HLSABC

Overview

Estimating high-level metrics for high-level synthesis (HLS) with large language models (LLMs).

Features

• Benchmarks: Polybench and ABC benchmarks for HLS evaluation.
• Utilities: Tools for logging, performance feedback, and estimations.
• Tests: Comprehensive test suite for validation.

Repository Structure

• benchmarks/: Contains benchmark programs for HLS evaluation.
• input/: Input files for HLS designs.
• logs/: Log files generated during execution.
• scratch/: Temporary files and scripts.
• scripts/: Helper scripts for testing and execution.
• tests/: Test cases for utilities and benchmarks.
• utils/: Utility modules for logging, benchmarking, and estimations.
• vivado/: Vivado project files and reports.

Getting Started

• Clone the repository:

git clone https://github.com/xli562/hlsllm.git

• Install Allo

• Install dependencies:

conda env create -n hlsabc -f environment.yml
pip install -r requirements.txt

• Run tests:

./scripts/pytest --parallel

Testing

Some tests require certain files in tests/resource.

Running tests

./scripts/pytest --parallel

If --parallel is specified, all tests not marked by @pytest.mark.serial will run in parallel. Then, all tests marked as serial will run in serial.

If --parallel is not specified, all tests will run in serial.

Configuring tests

In pytest.ini,

• testpaths specifies the directory / directories to collect tests from.
• markers defines pytest markers such as serial or slow.
• addopts can specify which specific files or functions should be run.

For example,

testpaths = 
    tests

markers =
    serial: tests that need to be run serially
    slow: slow tests (runs csynth or LLM)

addopts = 
    --color=yes
    tests/test_llm.py::test_simple

runs test_simple in tests/test_llm.py.