How divanshu-walia/c_rtl_coverage works
This is an HLS testbench-generation and coverage-evaluation project. Users select benchmarks and strategies; the pipeline generates or reuses C testbenches, synthesizes the benchmark with Bambu, measures C coverage, replays captured calls in Verilator, and records coverage results for plotting. Separate workflows refine LLM testbenches iteratively and run/analyze independent trials, including reachability-ceiling and remeasurement utilities. The graph reflects documented workflows and visible module references; several implementation details and benchmark sources are not present in the supplied tree or excerpts.
Main entry points: run_pipeline.py Coverage pipeline · agentic_loop.py Agentic refinement · generate_tb.py Generation CLI · run_replicates.py Repeated trials
Components 22
Testbench generation
- Generation CLI
pipeline/generate_tb.py - Manual and random
pipeline/generate_tb.py - LLM generation
pipeline/generate_tb.py - Claude client
pipeline/llm_client.py - Generated testbenches
testbenches
Coverage measurement
- Coverage pipeline
pipeline/run_pipeline.py - C coverage run
pipeline/run_pipeline.py - Call capture
pipeline/call_capture.py - RTL replay
pipeline/run_pipeline.py - Coverage parser
pipeline/extract_coverage.py
Experiment workflows
- Agentic refinement
pipeline/agentic_loop.py - Repeated trials
pipeline/run_replicates.py - Reachability ceiling
pipeline/run_ceiling.py - Trial remeasurement
pipeline/remeasure_experiment.py
Results and analysis
- Coverage results
pipeline/run_pipeline.py - Coverage plots
pipeline/plot_results.py - Replicate analysis
pipeline/analyze_replicates.py - Model comparison
model_comparison/compare_models.py
Other components
- Engineer
- Anthropic API
- Bambu HLS
- Verilator
Connections 30
- Engineer → Generation CLI: selects strategy
README.md - Generation CLI → Manual and random: builds manual/random
pipeline/generate_tb.py - Generation CLI → LLM generation: requests LLM
pipeline/generate_tb.py - LLM generation → Claude client: calls
pipeline/generate_tb.py - Claude client → Anthropic API: requests completion
pipeline/llm_client.py - Manual and random → Generated testbenches: writes
README.md - LLM generation → Generated testbenches: writes
README.md - Engineer → Coverage pipeline: runs benchmarks
README.md - Coverage pipeline → Generated testbenches: uses
pipeline/run_pipeline.py - Coverage pipeline → Bambu HLS: synthesizes with
pipeline/run_pipeline.py - Coverage pipeline → C coverage run: runs C coverage
README.md - Coverage pipeline → Call capture: captures calls
README.md - Coverage pipeline → RTL replay: replays stimulus
README.md - RTL replay → Verilator: simulates RTL
README.md - Coverage pipeline → Coverage parser: parses reports
pipeline/run_pipeline.py - Coverage parser → Coverage results: provides metrics
README.md - Coverage pipeline → Coverage results: records results
README.md - Engineer → Coverage plots: requests plots
README.md - Coverage plots → Coverage results: reads
README.md - Engineer → Agentic refinement: starts refinement
README.md - Agentic refinement → Claude client: requests refinements
pipeline/agentic_loop.py - Agentic refinement → Coverage pipeline: measures iterations
pipeline/agentic_loop.py - Engineer → Repeated trials: starts trials
README.md - Repeated trials → Coverage pipeline: runs measurements
pipeline/run_replicates.py - Repeated trials → Agentic refinement: can seed refinement
README.md - Engineer → Reachability ceiling: requests stress test
README.md - Engineer → Trial remeasurement: requests remeasurement
README.md - Engineer → Replicate analysis: analyzes trials
README.md - Replicate analysis → Coverage parser: uses coverage helpers
pipeline/analyze_replicates.py - Engineer → Model comparison: compares models