These entities were drawn every frame wherever they were in the loaded
world: every armor stand, item frame and painting of every loaded village
or base went through its model, item and per-vertex occlusion test even
far behind the camera. EntityRenderer.shouldRender now decides, as for
mobs and minecarts:
- Entity.shouldRenderAtSqrDistance from the current position: 64 times the
bounding box's mean edge times the entity distance option (boats about
71 blocks, primed TNT 63, paintings by size); item frames 1024 blocks;
armor stands 256 times the box's mean edge whatever the option (4 for a
marker's empty box); tridents ten times the default (AbstractArrow);
firework rockets 64 blocks.
- the interpolated culling box grown by 0.5 in the view; tridents grow it
by 1.5 more (ThrownTridentRenderer), armor stands with a dragon head by
0.5 (LivingEntityRenderer), item frames holding a map use the block-wide
box.
The fishing bobber is unchanged: vanilla never frustum-culls it
(FishingHookRenderer.affectedByCulling) and Horizon only draws the
player's own, always within its 64 blocks.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The section occlusion search (SectionOcclusionGraph) ran on the render
thread whenever a section was meshed or the camera entered another
section: at render distance 32 that is a breadth-first walk over about
124,000 cells, and while exploring it ran nearly every frame.
As in vanilla 26.3, the full search now runs off the render thread. A
worker keeps its own copy of every section's face connectivity, fed the
changes with each job, and searches over dense arrays around the camera.
Frames use the last finished graph and never wait for the next. The search
starts at the top of the frame, as soon as the camera is known, so it
usually lands before the frame builds its visible list. A section meshed in
between propagates into the graph the frame has (vanilla's
runPartialUpdate), so a wall that opens shows what is behind it at once; a
section that closes goes at the next full search. A graph that never
reached the camera's section (after a teleport, say) is not used: until the
next one lands every section in the frustum is drawn.
The search itself is unchanged: the same box, steps and first-entry rule,
checked cell for cell against the previous code in a test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
light_chunk's sky and block spreads read LightTables::cells for every
neighbour, and the fill and both spreads dominated lighting. Pack each
cell into two bytes (occlusion type and dampening) with a per-type
blocking mask, step neighbours by linear offset with an axis bounds
check, and precompute the per-column present table.
1156-column probe: 1.81 -> 1.31 ms/col (~28%). Vanilla light parity is
unchanged (1,179,648 sky/block values, 0 differences).
entity_motion_events looked up each moving mob in a SipHash map every
tick. The keys are mob ids the server gives out, so the fast hasher is
safe there.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Their particles, ambient sounds and the turning display mob are for a
window. A dedicated or integrated server has none, yet made a display
mob for every spawner it ticked.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The list kept its slots for the whole tick; sweeping it every tick cost
two hash lookups per ticker. Now it is swept only when a slot was freed
or a block entity was taken out and not brought back.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
BlockSection::get and set check the section index and the state id once,
then read and write through new PackedStorage::read/write, which skip the
storage-level bounds and value checks and the Result that get/set keep for
external callers. A section index is below 4,096 and a palette id always
fits the width bits_for_palette chose; both methods debug_assert that
contract. Used for the hot read, the hot write and the growth re-encode.
Benchmark (ignored tests, best of alternating-order rounds, fast profile):
BlockSection::set 13.15 -> 9.12 ns/write (1.44x), get 2.33 -> 1.45 ns/read
(1.61x). See docs/performance/paletted-container-access-results.md.
World::block_entities keeps vanilla's Level.blockEntityTickers beside the
map: a slot when a block entity whose state has a ticker is added, kept
when it is replaced, freed when it is removed, unloaded or its state loses
the ticker; a loaded chunk's in its block entity map's order. The tick
visits that list once, in block-ticking chunks within the world border,
instead of scanning the whole map once per kind.
New harness block_entity_tickers: real LevelChunks of a server level
against BlockEntities.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: merged by Ludovic's AI agent. CI parity passed (harness paths stay synchronous). Smoke test of the new path on current main (macOS, --offscreen, normal launch): resources loaded on the worker in 9.3 s, renderer swapped, title screen up and running for 60 s, no panic. Thanks @rdbtCVS!
The NBT tables stream, the fluid tables reuse the game's facts and the content
root is kept between starts (#753, #754, #756). Horizon with ATM10's 40 data-tier
mods against the same mods on NeoForge 21.1.259. Worldgen carry-over and what it
does not cover yet (#763, #764). Chipped's harness row with its self-drop tables
(#762). The overlay cache line said it was cached; it is now.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
BlockSection::set resolved the old cell through get and then probed the
section palette three times per non-trivial write: keying the old state by
id, testing the incoming state and finding it. The slot at the written index
already is the old state's id, so read it straight from storage and resolve
only the incoming state, with a single map probe. Palette reuse, free slots
and growth are unchanged, so palette, counts, free list and NBT are identical.
PalettedContainer::unpack and unpack_u16 matched the palette variant inside
the per-cell loop; dispatch Single/List/Global once before it.
Measured against copies of the pre-optimization code (ignored tests):
BlockSection::set 12.73 -> 11.37 ns/write (1.12x), unpack list 1.956 ->
1.860 us/call (1.05x), global 2.255 -> 2.172 (1.04x). See
docs/performance/paletted-container-access-results.md.
set_matches_baseline_under_random_edits drives both implementations over the
same 50k-edit streams for 1/2/16/17/24/64 distinct states, with periodic
whole-section collapses, and asserts equal results, bits, palette, counts,
free list and NBT.
Structure mods build on libraries whose structure, placement and pool element
types are vanilla's with extra rules: Lithostitched's, Moog's Structure Lib's
and Repurposed Structures' jigsaw structures, their advanced random spreads,
versioned and ocean-bottom single pool elements. Carry rewrites them as
vanilla's types; the extra rules (no start in liquid, a minimum distance from
the origin) are lost, so a structure may start where the mod would not. Types
with no vanilla counterpart (YUNG's, Formations', the nether variants) are left
as they are and counted in the carry report.
The structures' biomes name the loader's convention tags (#c:is_plains). fetch
now also gets the pack's NeoForge jar from NeoForge's Maven repository, checked
against its SHA-1, and carry copies the convention biome tags each mod's
structures use, with those they include, into the mod.
ATM10's 42 carried mods: 1,061 files rewritten, 103 convention biome tags.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Chipped registers its blocks' loot in code: its jar ships 7 loot tables for
about 7,000 blocks, so carried over they dropped nothing, while in the NeoForge
oracle every one drops itself. When a declared block has no loot table in the
jar and the oracle shows vanilla's dropSelf (one of itself with every tool,
nothing only where the state needs a tool it did not get), carry writes that
table in 26.3's shape. When only the states with one property value drop it
and the rest drop nothing (a door's lower half), the entry gets vanilla's
match_block condition on that value, as createDoorTable writes. Chipped:
6,843 tables added.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The nether-portal section precheck from #661 answered "does this section's
palette hold a portal?" per section, per tick. Keep the answer per resident
chunk in `portal_sections`, keyed by a weak handle: an edit or a chunk
replacement rebuilds it, and it is dropped when a chunk unloads and cleared
when the block facts change, so a portal block always forces the sweep. The
entity sweep now finds portal blocks without the per-tick map.
This is a scan microbenchmark, not a whole-game claim: steady improves 2.4x
to 17x over the fixtures and the run parity-checks the cached and per-tick
answers. Rescanning every chunk every tick (cold, reload, edit_all) is slower
because a miss reads all 24 sections of a chunk; see
docs/portal-scan-benchmark.md.
Halves the CPU of deflating a chunk column and makes small packets about
20 % cheaper. The compressed bytes change; what they inflate to doesn't.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
getFiddledDistance restarts a six-step LinearCongruentialGenerator mix for
each of the eight quart corners. Its first two steps depend only on the X
and Y corner offsets, so quart_at_block now mixes the two X states and the
four XY states once, then runs the remaining steps across the eight corners.
next(m, salt + offset) == next(m, salt) + offset, so a coordinate folds into
its step with one addition.
Bit-identical to BiomeManager.getBiome and getFiddledDistance; a reference
oracle test pins the equivalence over a dense grid and wrapping extremes.
🤖 Agent: merged by Ludovic's AI agent: perf-server-5. Before/after in the PR body; agent parity 87/87 stacked; orchestrator combined check with the perf-ticks-5 stack on current main: clippy clean, client+protocol tests pass (one flaky world-delete test, #749), dedicated_server/lan/chunk_stage/block_entity_storage PASS.