Automates iterative design space exploration for computer architecture and EDA parameter tuning.
npx skills add https://github.com/OpenLAIR/dr-claw --skill aris-dse-loop
Please help me install this Agent Skill. Skill: aris-dse-loop Source: https://github.com/OpenLAIR/dr-claw/tree/main/skills/aris-dse-loop Command: npx skills add https://github.com/OpenLAIR/dr-claw --skill aris-dse-loop
Manually sweeping hardware or toolchain parameters to find an optimal configuration is slow, error-prone, and hard to track. This Skill runs an autonomous run-analyze-tune loop that explores a design space until an objective is met, a timeout is reached, or improvements stall.
Ask the agent to run a design space exploration on your simulation or synthesis command, naming the parameters to tune, the objective metric to optimize, and a wall-clock timeout.
High-intent search queries and answers about installing and using this skill.
Invoke the skill with the simulation command, the parameters to tune such as L1D_SIZE, L2_SIZE, and ROB_ENTRIES, the objective like maximizing IPC, and a timeout. The loop runs baseline and exploration iterations, logging every result to dse_log.csv.
Provide the make or script target, parameters such as clock period and optimization passes, and an objective like minimizing area at timing closure. The skill sweeps configurations, checks constraints each run, and reports the best feasible design point.
The skill reads the source code for argparse definitions, config files, and hardware parameters, then infers ranges using domain heuristics such as powers of two for cache sizes. Inferred ranges are written to dse_results/inferred_params.md for review.
Yes. State is persisted in DSE_STATE.json and dse_log.csv after every iteration, so the loop can reload the current iteration, best parameters, and patience counter and resume from where it stopped.
The loop stops when the timeout is reached, the iteration cap is hit, no improvement occurs for the configured patience count, or the success criteria are met. Constraint-violating points never count as improvements.
londeyEnforce consistent Verilog and SystemVerilog coding style for Verilator linting and Yosys synthesis.
tnn1t1sFormat benchmark results into tables, charts, and Markdown summaries.
Kingdaddy007Diagnose performance bottlenecks in backend services using profiling and baseline metrics.
AbhinavbwjEvaluates public space designs using Jan Gehl's 12 quality criteria and PPS placemaking methodologies.
