Rewrite LODSphereMesh as adaptive chunked-LOD single-draw renderer#2618
Draft
levinli303 wants to merge 9 commits into
Draft
Rewrite LODSphereMesh as adaptive chunked-LOD single-draw renderer#2618levinli303 wants to merge 9 commits into
levinli303 wants to merge 9 commits into
Conversation
levinli303
force-pushed
the
lodsphere-chunked-lod
branch
from
June 30, 2026 23:19
918609e to
d815b4d
Compare
Port the cube-sphere adaptive LOD and single-draw-call batching into the lat/long LODSphereMesh, adapting the cube blueprint to a single equirectangular (longitude, latitude) parameterization: - Quadtree roots at 2 longitude cells x 1 latitude cell; screen-space sagitta error drives splitting, with 2:1 balance and one-level stitch templates sealing LOD seams in a single batched draw per texID group. - Longitude edges wrap across the meridian seam; pole edges have no neighbour and are left unstitched (each pole is a single point, the collapsed triangles are degenerate and harmless). - resolveTile reproduces the legacy renderSection texcoord transform, so leaves map directly onto VirtualTexture tiles without a new texture type. - render() takes eyePos, pixelSize, enableHorizonCull and geometryScale; cull bounds and the LOD metric are scaled by geometryScale so the cloud and atmosphere shells (scaled in the modelview) are not over-culled. - Optional macro-gated wireframe mode mirroring the cube-sphere renderer. Plumb pixelSize through RenderInfo and update the surface, cloud and atmosphere call sites to the new signature.
levinli303
force-pushed
the
lodsphere-chunked-lod
branch
from
June 30, 2026 23:45
d815b4d to
78a26c9
Compare
Divide the screen-space split error by the grazing cosine so limb chunks, compressed tangentially into thin screen slivers, refine while head-on chunks stay coarse. This targets the hard lit/sky edge aliasing without globally lowering the pixel-error threshold and over-tessellating the front face. Also fold the chunkRadius/edge reductions into single std::max initializer lists and hoist the edge sqrt out of the per-corner norms.
levinli303
force-pushed
the
lodsphere-chunked-lod
branch
from
July 1, 2026 04:52
6ae3203 to
2954633
Compare
- render(): extract cullLeaves/buildBatch/uploadAndBindBatch/drawBatch to cut cognitive complexity from 60 to within limit. - ensureStitchTemplates(): extract buildStitchTriangles (and the wireframe buildStitchLines) file-local helpers to cut complexity from 33. - bufferOffset(): annotate the GL-mandated void* return with NOSONAR. - LODSPHERE_WIREFRAME: keep the debug macro but mark it NOSONAR.
levinli303
force-pushed
the
lodsphere-chunked-lod
branch
2 times, most recently
from
July 1, 2026 05:18
104ddfb to
6a42249
Compare
MSVC rejects using the constexpr locals N/stride inside the vid/snap lambdas without an explicit capture (C3493), while capturing them trips Clang's unused-capture warning. Reference the namespace-scope CHUNK_RES directly so no capture is needed on either compiler.
levinli303
force-pushed
the
lodsphere-chunked-lod
branch
from
July 1, 2026 05:19
6a42249 to
106db31
Compare
The batch buffers are rebuilt every frame from every visible chunk. appendChunk grew them with per-element push_back and no reserved capacity, so it repeatedly reallocated mid-frame and paid a capacity check per float (a profile hot spot). Reserve the exact vertex total (constant per chunk) plus an index upper bound once in buildBatch, then write through raw pointers in appendChunk. Also raise CHUNK_PIXEL_ERROR 0.5 -> 1.0: the grazing-angle term now sharpens the silhouette on its own, so the front-face budget can loosen without the limb degrading (effective limb threshold is CHUNK_GRAZING_FLOOR * this).
Replace the per-leaf root-descent balancing and edge-mask stitching (coveringDepth/neighborOf/computeEdgeMask/gatherLeaves) with a single top-down walk that carries each node's four same-level neighbour indices. Child neighbours are derived O(1) from the parent context, and imbalanced leaves are collected to a fixpoint split list rather than re-descending from the roots each query. QuadNode gains an int depth; LeafRef/balanceScratch are removed in favour of a reused splitList. Cuts balanceLeaves inclusive cost from ~4.1% to ~2.4% of frame time on the close-surface benchmark.
Add a flat, uniform-LOD path (setTerrainEnabled=false by default) that skips the adaptive quadtree, 2:1 balance and stitch when no terrain displacement is used, matching/beating the pre-chunked renderer on close-surface views while keeping the adaptive path available for future terrain. Also resolve SonarCloud issues on the PR: C-style arrays -> std::array, if init-statements, const member functions, an error-prone loop, a structured binding, group batch/draw members into a Batch struct (<=20 fields), and bundle renderCloudsUnlit cloud params into a struct (<=7 params).
Drop the single-use wantSplit init-statement variable (flagged as assigned but never read); test the split predicate directly in the if.
|
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.
wip