Repository navigation
h12b: 45-degree kinds, stage 2 continued - #36
Merged
Merged
Conversation
The enclosure modes the foundries' KLayout decks check by edge pairs (enclosing without touching edges, enclosed and enclosing by projection) need KLayout's pair semantics on 45-degree edges: which pairs count at all, how each edge is cut to the other's inner side, and the near part each keeps. relation::pair_threshold2 ports EdgeRelationFilter::check (default angle_limit, no projection bounds) with euclidian_near_part_of_edge and var_near_part_of_edge, and returns the least distance, twice squared, at which a pair is reported. KLayout computes in floating point; here every crossing of an axis edge with a 45-degree line is a lattice point and every distance exact. projected_reach2 reads whether a reported part reaches a corner (sky130's adjacent-sides idiom). derive::inside_flags exposes inside_into's covered-polygon test. Oracle: differential against narrowest_pair2, itself held to KLayout by the nightly mirror: the width read pair by pair through the port equals it on random merged octilinear shapes over 300 named seeds. A port that misreads a cut, an angle, a zero-distance mode or a near part disagrees on some figure.
min_enclosure, noncoincident_enclosure, asymmetric_enclosure and adjacent_enclosure refused any layer with a 45-degree edge. Where either layer has one, the rule now hosts each inner polygon in the merged outer figure that covers it (inside_flags) and reads: - every side: the Euclidean distance between the boundaries, exact; - non-coincident, opposite and adjacent: the edge pairs the foundry decks check them by (relation::pair_threshold2), so an inner edge also pairs with a host edge at 45 degrees to it. Opposite and adjacent take rectangular inner shapes (cuts), as the decks do, and refuse one with a 45-degree edge. full_width and an enclosure with a wire band still refuse: both cut their layers into rectangles. Rectilinear layers take the old path unchanged. The octilinear KLayout differential gains a cut layer (squares placed in met1 shapes' boxes, never meeting, as the mirror's enclosed reads their merge) and the four kinds, plus min_enclosure and noncoincident_enclosure of 45-degree met2 in met1. Oracles: hand-derived cases in tests/drc_octilinear.rs (a square `sqrt(1250)` from a chamfer, a diamond, an unhosted square, a triangle on a chamfer, a square read past chamfers), each under the eight grid symmetries; the KLayout mirror on 1500 random octilinear layouts, which agrees on every layout (min_enclosure flagged on 2736 runs, noncoincident 2704, opposite 1062, adjacent 1042). Goldens of the three 45-degree fixtures record the enclosure rules as ran. A host edge ignored, a misread pair angle or a missing near-part cut fails the hand cases or the differential.
Reading IHP's sg13g2_stdcell.gds whole, where every top cell sits at
the origin, tripped decompose_into's debug assertion ("a polygon's
rectangles cover exactly its area", 5563800 vs 2922600). The
decomposition was right; the polygon reaching it was not.
sg13g2_dfrbpq_2's Activ is one keyhole BOUNDARY around a hole; the
reader splits it into its outer and the hole, stored one after the
other. Validation then bound each hole to the smallest-boxed outer
holding its first vertex, and another cell's Activ drawn over the hole
(holding its corners, not its middle) won: that shape gained a hole
it does not cover, so its even-odd fill exceeded its claimed area,
and the keyhole lost its hole.
A reader stores a polygon's holes right after its outer (a keyhole, a
boolean's result), so validation now binds a hole to the nearest
outer row before it when that outer holds it, and falls back to the
innermost outer holding it only otherwise.
Oracles: construct-from-answer. A notched square stored before a
100-square and its hole: the hole is the 100-square's and both
decompose to exactly their areas (geom test); the same reduced from
the GDS, a keyhole written after the overlapping shape and read back
(tests/test_all.rs), which trips the original assertion without the
fix. The whole library now loads and runs every rule of
pdks/ihp_sg13g2.deck in a debug build.
must_be_inside and must_contain refused any layer with a 45-degree edge. Where either layer has one, both layers are merged and read exactly: an inner figure is held where the outer covers it (inside_flags) and counted in the figure holding it; the area an inner figure leaves outside is the sum of the outside parts (subtraction_into) inside it, exact, reported rounded up so half a unit outside reads 1. Rectilinear layers take the old path. The octilinear KLayout differential gains inside(met2, met1), inside(via1, met1) and contains(met1, via1) >= 1. Oracles: hand-derived (a diamond inside an octagon, one halved by its chamfer, outside 100; a half-unit triangle outside, reported 1; one diamond held where two are asked) under the eight grid symmetries; the KLayout mirror on 1500 random octilinear layouts agrees on every one (must_be_inside flagged on 2336 runs, must_contain 1297). The full local suite passes; CI was not waited for (stopped by the coordinator).
Contributor
Author
|
Landed on verification-hardening (workspace tests run locally; merge tier green on ab6b9f1). |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
H12 stage 2 for the remaining geometric kinds on 45-degree geometry, kind by kind (enclosure first), each with hand-derived cases under the eight grid symmetries, the KLayout mirror on random octilinear layouts and the foundry decks; then stage 3 (DRC-116). Commits say per kind what measures 45-degree shapes and what still refuses.