Preserve pedometer layer transit and via continuity - #2548
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Benchmark This PRRun benchmarks by commenting on this PR: Comment Everything after Use Any PR whose title contains |
|
/benchmark --same-machine --dataset 18 --pipeline 9 |
Same Machine Benchmark ResultsBoth revisions ran sequentially in one Blacksmith job on Dataset:
Outcome changes: 0 improved, 1 regressed. DRC issues are totaled across solved samples. Timing percentiles include solved and timed-out samples; negative timing deltas are faster. Main pipeline stage timingsAutoroutingPipelineSolver9_PreloadedTraceGraph Recorded timings: 16/16 samples (3 partial, including failed or timed-out samples).
PR pipeline stage timingsAutoroutingPipelineSolver9_PreloadedTraceGraph Recorded timings: 16/16 samples (4 partial, including failed or timed-out samples).
Changed outcomes (1)
Workflow: View run |
|
/benchmark --same-machine --dataset 18 --pipeline |
Same Machine Benchmark ResultsThe paired benchmark ended with failure before both reports were produced. Workflow: View run |
|
/benchmark --same-machine --dataset 18 --pipeline 9 |
Same Machine Benchmark ResultsBoth revisions ran sequentially in one Blacksmith job on Dataset:
Outcome changes: 0 improved, 1 regressed. DRC issues are totaled across solved samples. Timing percentiles include solved and timed-out samples; negative timing deltas are faster. Main pipeline stage timingsAutoroutingPipelineSolver9_PreloadedTraceGraph Recorded timings: 16/16 samples (3 partial, including failed or timed-out samples).
PR pipeline stage timingsAutoroutingPipelineSolver9_PreloadedTraceGraph Recorded timings: 16/16 samples (4 partial, including failed or timed-out samples).
Changed outcomes (1)
Workflow: View run |
AnasSarkiz
changed the title
|
/benchmark --same-machine --dataset 18 --pipeline 9 |
Same Machine Benchmark ResultsThe paired benchmark ended with cancelled before both reports were produced. Workflow: View run |
|
/benchmark --same-machine --dataset 18 --pipeline 9 |
Same Machine Benchmark ResultsBoth revisions ran sequentially in one Blacksmith job on Dataset:
Outcome changes: 0 improved, 0 regressed. DRC issues are totaled across solved samples. Timing percentiles include solved and timed-out samples; negative timing deltas are faster. Main pipeline stage timingsAutoroutingPipelineSolver9_PreloadedTraceGraph Recorded timings: 16/16 samples (3 partial, including failed or timed-out samples).
PR pipeline stage timingsAutoroutingPipelineSolver9_PreloadedTraceGraph Recorded timings: 16/16 samples (3 partial, including failed or timed-out samples).
Workflow: View run |
2 participants
Pedometer bug report 103 could finish port-point pathing but stall in high-density routing. Preserve outer-layer transit across an inner copper plane and an empty inner layer, then keep via endpoints connected through boundary cleanup, final repair, and SRJ export.
The original tiny-hypergraph pin c1043b3043ddf0c4d841fe5a6d9a515165960911 and UniformPortDistributionSolver behavior are restored. The local changes expose high-density progress, preserve terminal-wire continuity, and share the existing via conversion between repair and export. Routing budgets, timeout acceptance rules, and existing test assertions are unchanged.
Dependency pins:
The requested same-machine dataset 18 / pipeline 9 benchmark completed on 16 scenarios:
There are zero outcome regressions. Samples 6 and 15 fail and sample 14 times out on both revisions. These results replace the earlier broad-change benchmark that lost one completion and increased median runtime by 12%.
The benchmark compared main 109c67b with b24c94f. Current head b387501 differs only by line wrapping in the new regression test; production code, dependency pins, and benchmark scripts are identical.
Validation on b387501: build, TypeScript, format, added-code checks, and preview build pass. Eight of nine CI test jobs pass. The sole failing test is sample 3's SVG snapshot (27.571% difference); main 109c67b reproduces the failure and generates a byte-for-byte identical SVG (SHA-256 4e3da8d6a0d6d5e470b4e9d249bf5555ce3e9402055d752c2effec2d3728121d). Both pass the zero-DRC assertion. The unrelated snapshot is unchanged.
The original low-effort pedometer test remains unchanged. The new default-effort test completes all 1,106 high-density regions and validates via representation. The export and repair regressions fail without their fixes. Existing sample 4 work-budget, sample 8, sample 9 exact-output, and other formerly regressed board checks pass. Rectdiff has 78 passing tests and one existing skip; repair02 has 77 passing tests.
Still draft: a fresh default-effort, cache-disabled pedometer run completes with zero connectivity rollbacks, but has 13 relaxed-DRC errors: one trace-pad contact, four pad clearances, and eight same-net via clearances. Power expansion reaches its existing budget. This fixes the stall and via continuity, but does not produce a zero-error pedometer board.