AI Coding Agent for specification to hardware code.

The HDL Agent uses self-reflection, context, and grounding with EDA tools to generate hardware description language (HDL) code from specifications.

Installation instructions

Prerequisites

• Python 3.8 or higher
• Poetry for managing dependencies
• Yosys: Required for benchmarking and testing

Install with python poetry:

poetry install

If updating hdlagent, you may need to update poetry dependencies too:

poetry lock
poetry install

Set the required keys (depends on the model that you use)

export OPENAI_API_KEY=.....
export SAMBANOVA_API_KEY=....

Usage

The hdlagent script provides several commands to help you generate HDL code from specifications, run benchmarks, build code, and more.

General Help

To display general help with a list of all commands:

poetry run ../hdlagent/cli_agent.py --help

Available Commands

Here are the available commands:

• start - Start a new spec from a simple explanation
• bench - Run benchmarks using YAML specification files
• build - Build the code from spec files
• list-models - List available models
• log - Collect and display logs from previous runs

start Command

Description: Start a new structured spec file from a plain text problem description.

poetry run hdlagent/cli_agent.py start [options] <text_file1> [<text_file2> ...]

Example:

poetry run hdlagent/cli_agent.py start description.txt

Options:

• --llm LLM_MODEL: Specify the LLM model to use (default: gpt-4o)
• --lang {Verilog,Chisel,PyRTL,DSLX}: Language for code generation (default: Verilog)
• Additional options are available; run hdlagent start --help for more details.

bench Command

Description: Run benchmarks using YAML specification files.

Usage:

poetry run hdlagent/cli_agent.py bench [options] <spec_file1.yaml> [<spec_file2.yaml> ...]

Example:

poetry run hdlagent/cli_agent.py bench sample/RCA_spec.yaml

Options:

• --skip_completed: Skip already generated tests
• --llm LLM_MODEL: Specify the LLM model to use
• Additional options are available; run hdlagent bench --help for more details.

build Command

Description: Build the HDL code (and test benches) from spec files.

Usage:

poetry run hdlagent/cli_agent.py build [options] [spec_file1.yaml spec_file2.yaml ...]

If no spec files are specified, it will find all *spec.yaml files in the working directory.

Example:

poetry run hdlagent/cli_agent.py build

Options:

• --llm LLM_MODEL: Specify the LLM model to use
• Additional options are available; run hdlagent build --help for more details.

list-models Command

Description: List the existing models available for use.

Usage:

poetry run hdlagent/cli_agent.py list-models [options]

Example:

poetry run hdlagent/cli_agent.py list-models

log Command

Description: Collect and save RESULTS lines from previous runs, or display details of a specific run.

Usage:

poetry run hdlagent/cli_agent.py log [options] [benchmark_name]

Examples:

• Collect all RESULTS lines:

  poetry run hdlagent/cli_agent.py log
  

• List all available runs:

  poetry run hdlagent/cli_agent.py log --list-runs
  

• Display details of a specific run:

  poetry run hdlagent/cli_agent.py log RCA_spec.yaml
  

• Filter logs:

  poetry run hdlagent/cli_agent.py log --status=failed --date-from=2023-10-01 --date-to=2023-10-05
  

Options:

• --output-file OUTPUT_FILE: Specify the output file to save collected RESULTS lines.
• --list-runs: List all available runs and their timestamps.
• --status {all,success,failed}: Filter logs by run status (default: all).
• --date-from YYYY-MM-DD: Show logs from this date.
• --date-to YYYY-MM-DD: Show logs up to this date.
• --top_k TOP_K_VALUE: Filter logs where top_k is greater than or equal to the specified value.

Run a Simple test from json

cd sample
poetry run ../hdlagent/hdlagent.py --supp_context  --llm gpt-4-turbo-preview --bench ./sample-test.json

Contributing

Contributions are welcome! Please open an issue or submit a pull request on GitHub.