hw_model_sweep.help

usage: hw_model_sweep.py [-h] -s {icache,dcache,bpred} -p PARAMS [--save_sim]
                         [--save_stats] [--load_stats] [--work_dir WORK_DIR]
                         [--track] [--max_workers MAX_WORKERS]
                         [--bp_top_acc_num BP_TOP_ACC_NUM]
                         [--bp_top_acc_thr BP_TOP_ACC_THR]
                         [--bp_top_mpki_num BP_TOP_MPKI_NUM]
                         [--bp_top_mpki_thr BP_TOP_MPKI_THR]
                         [--bp_run_best_for_all_workloads]
                         [--add_all_workloads] [--chart_title CHART_TITLE]
                         [--save_png] [--save_pdf] [--silent]
                         [--plot_skip_searched] [--plot_hr_thr PLOT_HR_THR]
                         [--plot_ct_thr PLOT_CT_THR] [--plot_no_ct_thr]
                         [--plot_name_filter PLOT_NAME_FILTER]
                         [--plot_acc_thr_min PLOT_ACC_THR_MIN]
                         [--plot_acc_thr_max PLOT_ACC_THR_MAX]
                         [--plot_mpki_thr_min PLOT_MPKI_THR_MIN]
                         [--plot_mpki_thr_max PLOT_MPKI_THR_MAX]
                         [--plot_per_workload]

Sweep through specified hardware models for a given workload

options:
  -h, --help            show this help message and exit
  -s {icache,dcache,bpred}, --sweep {icache,dcache,bpred}
                        Select the hardware model to sweep
  -p PARAMS, --params PARAMS
                        Path to the hardware model sweep params file
  --save_sim            Save simulation stdout in a log file
  --save_stats          Save combined simulation stats as json
  --load_stats          Load the previously saved stats from a json file
                        instead of running the sweep. Takes priority over
                        --save_stats
  --work_dir WORK_DIR   Path to run the sweep in and store results. Either
                        absolute, or relative to this script
  --track               Print the sweep progress
  --max_workers MAX_WORKERS
                        Maximum number of parallel workers to be dispatched,
                        if the sweep allows it
  --bp_top_acc_num BP_TOP_ACC_NUM
                        Number of top branch predictor configs to always
                        include based only on accuracy. Ignored for
                        --load_stats. Useful to ensure at least this many will
                        be included. Combined with other --bp_top_* results,
                        if used together
  --bp_top_acc_thr BP_TOP_ACC_THR
                        Accuracy threshold for the best branch predictor
                        configs to always include. Ignored for --load_stats.
                        Applies to average accuracy if run for multiple
                        workloads. Useful to ensure all good predictors are
                        included. Combined with other --bp_top_* results, if
                        used together
  --bp_top_mpki_num BP_TOP_MPKI_NUM
                        Number of top branch predictor configs to always
                        include based only on MPKI. Ignored for --load_stats.
                        Useful to ensure at least this many will be included.
                        Combined with other --bp_top_* results, if used
                        together
  --bp_top_mpki_thr BP_TOP_MPKI_THR
                        MPKI threshold for the best branch predictor configs
                        to always include. Ignored for --load_stats. Applies
                        to average MPKI if run for multiple workloads. Useful
                        to ensure all good predictors are included. Combined
                        with other --bp_top_* results, if used together
  --bp_run_best_for_all_workloads
                        Use 'best' predictors instead of 'binned' for the all
                        workloads run. In general, this means more predictors
                        to run for all workloads, and will bias towards better
                        performing (and usually larger) predictors. Useful for
                        very contrained runs, but will (significantly)
                        increase runtime for larger runs
  --add_all_workloads   Run with all workloads from the config (i.e. including
                        'skip_search: true' ones) after the main sweep
                        finished. Creates a single chart as average values
                        across all workloads. Used to validate HW model across
                        all benchmarks
  --chart_title CHART_TITLE
                        Adds this string to each chart's title
  --save_png            Save chart(s) as PNG
  --save_pdf            Save all charts as a single PDF
  --silent              Don't display chart(s) in pop-up window
  --plot_skip_searched  Don't plot searched workloads. Applicable only when
                        stats are loaded with --load_stats
  --plot_hr_thr PLOT_HR_THR
                        Set the lower limit for the plot y-axis for cache hit
                        rate
  --plot_ct_thr PLOT_CT_THR
                        Set the upper limit for the plot y-axis for cache
                        traffic
  --plot_no_ct_thr      Don't set the upper limit for the plot y-axis for
                        cache traffic
  --plot_name_filter PLOT_NAME_FILTER
                        Filter (regex) for this string only during plotting
  --plot_acc_thr_min PLOT_ACC_THR_MIN
                        Override the lower limit for the plot y-axis for
                        branch predictor accuracy
  --plot_acc_thr_max PLOT_ACC_THR_MAX
                        Override the upper limit for the plot y-axis for
                        branch predictor accuracy
  --plot_mpki_thr_min PLOT_MPKI_THR_MIN
                        Override the lower limit for the plot y-axis for
                        branch predictor MPKI
  --plot_mpki_thr_max PLOT_MPKI_THR_MAX
                        Override the upper limit for the plot y-axis for
                        branch predictor MPKI
  --plot_per_workload   Plot results from '--add_all_workloads' with one
                        workload per chart. Used to validate HW model across
                        all benchmarks. Requires running with '--
                        add_all_workloads'