dse-loop

Automate parameter tuning for computer architecture simulations by running a design-space exploration loop that executes the target program, analyzes outputs, and iteratively adjusts parameters to optimize your objective within a specified timeout.

Autonomous design-space exploration is an iterative workflow that runs experiments, evaluates results, selects the next parameter sets, and repeats until an objective is met or a time budget expires. It automates the run, analyze, and adjust cycle for architecture and EDA problems.

You do not need to manually define all parameter ranges to start an EDA optimization workflow. The exploration loop can infer missing parameter ranges directly from the source code when they are not explicitly provided in the task description.

Recover and continue an interrupted architecture simulation optimization using the dedicated state recovery mechanism. The loop generates a DSE_STATE.json file and a dse_results directory, allowing the autonomous exploration to resume from its last saved state.

You can set a timeout for automated parameter tuning to limit execution time. The exploration loop accepts a time budget argument and will safely terminate the iterative parameter adjustment process once that limit is reached.

The best way to log results from an autonomous design-space exploration loop is to use the built-in logging mechanism. It automatically creates a dse_results directory and uses a dedicated parse_result script to record and evaluate experiment outputs.