Apple Silicon support for texture compression - #284
Open
rromanchuk wants to merge 2 commits into
Open
Conversation
|
Curious about pulling this PR into my forked version of Ortho4XP. Does anyone happen to know why nvcompress is used instead of DDSTool? I suspect there was a reason since it's been around since 2019. |
Collaborator
Maybe because it used to be faster back in the days (due to CUDA support) and accepts JPG as an input format? But that's just a guess. |
Ah, good point - looking at the readme for DDSTool, I see a note with the latest version 1.4 from 5/07/24 mentioning it can accept any format (png, tif, jpg, bmp). I wonder if prior to that it didn't accept jpg which would make sense. As for CUDA support, also makes sense. However, someone just made a PR in my forked version to add a newer version of nvcompress for CUDA support so I'm wondering if it's broken in this older version. Anyways, I think I may just switch to DDSTool for macOS since CUDA support is not applicable and use the newer version of nvcompress for Windows. I need to figure out if there's a way to get an updated version of nvcompress for Linux users to take advantage of CUDA. |
shizumaat
added a commit
to shizumaat/Ortho4XP-novemberlima
that referenced
this pull request
Jun 15, 2026
…ension User ruling (2026-06-10): the 1.5% cap applies along the taxi CENTERLINE; cross-axis diagonal chords are unregulated (ICAO Annex 14 / EASA CS-ADR-DSN.D.265/.280) and inside-of-curve edges legitimately exceed the cap. All-pair/visibility chords had pinned high-speed exit junctions flat (HECA oscarpilote#282 could not rise toward A4). Built: - _PER_AXIS_JUNCTIONS=True + the per-axis rule applied where junction constraints are actually built (_build_shape_constraints visibility path; the flag previously only touched the unused _build_edges path): along-axis pairs cap at the ARC between projections, cross-axis diagonals drop, ring-adjacent pairs survive. - Curve-aware corridor distances: _junc_axis_arc supersedes shorter geodesics in _junc_geo_dist (hard bands, twist clamps, ties). - RUNWAY-EXIT EXTENSION: chain termini at runway-touching junctions get a virtual HARD station at the runway contact; contact = the qualifying centerline's nearest runway intersection (extending ingest-cut ends straight) or the fan's THROAT vertex by in-junction geodesic when the apt.dat line cuts the corner (A4's line: 40 m in-junction where the fan flow runs ~190 m); value = runway edge interpolation; the virtual gap is recorded as a twist line source so the junction interior is painted; singletons survive when extended; extended singletons skip the anchor relax (it crushed A5 flat and left 3.9 m at the oscarpilote#284 seam). MEASURED (HECA): junction oscarpilote#282 now carries the climb (59.4 throat -> 62.6 at A4's mouth, the user-specified behaviour); A4 7% -> 1.4%. NOT shippable yet: A5/oscarpilote#283 unchanged (its far-mouth nodes were already relief-crushed to 60.5 pre-corridor - different mechanism), oscarpilote#284 3.9 m, new oscarpilote#302 3.7 m / oscarpilote#192 1.2 m, and the 05C dip regressed 107.9 -> 110.4 (virtual anchors changed the demand measurement). Parked for tuning. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
shizumaat
added a commit
to shizumaat/Ortho4XP-novemberlima
that referenced
this pull request
Jun 15, 2026
…8.7) Diagnosed with full demand instrumentation (the 'regression' was three separable bugs, not a design flaw): 1. THROAT fallback gated to SINGLETON exit stubs: HECA's diagonal runways made T4+U read tangential and the 216 m throat budget dissolved its runway-flex demand (05C 107.9 -> 110.4). Multi-rect through-corridors keep the axis contact. 2. Virtual-blocker demands carry the CONTACT need to nearby runway vertices (<=60 m) instead of adding the contact->vertex leg, which diluted depth (109.6 -> 108.7; matches the user-validated p9 semantics: the dip centres at the contact). 3. Route-reach bands relax to the chain's own curve-aware distance at stations of virtual-anchored chains (the reach graph cuts the same curve corners as the exit centerlines and re-clamped the climb the virtual granted). 4. Extended-singleton far termini try to anchor at the far junction's uncarved side (max, chain-feasibility-clamped) - NOT yet effective for A5: its entire far junction reads carved (~60.5), so the 64.4 uncarved level lives on a different shape; next probe = map A5's far juncs-index to the shape that holds 64.4 and anchor against it. State: 05C 108.7 (user ~108), 05L untouched 57.9-60.7, A4 1.1% with oscarpilote#282 carrying the climb. Open: A5/oscarpilote#284 3.7 m, #350 1.9 m, #317 1.6 m. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
shizumaat
added a commit
to shizumaat/Ortho4XP-novemberlima
that referenced
this pull request
Jun 15, 2026
…root cause) The oscarpilote#284/oscarpilote#232 A-complex cliff was NOT unmerged same-ref chains: the two A chains are 1.86 km apart with apron oscarpilote#190 carrying the route between them. The high chain's SW terminus is an APRON MOUTH (junc=None, ring nodes shared with the apron) anchored at the apron edge's DEM-settled value (64.67) at station construction; nothing downstream could move it (ties form chain-to-chain only, anchor self-consistency is intra-chain, the freeze-veto's only escape requires a runway blocker) so it froze out the A5-complex consensus tie (61.76, route ceiling 62.4 at the junction side) and held the 63.3-vs-60.7 junction cliff. Fix: in the tie freeze, when every blocking anchor is a non-hard UNTIED terminus whose mouth opens into a free apron, project it MINIMALLY into the tie's cap reach and accept the tie. Projection, NOT un-anchoring: dropping the anchor let HECA J's tail (0.33 m infeasible) fall 4.7 m to the flat extension - a manufactured 5.3 m wall against G2's legitimate writes in the same apron. Measured (HECA, new apt.dat): oscarpilote#232/oscarpilote#233/oscarpilote#228 A-complex cliffs cleared, A profile 62.5..61.8..63.4 with the tie landed, 05C 108.7 / 05L 57.9-60.7 / A4 climb / A5 flat all held. CYXY relax never fires (10 pre-existing exit-junction viol unchanged), SPJC untouched (R1/R2 residual), suite 305p/4f = branch baseline. Remaining: apron band-noise residue (free apron vertices pinned at route-band floors ~0.6-1.7 m above moved corridor writes - the p4/p7 band-pinning class, needs the junction band_exempt treatment extended to aprons near corridor writes). Also: chain-ends debug print (head/tail junc+mid per multi-rect chain). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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
3 participants
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.
Remove dependency because:
Usage:
Flags: