[oharboe]

Report what is free unconditionally. With SKIP_REPORT_METRICS=1, report_metrics wrote no metrics at all, so a flow skipping it for speed also lost design area, instance counts and utilization. Those are a walk over the instances and need no timer. They are now written before the skip; SKIP_REPORT_METRICS skips only what costs time.

What SKIP_REPORT_METRICS=1 skips, measured on asap7/gcd (ms, summed over the stage's report_metrics calls, DETAILED_METRICS=1):

call	floorplan	detailed place	cts	grt	final	under log_cmd
report_power	49	48	65	203	60	yes
first report_tns (the timing update)	0.1	23	49	95	8	yes
input→register find_timing_paths ×2	–	17	42	143	94	yes (eval)
register→register report_checks ×2	–	14	33	115	67	yes (eval)
report_erc_metrics	–	17	27	62	20	yes
report_check_types	–	5	7	20	6	yes
report_clock_min_period	1.6	1.6	6	18	–	yes
report_clock_skew, report_wns, report_worst_slack, unfiltered report_checks, other *_metric	≤ 4 each					no

report_design_area_metrics, now unconditional: 0.07 ms. The tools' own counters (gpl, dpl, grt, drt, repair buffer counts, warnings) never went through report_metrics and are written either way.

The calls that cost time even on gcd run under log_cmd, so on a large design the log names any that takes five seconds or more. The register path searches go through log_cmd eval { ... } so the log prints the command, not the expanded register list.

Tests, asap7/gcd:

• metrics on vs SKIP_REPORT_METRICS=1: byte-identical .odb at every stage and identical 6_final.v;
• with SKIP_REPORT_METRICS=1, every stage's metrics .json has design__instance__area and no timing__setup__ws;
• with metrics on, the metric keys are the same as before the change.

tclfmt/tclint 0.7.0, generate-variables-docs.py, yaml_to_json.py and yamlfix leave the tree unchanged.

Integration test: The-OpenROAD-Project/bazel-orfs#1433.

[gemini-code-assist]

Code Review

This pull request updates the behavior of SKIP_REPORT_METRICS so that it only skips metrics requiring heavy analysis (such as timing, power, and clock skew) while still reporting basic metrics like design area, instance counts, and utilization. To support this, report_design_area_metrics is now called before checking the SKIP_REPORT_METRICS flag, and various reporting commands in report_metrics.tcl have been wrapped with log_cmd. Documentation and variable definitions have been updated accordingly. The reviewer pointed out that the path reports/<stage>_*.rpt in the description contains <stage>, which resembles an HTML tag and could cause rendering issues in Markdown; they suggested wrapping it in backticks.

[openroad-ci]

🔍 QoR check

Metrics reflect the PR merge build — i.e. what will land on the target branch.

Commit 9a91cc9 · Jenkins build #3 · Baseline: build · View build on dashboard

62 design(s) checked — 1 with regression(s), 1 without a comparable baseline.

❌ asap7/aes-block base — 2 failing metric(s)

Metric	target	base	delta	limit	band
cts__timing__setup__tns	-2405.450	-409.639	-487.212%	-499.639	PeriodPadding 20.000%
cts__timing__setup__ws	-57.057	-27.467	-107.733%	-49.967	PeriodPadding 5.000%

[oharboe]

@maliberty I want as much free(fast) information as possible with the fewest ORFS variables

Hence I changed SKIP_REPORT_METRICS to report what it can for free and only disable reporting that takes time.