dse-loop

Automate parameter sweeping by running an iterative design space exploration loop that executes candidate configurations, parses objective metrics from outputs, and selects improved parameters until objectives are met or timeouts occur.

Tune RTL compile flags and memory subsystem parameters by defining YAML task instructions to run candidate parameter sets, extract numeric objectives programmatically, and let the adaptive search strategy iteratively select optimal configurations.

Design space exploration applies adaptive search behaviors that automatically adjust strategy based on parameter cardinality and type, selecting different optimization approaches for categorical, continuous, or small-cardinality variables.

Resume architecture tuning experiments from saved state because the loop records every run, including failures, into structured logs and supports recovery to continue iterative optimization without losing previous progress.

Iterative synthesis tuning and simulator sweeps enforce conservative safety rules that prevent destructive actions, ensuring experiments run non-destructively while logging all configurations and metrics for analysis.

Define parameter ranges in YAML task instructions if known, specifying which parameters to tune, the objective to optimize, and total timeout, allowing the loop to autonomously explore within those boundaries.