flow: Enable PARALLEL_EVAL setup repair in 38 designs by jhkim-pii · Pull Request #4760 · The-OpenROAD-Project/OpenROAD-flow-scripts

jhkim-pii · 2026-10-28T19:12:17Z

Summary

Add SETUP_REPAIR_PHASES (passed as repair_timing -phases) and enable the experimental PARALLEL_EVAL setup repair policy in the 38 designs (half of the 76 CI designs) with the largest finish timing gain.

Impact

Metric (38 enabled designs, vs LEGACY)	Median
Finish setup TNS reduction	+44.9%
Finish fmax	+1.02%
Finish std cell area	-0.11%
Finish power	-0.02%
Total flow runtime	-20.1%

Enabled designs

#	Design	Setup TNS reduction	Finish fmax	Area	Runtime
1	`sky130hs/jpeg`	+97.8%	+3.8%	+0.1%	+35.4%
2	`nangate45/bp_be_top`	+100.0%	+1.3%	-0.2%	+0.3%
3	`gf180/riscv32i`	+93.3%	+5.8%	+0.6%	-40.4%
4	`gf12/ibex`	+98.7%	+0.4%	+3.6%	-23.0%
5	`sky130hd/riscv32i`	+91.3%	+6.2%	-0.4%	-59.6%
6	`ihp-sg13g2/ibex`	+95.7%	-1.3%	+0.6%	+5.4%
7	`asap7/riscv32i`	+80.8%	+0.7%	+0.2%	-43.5%
8	`asap7/ibex`	+84.0%	-6.2%	+0.6%	-37.7%
9	`sky130hs/riscv32i`	+72.9%	+1.9%	+0.9%	-48.1%
10	`nangate45/bp_multi_top`	+72.5%	+0.2%	-0.3%	+4.4%
11	`sky130hd/chameleon`	+69.1%	+1.0%	-0.8%	-0.2%
12	`nangate45/ibex`	+74.1%	-5.0%	-0.2%	+6.9%
13	`asap7/riscv32i-mock-sram`	+54.7%	-0.5%	+1.8%	-24.1%
14	`sky130hd/ibex`	+45.1%	+7.7%	+2.7%	-31.2%
15	`asap7/aes_lvt`	+50.9%	+1.1%	+0.5%	+3.3%
16	`gf12/aes`	+49.5%	+1.3%	-0.7%	-41.9%
17	`asap7/chi_async_bridge`	+47.8%	+1.0%	-0.3%	-13.4%
18	`asap7/aes`	+48.0%	+0.4%	-0.2%	-0.1%
19	`asap7/uart`	+44.8%	+2.5%	-3.1%	+20.0%
20	`ihp-sg13g2/gcd`	+41.6%	+2.9%	-3.7%	+52.4%
21	`nangate45/ariane133`	+45.1%	-2.3%	+0.7%	-23.5%
22	`sky130hd/jpeg`	+41.1%	+1.3%	-2.1%	-28.4%
23	`gf12/tinyRocket`	+36.8%	+3.9%	-0.0%	-24.9%
24	`sky130hs/aes`	+35.0%	+1.7%	-0.3%	-11.5%
25	`asap7/gcd-ccs`	+31.1%	+1.4%	-2.8%	+0.7%
26	`sky130hs/ibex`	+20.0%	+3.3%	-0.4%	-42.5%
27	`nangate45/aes`	+20.3%	+0.0%	-1.0%	-5.5%
28	`nangate45/tinyRocket`	+17.9%	+0.8%	-0.1%	-41.9%
29	`gf180/ibex`	+17.1%	-0.4%	+2.5%	-35.5%
30	`ihp-sg13g2/riscv32i`	+9.4%	+6.6%	-1.1%	-50.3%
31	`sky130hd/aes`	+14.3%	+0.7%	+1.0%	-5.6%
32	`asap7/ethmac_lvt`	+14.5%	+0.1%	-0.1%	+0.4%
33	`nangate45/dynamic_node`	+6.6%	+7.4%	+2.3%	-4.3%
34	`sky130hd/gcd`	+8.5%	+3.0%	-5.7%	-69.4%
35	`gf180/aes-hybrid`	+7.7%	+1.0%	+8.2%	-23.5%
36	`asap7/jpeg`	+7.0%	+0.4%	-0.5%	-17.3%
37	`nangate45/cva6`	+10.6%	-3.2%	+5.7%	-29.1%
38	`asap7/gcd`	+4.0%	+0.9%	+1.0%	+1.3%

Code Review

This pull request introduces the SETUP_REPAIR_PHASES configuration variable, which allows specifying setup repair phase names (such as the experimental PARALLEL_EVAL policy) for repair_timing. The variable has been documented, integrated into the flow scripts, and enabled across various design configurations. The review feedback identifies redundant definitions of SETUP_REPAIR_PHASES in gcd-ccs/config.mk and riscv32i-mock-sram/config.mk, as these files already inherit the setting from their respective base configurations.

repair_timing selects its setup repair pipeline with -phases, but ORFS had no way to pass it, so a design could not opt into the experimental PARALLEL_EVAL policy. Add SETUP_REPAIR_PHASES, passed as -phases by repair_timing_helper in the same way SETUP_MOVE_SEQUENCE is passed as -sequence. It applies to every repair_timing_helper call (floorplan, place, cts and grt). When unset, repair_timing keeps its default phases, so no design changes. Signed-off-by: Jaehyun Kim <jhkim@precisioninno.com>

Set SETUP_REPAIR_PHASES = PARALLEL_EVAL in the 38 designs with the largest final timing gain when every CI design ran the policy, half of the 76 designs in the public and private CI. A CI run with PARALLEL_EVAL enabled for every design was compared against the same ORFS and OpenROAD commits without it. The gain score is the mean of the finish fmax gain and the finish setup TNS reduction (clipped to +-100%). A design was eligible when its TEST flow completed, passed the dashboard QoR check, gained timing, and added no DRC, antenna or hold TNS degradation. rapidus2hp designs are not in this repository and were left out. asap7/gcd-ccs and asap7/riscv32i-mock-sram inherit the setting from the included asap7/gcd and asap7/riscv32i configs. | Rank | Design | Score | fmax | Setup TNS | | ---: | --- | ---: | ---: | ---: | | 1 | sky130hs/jpeg | +50.8 | +3.8% | +97.8% | | 2 | nangate45/bp_be_top | +50.6 | +1.3% | +100.0% | | 3 | gf180/riscv32i | +49.5 | +5.8% | +93.3% | | 4 | gf12/ibex | +49.5 | +0.4% | +98.7% | | 5 | sky130hd/riscv32i | +48.7 | +6.2% | +91.3% | | 6 | ihp-sg13g2/ibex | +47.2 | -1.3% | +95.7% | | 7 | asap7/riscv32i | +40.7 | +0.7% | +80.8% | | 8 | asap7/ibex | +38.9 | -6.2% | +84.0% | | 9 | sky130hs/riscv32i | +37.4 | +1.9% | +72.9% | | 10 | nangate45/bp_multi_top | +36.4 | +0.2% | +72.5% | | 11 | sky130hd/chameleon | +35.1 | +1.0% | +69.1% | | 12 | nangate45/ibex | +34.6 | -5.0% | +74.1% | | 13 | asap7/riscv32i-mock-sram | +27.1 | -0.5% | +54.7% | | 14 | sky130hd/ibex | +26.4 | +7.7% | +45.1% | | 15 | asap7/aes_lvt | +26.0 | +1.1% | +50.9% | | 16 | gf12/aes | +25.4 | +1.3% | +49.5% | | 17 | asap7/chi_async_bridge | +24.4 | +1.0% | +47.8% | | 18 | asap7/aes | +24.2 | +0.4% | +48.0% | | 19 | asap7/uart | +23.7 | +2.5% | +44.8% | | 20 | ihp-sg13g2/gcd | +22.2 | +2.9% | +41.6% | | 21 | nangate45/ariane133 | +21.4 | -2.3% | +45.1% | | 22 | sky130hd/jpeg | +21.2 | +1.3% | +41.1% | | 23 | gf12/tinyRocket | +20.4 | +3.9% | +36.8% | | 24 | sky130hs/aes | +18.3 | +1.7% | +35.0% | | 25 | asap7/gcd-ccs | +16.3 | +1.4% | +31.1% | | 26 | sky130hs/ibex | +11.7 | +3.3% | +20.0% | | 27 | nangate45/aes | +10.2 | +0.0% | +20.3% | | 28 | nangate45/tinyRocket | +9.4 | +0.8% | +17.9% | | 29 | gf180/ibex | +8.4 | -0.4% | +17.1% | | 30 | ihp-sg13g2/riscv32i | +8.0 | +6.6% | +9.4% | | 31 | sky130hd/aes | +7.5 | +0.7% | +14.3% | | 32 | asap7/ethmac_lvt | +7.3 | +0.1% | +14.5% | | 33 | nangate45/dynamic_node | +7.0 | +7.4% | +6.6% | | 34 | sky130hd/gcd | +5.7 | +3.0% | +8.5% | | 35 | gf180/aes-hybrid | +4.4 | +1.0% | +7.7% | | 36 | asap7/jpeg | +3.7 | +0.4% | +7.0% | | 37 | nangate45/cva6 | +3.7 | -3.2% | +10.6% | | 38 | asap7/gcd | +2.4 | +0.9% | +4.0% | Signed-off-by: Jaehyun Kim <jhkim@precisioninno.com>

openroad-ci · 2026-10-28T21:07:56Z

🔍 QoR check

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

Commit 2b5b8d9 · Jenkins build #2 · Baseline: build · View build on dashboard

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

❌ nangate45/cva6 base — 2 failing metric(s)

Metric	target	base	delta	limit	band
`finish__timing__setup__ws`	-1.438	-1.359	-5.813%	-1.427	PeriodPadding 5.000%
`globalroute__timing__setup__ws`	-1.426	-1.347	-5.802%	-1.415	PeriodPadding 5.000%

jhkim-pii · 2026-10-28T21:43:03Z

@maliberty This PR enables PARALLEL_EVAL in 50% of ORFS designs.

Failure

asap7/swerv_wrapper: [ERROR RSZ-1008] FixedDelay conversion out of range: -inf

Nightly CI shows the same fail. Different issue.

oharboe · 2026-10-29T07:59:50Z

🥳

jhkim-pii self-assigned this Oct 28, 2026

gemini-code-assist Bot reviewed Oct 28, 2026

View reviewed changes

Comment thread flow/designs/asap7/gcd-ccs/config.mk Outdated

Comment thread flow/designs/asap7/riscv32i-mock-sram/config.mk Outdated

jhkim-pii added 2 commits October 28, 2026 12:18

openroad-ci force-pushed the secure-parallel-eval-rollout branch from bd6d30d to 69cc62a Compare October 28, 2026 19:23

jhkim-pii requested a review from maliberty October 28, 2026 20:35

jhkim-pii requested a review from precisionmoon October 28, 2026 21:43

maliberty approved these changes Oct 28, 2026

View reviewed changes

maliberty merged commit 7bf5a81 into The-OpenROAD-Project:master Oct 28, 2026
10 of 12 checks passed

maliberty deleted the secure-parallel-eval-rollout branch October 28, 2026 21:55

Navigation Menu

Repository navigation

flow: Enable PARALLEL_EVAL setup repair in 38 designs - #4760

flow: Enable PARALLEL_EVAL setup repair in 38 designs#4760
maliberty merged 2 commits into
The-OpenROAD-Project:masterfrom
The-OpenROAD-Project-staging:secure-parallel-eval-rollout

jhkim-pii commented Oct 28, 2026 •

edited

Loading

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

Uh oh!

Uh oh!

openroad-ci commented Oct 28, 2026 •

edited

Loading

Uh oh!

jhkim-pii commented Oct 28, 2026

Uh oh!

Uh oh!

oharboe commented Oct 29, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

jhkim-pii commented Oct 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Impact

Related

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

openroad-ci commented Oct 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔍 QoR check

Uh oh!

jhkim-pii commented Oct 28, 2026

Failure

Uh oh!

Uh oh!

oharboe commented Oct 29, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

jhkim-pii commented Oct 28, 2026 •

edited

Loading

openroad-ci commented Oct 28, 2026 •

edited

Loading