The build's placements ask exact heights over the whole map before any ring exists, and through the
tiles that read the file ~25 000 times in one frame. The copies now stay after the cut and the game
calls terrain_tiles_build_done() when the world is built (the end of world_things, and after a swap's
terrain_reload); every answer is the same before and after, from the copy or the tile. The GPU half
merged here (r3d/tp-rt, r3d/tp-shd: the page table, the pool and the shaders) is unchanged by it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
At the cut the whole 4096 height, normal and photograph textures go, replaced by a coarse
2048 level (the CPU's tt_coarse uploaded; the normal and photograph blitted down on the GPU,
the photograph mipmapped) and a pool of fine tiles in three array textures - heights, normals,
photograph, each layer a tile with a one-texel border - addressed through a tt_n^2 page table
(u_tp_page: slot + 1, 0 = the coarse level). The pool holds the tiles within the reach (900 m,
or the fog wall's when nearer) and a ring, and is made again when the fog wall changes its size
(terrain_pages_fog). Once a frame (tp_frame, beside tt_frame) tiles past the reach and two tiles
go and the wanted ones come in nearest first, 8 a frame, written into a staging buffer kept for
the process (two halves, one per frame in flight) and copied into their layers inside the frame's
own command buffer - no submit of their own - with the page table re-uploaded only when it
changed. tp_bind binds the pool (or 1-layer stand-in arrays while paging is off, u_tp_on = 0) for
every program that reads the ground: terrain_bind_height (models, scatter, shadow_bind, grass),
terrain_bind_prog, the sun pass and the shadow bake. grass_cull.comp reads through the same page
table. R3D_VKMEM prints the pool's line. Tiles off, nothing changes. Compile-only: not run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Most of a build was one clang -O2 on one .ll (the game: 32 s of a 41 s headless build, one core).
Both paths that assemble - the CLI's (build.ludic: ludicc --emit-llvm, then clang) and ludicc's own
(-o, which ludic bundle and the examples use) - now cut the program's IR into N parts with llvm-split
(externalizing what the parts share), compile them with one clang each in parallel (-x ir -O<opt>
-mmacosx-version-min=11.0, the link's own clang taking the objects where it took the .ll), and remove
the parts and objects after. N is $LUDIC_JOBS, else min(cores, free GB / 1.5).
It needs an llvm-split and a clang of the same LLVM (Homebrew's LLVM 22 writes attributes Apple's
clang 17 cannot read): $LUDIC_LLVM, else /opt/homebrew/opt/llvm/bin. With either missing, on Windows
(its shell cannot run the parts at once yet), with LUDIC_SPLIT=0, or when a part fails, it compiles the
.ll whole as before.
$LUDIC_OPT=1 is a developer's faster build; ludic bundle sets LUDIC_OPT=2 for its compile whatever the
shell says.
Measured before the compile-only rule (this Mac, 12 cores, one build at a time):
- the game headless: 37-41 s -> 13-15.5 s (8 parts / by free memory), peak 2.1 GB -> 1.0-1.1 GB;
- the lab headless: 43.1 s -> 12.7 s, peak 2.4 GB -> 1.0 GB;
- the game at LUDIC_OPT=1, split: 11.8 s (fps cost not measured).
Both built and linked clean; the goldens and a headless shot of the result are not run here.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Within a pass a row writes only its own still-masked texels and reads only neighbours already let go,
which no row writes that pass, so the result is the single-threaded one. The worker takes a DilateJob
of plain buffers and allocates nothing. tex_dilate_bytes (safe_api) and examples/rendering/dilate.ludic,
which checks it against the old loop on RGB and RGBA atlases of sizes that do not divide (DILATE OK).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
terpage.glsl is the reference block (tpSlot, tpUV, terHeight, terHeightSmooth, terNormalXZ,
terOrtho, tpOrthoRes, tpOrthoLod), pasted by section into terrain.vert, terrain.frag,
tersun.frag, model.vert, grass.vert and grass.mesh. u_tp_on = 0 reads the old samplers with the
old coordinates and filtering; on, a resident tile is read at level 0 from u_tp_h / u_tp_nrm /
u_tp_ortho, anything else from the coarse map now bound under the old names. The B-splines use
the FULL map's texel and take every tap through the page, so a tile edge stays one surface;
blurred photograph reads (lod 1-2.5) stay on u_ortho with the level moved down by the coarse
map's ratio. The fragment stages drop their unused u_height. SPIR-V rebuilt: terrain programs
carry 23 samplers (21 in the fragment stage), up from 19.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A switch a library reads only as a process starts - libmalloc's MallocLargeCache, which on a game
keeps ~200-300 MB of freed load buffers - has to be in the environment before main. For a .app started
from Finder, the Dock or `open` that is Info.plist's LSEnvironment; `app env` is repeatable and each
K=V becomes a string in it. bundle_case's probe app carries one and checks it with plutil.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The volumetric pass marched 800 m in a fixed number of steps whatever the wall: it now stops at the
wall with steps in proportion (at least 8) - 0.57 -> 0.33 ms at 30 m by the pass timers. SSGI skipped
only past 900 m; it now also skips past 0.85 of a wall, where the fog is solid (u_ao_far). applyFog
samples the fog's colour only where its weight is above zero. A card layer whose casters inside the
wall are 80% or more of it (a kilometre's wall) casts from its static list and is not refiltered and
re-uploaded every 4 m of camera motion. Fog off: the frame unchanged (0 pixels over 8).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
terrain_tiles_to(dir, key) before a map is made: once made, its heights, photograph and baked normals
are written tile by tile (64 m, TT_TEX) to <dir>/<key>.tiles - fresh every time, header last, so no
other generator's or a torn file is ever read - and the whole copies go. Every read goes through the
tile: resident, else read from the file there and then into a 2048-tile clock, so terrain_height,
terrain_height_smooth, terrain_ortho* and terrain_chunk_heights answer exactly what the whole copy
did (R3D_TT_CHECK: worst 0.0 m over 4000 points) whatever is resident - two machines and the boot's
placements agree to the bit. terrain_chunk_heights takes render3d_st mut for it.
terrain_height_near(read-only): the tile if it is in, else a coarse 2048^2 level (worst 0.91 m), never
the file - for line checks that must not take render3d_st mut. terrain_texel() is the texel size,
terrain_tiles_prefetch(x, z, r, budget) reads ahead, and a frame that reads more than 8 tiles says so
once. R3D_TERRAIN_TILES=<dir> turns it on for a run.
At the overlook with MallocLargeCache=0: 1731 MB off, 1658 MB on; 192 MB written in ~200 ms; the
frame unchanged (0 pixels over 8).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The DEM (R16 with mips, 44 MB) was read only by terrain_generate and kept for the map's life; a
world swap loads its own again. The height-field sun shadow was one RGBA32F for three values and an
unused fourth: the lowest lit height stays 32-bit (a receiver 3000 m up compares against it to a
quarter metre), the occluder distance and the cloud mask go beside it in RG16F (64 -> 32 MB), baked
in a pass of their own (TS_AUX): a pipeline takes one colour format for all its targets, and drawn
together into R32F + RG16F the writes went nowhere and nothing was lit.
-75 MB at every fog level, off included (2008 -> 1933 MB at the overlook); the fog-off frame is
unchanged (0 pixels over 8).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A 1 km fog gave a 1040 m zone, more of the world than no fog's 900 m (2405 MB and 1029 nav tiles against
2294 MB and 753 in the fog profile). zones_radius is min(max(fog + 40, 150), 900); the test holds 1 km
and 860 m at 900.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The ring held 2 x 64 MB of host memory for a frame that uses 3 MB at most (2761 draws at the
overlook, 1.7 MB in town): it starts at 16 MB a half, and a frame that ever runs out grows it for the
frames after (gvk_ring_grow, making the kept sets again; R3D_RING_KB=<n> starts small to watch it).
-95 MB at every fog level, off included.
A streamed layer's arrays and its stream's arena are made for the reach a fog wall leaves (twice its
area's share, at least 4096 instances) and emptied to refill within the frame budget
(fog_streams.ludic). The near streams' reach is already inside most walls, so it is -9 MB at 30 m.
R3D_VKMEM adds the ring's high-water mark, the largest buffers and the streamed layers' share.
Footprint at the overlook: 2008 MB off (2108 before this phase), 1748 MB at 175 m, 1682 MB at 30 m.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An empty world at the game's size held 51839 KB of Jolt's heap (200000 bodies, 65536 pairs, 20480
contacts, a 32 MB temp heap), the jolt= floor under every walk. Pairs and contacts come from what
moves - drops, boats, a few walkers - and a valley's still things make none. Now 11743 KB at 200000
bodies and 8461 KB at the 98304 the game opens for (jolt_floor_test holds it under 16 MB). The temp heap
still falls back to malloc for a step that wants more.
lib/macos-arm64 rebuilt; lib/windows-x64 needs native/build.sh on the PC (jph_world_new's new sizes).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A cascade that begins past the wall was fitted and cleared every frame for nothing: the array is now
made with the ones that begin inside it (2 at 30 m, 3 at 175 m, 5 without a fog) and made again when
the wall changes that count; shadow maps 80 -> 32 MB at 30 m, 48 MB at 175 m. A layer the array lacks
clamps to its last, read only past the wall where everything is fogged.
The sky: at walls of 110 m and nearer the whole sky is the fog's colour, overhead too, with a soft
glow at the sun or moon and no stars or clouds through it, easing out by 175 m; from 175 m on the
horizon band alone, and off as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Inside the wall every instance is a mesh whose LOD2 casts its shadow, so a card is neither drawn nor
cast there: fog_impostors.ludic frees a layer's atlases while wall + margin < its near (trees 420-480
m, rocks 320 m) and paints them again from the model when it opens past it (impostor_paint, split out
of the bake). Impostor atlases get memory of their own, so a release goes back to the driver instead
of leaving a hole in a shared block. R3D_FOG_AT=<frame> with R3D_FOG_TO=<m> changes the wall mid-run.
At the overlook: 2102 MB off, 1824 MB at 30 m (249 -> 67 MB of atlases; the rest is the ground cover's
card atlases, drawn inside the wall); opening re-bakes 16 layers in 42-46 ms, back to 2102 MB.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- The arrows, a press on the select and the pad step past a disabled option (ct_opt_step), wrapping
as before.
- A value naming a disabled option is shown as the nearest enabled one (the lower of two as near) and
set to it once the build is done: an event fired while the tree is built is cleared with the frame's,
so the clamp is queued (ct_clamp_n / _i) and fired after nt_fired_clear.
- The < or > whose next option is disabled is drawn disabled (its part's :disabled), so a cycler shows
where the options stop being on; it still steps past them.
- ct_build_select moves to controls_opts.ludic with the rest (controls_build.ludic 119 -> 104 lines).
Golden ui_select_disabled: options 3 and 4 disabled, a value of 4 clamps to 2, > is drawn disabled
there, > wraps to 0 and < from 0 steps back to 2. The 38 ui examples pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every allocation carries the owner its maker was wrapped in (models, terrain, impostor atlases,
scatter, grass, shadow maps, frame targets, sky, water, game textures, made in a frame), and the
report prints each owner's images and buffers, the 25 largest images and the scatter layers',
streams' and terrain's CPU copies. Nothing drawn changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
zones_set / zones_clear / zones_merge, zones_contains / zones_nearest and the clusters' bounding circles,
over fixed arrays of ZONES_MAX (16). WildlifeWorld.active freezes an animal outside (no state change, no
dice), NpcWorld.active a walker (and no birth outside), VehicleWorld.active a moored boat, and
ThingsWorld.restocks keeps a morning's restock inside; every default is "everywhere", so an unbound game
is unchanged. Tests: the merge (near players one cluster, far two, a player leaving splits a chain),
contains, nearest, the radius; wildlife, npc, vehicles and things pass as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
For loading the world by each player's fog rather than a fixed 900 m. keep_near_test: a zone round
one corner brings in its tile, a bigger one its neighbours, an edge nudged back and forth twenty times
changes nothing, and a zone moved 2 km away lets them all go.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Whatever is cut at the wall is now cut inside full fog: at 0.3-1.0 the last metres before the cut
were 80-95% fogged against fully fogged ground behind, which drew a line of tufts at 70 m and a
dark band of blade tips at 30 m. The blades thin out while the fog is still coming in.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
cam_sphere_visible refuses a sphere wholly past the wall (terrain patches, scatter cells, stream
chunks, grass tiles, water), actors past it are skipped for draws, casters and outlines, the grass
reach and the streams' generate-and-gather reach end at it, and the card shadows cast only from
the wall's share of a layer (fog_casters.ludic, rebuilt every 4 m). r3d_beyond_fog is exported for
the game to skip animating what will not be drawn. Render-only: no query of the ground or the world
changes, and off is the old frame (0 pixels over 8 against 160ce96, twice per side).
Draws per frame at the overlook: 2757 off, 1104 at 175 m, 484 at 30 m.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Master's windowed fog profile failed the fence at walk t 35 s: +512 B from value_new / value_put /
value_slot / value_lists under bd_class and cmp_keycap_model - a KeyCap first shown mid-walk.
It is a first build, bounded, not a per-show leak: an unmounted instance goes to in_free and the next
show of its class reuses it (in_reuse, `renew`), its props and model objects kept and filled in place.
Only a NEW instance makes them - bounded by how many of that class are up at once, and the ones
unmounted less than two builds ago. in_reuse already declared the record; the props and model objects
and their first fill, made afterwards in bd_class, were not. They are now: in_reuse marks a new record
`fresh`, and bd_class's first fill of it goes through bd_first_fill (@alloc_ok) - a reused instance's
fill stays judged, so a real per-show leak would still fail.
Golden ui_first_show: a component first shown at frame 700, once the fence is judging, then hidden and
shown twice more: bad 0 (the toolchain before this fails frame 701: +464 B value_new from value_slot
under cmp_cell_model, value_put grow - the profile's failure). The 37 ui examples pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every lit program blends toward the horizon colour (the prefiltered sky with the sun's inscatter,
so it carries the day's grade); the sky's horizon band is pulled to the same colour, wider the
closer the wall. The far plane, scatter and stream cull reach and the shadow cascades are clamped
to the wall; the reflection pass shares the shaders and the cull. 0 is off: every branch is gated
on u_fog_wall > 0 and r3d_reach hands back the far it was given. R3D_FOG_WALL=<m> for a shot.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The valley self-test died after the Lamar swap, loading Maroon's pine LODs:
gvk_fr_blk (the device suballocator's free ranges) grew past its @max(4096).
Nothing failed to coalesce: a range was carved at the buffer's bare size, so
what was left after it started unaligned and the next buffer, rounding its
start up, left a sliver before it that no later buffer could use. One host
block held 3,968 small buffers and 3,718 free ranges between them - 496 KB
free in all, 133 bytes a hole. Those merge away when a neighbour is freed, so
nothing leaked; the count simply rose with live buffers.
gvk_mem_new now carves `span`, the size rounded up to the alignment, and
records it as the allocation's length, so a free gives back exactly what was
carved and the rest of a range always starts aligned.
The same repro (the lab's `tests` scene, headless, a fresh test root): the list
never reached 256 slots (it passed 4096 before), the run completes (exit 0),
LAMAR OK and THINGS OK. r3d_small was 0 throughout, as ECS said.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A small window (r3d_small: the launcher's UI process) draws its interface flat
and tests no depth, but gvk_screen_make gave it a 32-bit depth image the size
of the screen - 8 MB at 1920 x 1080. In small mode the screen has none: the
pass and a clear with no depth attachment already leave depth alone (clearing
DEPTH_BUFFER_BIT with no attachment is skipped), and pipelines follow the
pass's formats.
With the launcher's own 1920 x 1080 backdrop (maroon-lake game/launcher-small),
the launcher window measured 378 -> 295 MB of footprint at 25 s; its headless
shot (R3D_SMALL=1, so no depth) differs from the unchanged build by 0.279% of
pixels, the resized backdrop, and two runs of each are identical.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The atlas is white glyphs on alpha and the overlay reads only the coverage,
but it was uploaded RGBA8: Maroon Lake's 3072 x 2688 atlas was 43 MB of Metal
memory with its mips, the largest single thing in the launcher window.
overlay_font now keeps the alpha alone as R8 (ov_font_tex; any other PNG
shape is uploaded as before) and overlay.frag reads .r - the white fallback
texture reads 1 there as it did from .a.
Launcher window, graphics regions: the atlas 43.0 -> 11.5 MB; process
footprint at 25 s 402 -> 378 MB. The headless launcher shot is identical to
the unchanged build's (0 pixels differ; two unchanged runs differ from each
other by 1.69%, and the new one by the same against them).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With one frame in flight, a buffer filled every frame (the interface's vertex
buffer, ov_vbo) was still being read by the frame on the GPU when the next
frame filled it, so gvk_buf_reserve released it and made a new VkBuffer and
its memory every frame - and every new buffer took a prime submit of its own.
Each handle now keeps a second buffer (gvk_bsp_*): a busy one swaps with it
when it is free, else the busy one becomes the second and one new buffer is
made; releasing a handle lets both go.
The launcher window (--launcher-ui, R3D_VK_PROF, per 120 frames): buffers
made/destroyed 120/120 -> 0/0, command buffers 240 -> 120 (one a frame). Its
footprint does not move with it (358 MB at 40 s, both): the churn was work,
not memory.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Beside Os.heap_bytes: malloc_zone_pressure_relief(NULL, 0) on macOS (weak, so a libc without it reads
0) and HeapCompact(GetProcessHeap(), 0) on Windows - kernel32 only, so the Windows build imports
nothing new; the bytes it says it released. Once after a load, never per frame.
Measured, for the record: on macOS it does NOT reach the large-block cache. A C program that frees six
15 MB blocks still holds 90 MB of MALLOC_LARGE (empty) after relief on every zone (it returns 0); only
MallocLargeCache=0 in the environment AT PROCESS START turns the cache off (read at malloc's init -
set later, it does nothing). Maroon Lake's watcher sets it for the processes it spawns.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Set before r3d_open: the frame's uniform ring is 2 x 4 MB (was 2 x 64), memory blocks are 8 MB with
anything over 4 MB on its own (was 64 and 16), and the descriptor pools are an eighth of a world's.
Maroon Lake's launcher window draws a picture and text and never becomes the game; its launcher home
and Settings page draw pixel-identical on it. The full-size path is unchanged (steady: STEADY OK).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
rt_init opened every windowed program's window before main, and render3d's gvk_open then only
retitled it. win_open makes the window when there is none (cocoa.ll and win32.ll alike), so for a
program that has gvk_open the runtime now leaves it (window_later(), an intrinsic: windowed and
render3d present): a process that never reaches the renderer - Maroon Lake's launcher watcher, which
only spawns the game and waits - never makes a window, an NSApplication or AppKit's heap.
Audited every window native reachable before the renderer opens (settings, telemetry, rescue, the
watcher reach App.* and Input.*): on macOS each that loads W_win / W_app / W_view / W_mtl / W_glctx
checks it for null; win_close, win_running, win_text, win_held, win_cursor_mode, win_gl_scale and the
pad and touch reads load none. On Windows each that loads W_hwnd / W_hdc checks it; the rest load none.
Measured, windowed, R3D_DEV=1 R3D_PLAYTEST=2, both killed after:
- the game straight to play: the window, the Vulkan swapchain (1920x1080) and the valley's models
come up, alive at 30 s;
- the launcher (R3D_GAME=launcher): the watcher 11 MB -> 3.7 MB, its launcher window alive at 10 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Maroon Lake's launcher watcher - a process that only spawns the game and waits - held 58 MB, 48 MB of
it MALLOC_SMALL. Measured with malloc stack logging it was not the state defaults but L_grow: every
program sized every property type's per-entity store to MAX_ENT (1024) slots at start, 1,583 stores,
entities or none. And every state record was made with its defaults in L_init_globals before Boot.
- emit_lazy.ludic: a program's (not the runtime's) state global is left out of L_init_globals, and
every read of it calls @S_<global>(), which makes it on first call from its own initializer - after
every registry and plain global, so a default may read them (the init-order crash cannot come back
through a state). What a getter makes is declared (@lp_fdecl): a state first touched in play is made
once and not judged as a frame's keep. A function value's trampoline calls the getter too.
- L_grow starts the stores at ECS_FIRST (8) and doubles as entities come, as it always did past MAX_ENT.
The watcher (with the game's watch step moved before the systems' defs): 57 MB -> 11 MB; the only
state it makes is UiState (14 KB). What is left: AppKit's window, opened by rt_init before main for any
windowed program (~4 MB), and the runtime's font, image and 2D inits (~1.2 MB).
Goldens: the arena, fence, value and json goldens; the 36 ui examples; 30 of the 32 ECS test cases
(sprite_render and sprite_atlas time out under a plain runner with the toolchain before this too).
Package tests: ludic.base, save, settings, i18n.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
