (see Depth tiers)
Breadth tier B4 Summary (see Breadth tiers)

How much Poka reads

B4 Summary. The container and metadata are read and summarised (counts, properties, structure), but the main content is not rendered.

Confidence: medium (the reviewer read the code). Basis: Reviewed 2026-10-26: read parsers/poka-mlir/src/main.rs (217 lines) and plugin.json in full. The crate has no unit tests and no tests/samples.rs, and the corpus has no .mlir sample, so the tier rests on reading the code. Covers .mlir. The output is a metadata card of counts found by matching text line by line: module blocks, func.func declarations, block arguments, regions, return statements, SSA value tokens, attribute blocks, private visibility, operations of well-known dialects and the most used dialects. Judgement: B4, because the structure is summarised but the source itself is never shown.

What Poka does not read

  • The source text is not shown: the card carries only counts and a few sample names, with no code view or highlighting
  • Constructs are found by matching text line by line, not by parsing, so multi-line declarations, nested or indented definitions and keywords inside strings or comments can be miscounted
  • Files over 256 MiB are cut to the first 256 MiB (the card says so)
  • Text that is not UTF-8 is decoded with replacement characters; no other encoding is tried
  • Operations are counted only for a list of well-known dialects; others are not identified
  • Function names, signatures and bodies are not shown

Libraries

Libraries Poka uses or has assessed for this format, with the status, licence and language the registry records.

  • melior (candidate; Apache-2.0; Rust; https://github.com/mlir-rs/melior) — Rust wrapper over the MLIR C API; source has Module::parse(context, &str). Needs an installed LLVM/MLIR, alpha API.
  • MLIR (LLVM project) parser and Python bindings (candidate; Apache-2.0 WITH LLVM-exception; C++, Python; https://mlir.llvm.org/docs/Bindings/Python/) — Reference parser; Python bindings show Module.parse on MLIR assembly. Licence from llvm-project LICENSE.TXT. Not portable to wasm easily.

Sources

  • parsers/poka-mlir/src/main.rs
  • parsers/poka-mlir/plugin.json