dse-loop

Automate parameter tuning by running a target program, collecting results, and iteratively adjusting configurations until a defined objective is achieved within a specified timeout. The autonomous run-analyze-tune loop handles iteration caps and logs results for architecture and EDA optimization.

Autonomous design space exploration reduces manual trial-and-error by executing a program, parsing outcomes into a numeric objective, and iteratively tuning parameters. It handles timeouts, patience-based stopping, and logs progress through DSE_STATE.json and dse_log.csv for recovery.

Explore compiler flags and hardware knobs by invoking the exploration loop with your parameterized program. Specify a numeric objective, such as maximizing IPC or minimizing area and latency, and let the autonomous process sweep configurations until the timeout or iteration limit is reached.

You can recover interrupted design space exploration using the DSE_STATE.json file. The loop supports state recovery alongside logging outcomes to dse_log.csv, allowing the autonomous tuning process to resume from its last saved checkpoint.

You need a runtime environment to execute programs, a mechanism to parse program outcomes into a numeric objective, and support for timeouts and iteration caps. These prerequisites enable the autonomous loop to sweep configurations for hardware-software co-design experiments.

Limitations include dependency on a defined numeric objective to evaluate outcomes and strict timeout or iteration caps that halt the process. The autonomous loop requires a runtime to execute programs and cannot tune parameters without a parseable success metric.