This repository contains the implementation of FSST-LIKE-Matching, a high-performance string matching engine published at DaMoN 2026. This work is built upon the FSST compression algorithm; the link to the official repository is https://github.com/cwida/fsst.

The algorithms in the repository provide efficient pattern matching (specifically LIKE patterns) over FSST-compressed data. By avoiding decompression, we achieve high throughput for analytical workloads.

The project is organized into several key directories:

• include/: Header files defining the core automata, codegen interfaces, and string search utilities.
• src/: Implementation of the pattern matching engine, including LLVM and C++ code generation.
• benchmark/: Performance measurement suite for datasets including IMDB, StackOverflow, and TPC-H.
• test/: Unit and integration tests for various matching scenarios (start, middle, end, and full pattern matching).
• fa-drawing/: A visualization tool (web-based) to draw and inspect the finite automata used in the matching process.

• Automata-based Matching: Uses specialized finite automata for LIKE pattern evaluation.
• Codegen Backend: Supports both LLVM-based JIT compilation and C++ source generation for matching kernels.
• FSST Integration: Direct matching on compressed data without full decompression.
• Multithreaded Benchmarking: Tools to measure throughput across multiple CPU cores.

• CMake (3.5+)
• LLVM16 (for LLVM codegen backend)
• Vectorscan / Hyperscan (for hybrid search support)
• C++20 compatible compiler

mkdir build && cd build
cmake ..
make -j$(nproc)

To use the browser client, run the following command:

cd fa-drawing
./run_server.sh

After starting the server, open up index.html in a browser. You must provide three values in order to generate the automaton correctly:

• Pattern: The LIKE pattern to generate the automaton for.
• Symbol Table Path: The path to the binary of the symbol tables relative to the main folder of the repository; example symbol tables are provided in data/.
• Automaton Type: One of four possible values:
  • start: For prefixes (the % wildcard is implicit at the end).
  • middle: For substrings (the % wildcard is implicit at both ends).
  • end: For suffixes (the % wildcard is implicit at the start).
  • full: May contain multiple LIKE subpatterns and multiple % wildcards.
• Compress File Path: If you want to generate a symbol table for a custom dataset, enter the path relative to the main folder of the repository. The format of the file is: one string entry per line.

Once the datasets have been installed, you can run the benchmark scripts by running the following commands:

cd benchmark
./run_benchmarks.sh