render3d: start-up declared, ov_nine and the caster test on buffers made once - 362 frame sites to 122

The renderer starts itself from the first frame (gpu_select), so the device, its tables, the
manifest and programs, grass, shadows, sky, terrain textures and the actor pool read as frame
allocations: each is @alloc_ok as start-up, with gvk_fail (a failure) and the actor census and
texture dump (debug switches). ov_nine's four corner/uv arrays are one floats(16) made in
gvk_startup_state; ac_in_light's four points are made there too. Compiled (steady, and ludic deps
over main 4316ff97).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

packages/ludic.render3d/actor.ludic

@@ -76,6 +76,7 @@ function ac_prog_new(render3d_st: mut Render3dState, vs: string, fs: string, def
   u_i(render3d_st, gpu_uniform(render3d_st, p, "u_diff"), 0);  u_i(render3d_st, gpu_uniform(render3d_st, p, "u_nrm"), 1);  u_i(render3d_st, gpu_uniform(render3d_st, p, "u_arm"), 2)
   return a
 }
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function actor_init(render3d_st: mut Render3dState) -> void {
   # the records play takes, made now: an actor first placed in play (a fish, a sign, a bed) made its
   # record, matrix and tint then (actor_release gives them back to this pool)
@@ -266,6 +267,7 @@ function outline_frame(render3d_st: mut Render3dState) -> void { if render3d_st.
 
 # R3D_ACTOR_CENSUS=<frame>: at that frame, the lit pass's actors grouped by model - skinned or rigid,
 # how many share it, the primitives each draws - so instancing is aimed at what repeats
+@alloc_ok("a debug census and dumps, only under their switches")
 function actor_census(render3d_st: Render3dState) -> void {
   let keys = new []Model
   let cnt = new []int
@@ -364,7 +366,6 @@ function actor_draw_outline_one(render3d_st: mut Render3dState, a: Actor, ap: Ac
 # depth, only the draw. (Each actor used to draw into every cascade out to 300 m - in town that was
 # about 1500 skinned shadow draws a frame against 69 in the lit pass.)
 function ac_in_light(render3d_st: mut Render3dState, vp: floats, x: float, y: float, z: float, r: float) -> bool {
-  if render3d_st.ac_lp0 == null { render3d_st.ac_lp0 = floats(3);  render3d_st.ac_lpx = floats(3);  render3d_st.ac_lpy = floats(3);  render3d_st.ac_lpz = floats(3) }
   m4_xform_point(render3d_st.ac_lp0, vp, x, y, z)
   m4_xform_point(render3d_st.ac_lpx, vp, x + r, y, z)
   m4_xform_point(render3d_st.ac_lpy, vp, x, y + r, z)

packages/ludic.render3d/env.ludic

@@ -692,6 +692,7 @@ export state Render3dState {
   stream_debug_n: int = 0
   stream_scratch: floats = null
   gvk_pipe_bufs: words = null      # a pipeline's vertex buffers, while it is made
+  ov_nine_buf: floats = null       # ov_nine's corners and uvs
   gvk_ac: pointer = null           # the VkAllocationCallbacks every create and destroy is given (R3D_ALLOC_VK=1)
   gvk_prime_b: words = null        # buffers made since the frame began, read once so MoltenVK makes their
   gvk_prime_h: []long = null       # Metal buffers now (gvk_prime), and their handles

packages/ludic.render3d/gpu.ludic

@@ -33,6 +33,7 @@ function gpu_name(kind: int) -> string { return "vulkan" }
 
 # Start Vulkan. There is nothing to fall back to: a machine whose Vulkan cannot start is told why
 # (gpu_fallback_reason, printed) and r3d_init fails, so the program can say so and stop.
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function gpu_select(render3d_st: mut Render3dState) -> int {
   if render3d_st.gpu_kind != 0 { return render3d_st.gpu_kind }
   # a window needs a surface: Win32 on Windows, a CAMetalLayer through MoltenVK on macOS

packages/ludic.render3d/gpu_manifest.ludic

@@ -35,6 +35,7 @@ property GpuVariant {
 }
 
 
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function gpu_variant_new(id: string, vs: string, fs: string, defs: string) -> GpuVariant {
   let v = new GpuVariant
   v.id = id;  v.vs = vs;  v.fs = fs;  v.defs = defs
@@ -46,6 +47,7 @@ function gpu_variant_new(id: string, vs: string, fs: string, defs: string) -> Gp
 }
 
 # Read a manifest; returns how many programs it holds (0: none, or no file).
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function gpu_manifest_load(render3d_st: mut Render3dState, path: string) -> int {
   render3d_st.gpu_variants = new []GpuVariant
   let text = Fs.read_text(path)

packages/ludic.render3d/gpu_vk.ludic

@@ -8,6 +8,7 @@
 
 # ---- the device ---------------------------------------------------------------------------
 
+@alloc_ok("a failure, said once: the message is built only when something is already wrong")
 function gvk_fail(render3d_st: mut Render3dState, what: string, r: int) -> bool {
   render3d_st.gvk_why = `{what} failed (VkResult {r})`
   gvk_note(render3d_st, `r3d: vulkan: {render3d_st.gvk_why}`)
@@ -57,6 +58,7 @@ function gvk_no_command_pooling() -> void {
   Os.set_env("MVK_CONFIG_USE_COMMAND_POOLING", "0")
 }
 
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function gvk_init(render3d_st: mut Render3dState) -> bool {
   if render3d_st.gvk_ready { return true }
   gvk_no_command_pooling()
@@ -514,6 +516,7 @@ function gvk_mem_id(x: long) -> int { return int(x) }
 # ---- one-shot commands --------------------------------------------------------------------
 # Uploads, bakes and read-backs record into a command buffer, submit it and wait. The frame
 # itself does not go through here.
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function gvk_cmd_init(render3d_st: mut Render3dState) -> bool {
   let cpi = bytes(VkCommandPoolCreateInfo_sizeof)
   Vk.zero(cpi, VkCommandPoolCreateInfo_sizeof)
@@ -730,6 +733,7 @@ function gvk_labels_close(render3d_st: mut Render3dState) -> void {
 # Everything a frame's draws use as scratch or as a table, made once the device is up: each was made
 # on the first draw that needed it, which put an allocation in whichever frame that was (plan 25.2)
 const GVK_PROGS_MAX: int = 4096
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function gvk_startup_state(render3d_st: mut Render3dState) -> void {
   let zero: long = 0
   render3d_st.gvk_tmp_buf = bytes(GVK_TMP_BYTES)
@@ -765,4 +769,6 @@ function gvk_startup_state(render3d_st: mut Render3dState) -> void {
     for i in 0 .. 32 { render3d_st.gpu_unit_2d[i] = -1 }
   }
   if render3d_st.cam_planes == null { render3d_st.cam_planes = floats(16) }
+  if render3d_st.ov_nine_buf == null { render3d_st.ov_nine_buf = floats(16) }
+  if render3d_st.ac_lp0 == null { render3d_st.ac_lp0 = floats(3);  render3d_st.ac_lpx = floats(3);  render3d_st.ac_lpy = floats(3);  render3d_st.ac_lpz = floats(3) }
 }

packages/ludic.render3d/gpu_vk_draw.ludic

@@ -267,6 +267,7 @@ function gvk_dispatch_tex(render3d_st: mut Render3dState, c: int, params: pointe
 }
 
 # R3D_VK_PROBE=1: one dispatch over a GPU-owned buffer, read back - the compute path end to end
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function gvk_compute_probe(render3d_st: mut Render3dState) -> void {
   let c = gvk_compute_new(render3d_st, "probe", 1)
   if c == 0 { print("r3d: vulkan compute probe: FAILED (no program)");  return }
@@ -681,6 +682,7 @@ function gvk_bind_texture(render3d_st: mut Render3dState, p: int, name: string,
 # gvk_frame_reset.
 const GVK_RING_BYTES: int = 64 * 1024 * 1024
 
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function gvk_frame_init(render3d_st: mut Render3dState) -> bool {
   let props = bytes(VkPhysicalDeviceProperties_sizeof)
   Vk.get_physical_device_properties(render3d_st.gvk_pd, props)
@@ -754,6 +756,7 @@ function gvk_ring_put(render3d_st: mut Render3dState, blk: pointer, n: int) -> i
 # sampler, so a set is never matched against a texture it no longer describes.
 const GVK_KPOOL_SETS: int = 8192
 
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function gvk_kpool_make(render3d_st: mut Render3dState) -> bool {
   let sizes = bytes(VkDescriptorPoolSize_sizeof * 2)
   Vk.put_i32(sizes, VkDescriptorPoolSize_type, VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC)
@@ -775,6 +778,7 @@ function gvk_kpool_make(render3d_st: mut Render3dState) -> bool {
   return true
 }
 
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function gvk_sc_clear(render3d_st: mut Render3dState) -> void {
   # emptied, not replaced: five fresh lists each time the pool filled were never given back
   if render3d_st.gvk_sc_prog == null {
@@ -1413,6 +1417,7 @@ function gvk_screenshot(render3d_st: mut Render3dState, path: string) -> bool {
 # ---- what gpu.ludic's Vulkan branches call -----------------------------------------------------
 
 # the manifest the programs are looked up in, from the renderer's own shader directory
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function gvk_manifest(render3d_st: mut Render3dState) -> bool {
   r3d_find_root(render3d_st)
   render3d_st.gvk_spv_dir = `{render3d_st.r3d_root}/shaders/spv`

packages/ludic.render3d/grass.ludic

@@ -87,6 +87,7 @@ function grass_blade_mesh(render3d_st: mut Render3dState, rows: int) -> Mesh {
   return m
 }
 
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function grass_init(render3d_st: mut Render3dState) -> void {
   render3d_st.grass_merge = gpu_has_mdi(render3d_st) and not r3d_env_has(render3d_st, "R3D_GRASS_TILES")
   var defs = "#define FOLIAGE\n#define BLADE\n#define GBLADE\n"

packages/ludic.render3d/grass_gpu.ludic

@@ -26,6 +26,7 @@ function gg_base(b: int) -> int {
 }
 
 # once, after grass_init: the programs, the buffers and the three instanced blades
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function gg_init(render3d_st: mut Render3dState) -> void {
   if not gpu_has_compute(render3d_st) or render3d_st.grass_prog == 0 { return }
   if r3d_env_has(render3d_st, "R3D_GRASS_GPU") and r3d_env(render3d_st, "R3D_GRASS_GPU") == "0" { return }

packages/ludic.render3d/overlay.ludic

@@ -363,15 +363,14 @@ function ov_sub(render3d_st: mut Render3dState, tex: int, x: int, y: int, w: int
 function ov_nine(render3d_st: mut Render3dState, tex: int, tw: int, src: int, x: int, y: int, w: int, h: int, dst: int, r: float, g: float, b: float, a: float) -> void {
   ov_use_tex(render3d_st, tex)
   let s = float(src) / float(tw)
-  let xs = floats(4);  let ys = floats(4);  let us = floats(4);  let vs = floats(4)
-  xs[0] = float(x);  xs[1] = float(x + dst);  xs[2] = float(x + w - dst);  xs[3] = float(x + w)
-  ys[0] = float(y);  ys[1] = float(y + dst);  ys[2] = float(y + h - dst);  ys[3] = float(y + h)
-  us[0] = 0.0;  us[1] = s;  us[2] = 1.0 - s;  us[3] = 1.0
-  vs[0] = 0.0;  vs[1] = s;  vs[2] = 1.0 - s;  vs[3] = 1.0
+  let nb = render3d_st.ov_nine_buf                  # made once (gvk_startup_state): xs, ys, us, vs
+  nb[0 + 0] = float(x);  nb[0 + 1] = float(x + dst);  nb[0 + 2] = float(x + w - dst);  nb[0 + 3] = float(x + w)
+  nb[4 + 0] = float(y);  nb[4 + 1] = float(y + dst);  nb[4 + 2] = float(y + h - dst);  nb[4 + 3] = float(y + h)
+  nb[8 + 0] = 0.0;  nb[8 + 1] = s;  nb[8 + 2] = 1.0 - s;  nb[8 + 3] = 1.0
+  nb[12 + 0] = 0.0;  nb[12 + 1] = s;  nb[12 + 2] = 1.0 - s;  nb[12 + 3] = 1.0
   for j in 0 .. 3 {
-    for i in 0 .. 3 { ov_quad(render3d_st, xs[i], ys[j], xs[i + 1], ys[j + 1], us[i], vs[j], us[i + 1], vs[j + 1], r, g, b, a) }
+    for i in 0 .. 3 { ov_quad(render3d_st, nb[0 + i], nb[4 + j], nb[0 + i + 1], nb[4 + j + 1], nb[8 + i], nb[12 + j], nb[8 + i + 1], nb[12 + j + 1], r, g, b, a) }
   }
-  free(xs);  free(ys);  free(us);  free(vs)
 }
 # an arbitrary quad (float-bit pixel corners, clockwise from top-left) of a texture
 function ov_quad4(render3d_st: mut Render3dState, x0: float, y0: float, x1: float, y1: float, x2: float, y2: float, x3: float, y3: float, u0: float, v0: float, u1: float, v1: float, r: float, g: float, b: float, a: float) -> void {

packages/ludic.render3d/prof.ludic

@@ -15,6 +15,7 @@ const PROF_SLOTS: int = 16
 const PROF_RING:  int = 4            # frames of latency before a result is read
 
 
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function prof_init(render3d_st: mut Render3dState) -> void {
   render3d_st.prof_on = r3d_env_has(render3d_st, "R3D_PROF")
   if not render3d_st.prof_on { return }

packages/ludic.render3d/programs.ludic

@@ -24,6 +24,7 @@ function r3d_home() -> string {
 
 # Settle r3d_root on first use: the package as checked out beside the project, else the
 # copy that ships with the toolchain.
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function r3d_find_root(render3d_st: mut Render3dState) -> void {
   if render3d_st.r3d_root_found { return }
   render3d_st.r3d_root_found = true

packages/ludic.render3d/scatter.ludic

@@ -505,6 +505,7 @@ function impostor_bake(render3d_st: mut Render3dState, model: Model, tiles: int,
 
 # Give a layer a LOD chain. `dists` (float bits) are the outer distances of each level;
 # the last one becomes the layer's `near` so the impostor (if any) starts there.
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function layer_set_lods(render3d_st: mut Render3dState, l: Layer, models: []Model, dists: floats) -> void {
   l.lods = models
   l.n_lods = len(models)

packages/ludic.render3d/shadow.ludic

@@ -8,6 +8,7 @@
 const SHADOW_CASCADES: int = 5
 
 
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function shadow_init(render3d_st: mut Render3dState) -> void {
   shadow_make_tex(render3d_st)
   render3d_st.sh_fbo = gpu_fb_new(render3d_st)

packages/ludic.render3d/sky.ludic

@@ -37,6 +37,7 @@ function sky_dir_from_uv(o: floats, u: float, v: float) -> void {
   v3_set(o, st * Math.sin(phi), Math.cos(theta), -(st * Math.cos(phi)))
 }
 
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function sky_load(render3d_st: mut Render3dState, path: string) -> bool {
   render3d_st.sky_tex = tex_load_hdr(render3d_st, path)
   if render3d_st.sky_tex == 0 { return false }

packages/ludic.render3d/streamline.ludic

@@ -66,6 +66,7 @@ function gsl_boot(render3d_st: mut Render3dState) -> void {
 }
 
 # After Vk.open(), before vkCreateInstance.
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function gsl_init(render3d_st: mut Render3dState) -> void {
   if render3d_st.gsl_on or Vk.sl_active() == 0 { return }
   render3d_st.gsl_feats = gsl_struct(20)

packages/ludic.render3d/terrain.ludic

@@ -325,6 +325,7 @@ function terrain_height_smooth(render3d_st: mut Render3dState, x: float, z: floa
   return h
 }
 
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function terrain_load_textures(render3d_st: mut Render3dState) -> void {
   render3d_st.ter_tex = words(15)
   let a = render3d_st.r3d_assets + "/textures/"
@@ -361,6 +362,7 @@ function terrain_init_step(render3d_st: mut Render3dState, i: int) -> void {
   terrain_init_finish(render3d_st)
 }
 
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function terrain_init_finish(render3d_st: mut Render3dState) -> void {
   cdlod_init(render3d_st)
   render3d_st.ter_wire = r3d_env_has(render3d_st, "R3D_WIRE")
@@ -591,6 +593,7 @@ function terrain_draw(render3d_st: mut Render3dState) -> void {
 # One 32x32 patch mesh (a_xz in 0..1) drawn once per selected quadtree node; the vertex
 # shader places, scales and morphs it. Levels are drawn out to cd_range[k] = 48 * 2^k m,
 # so cells are 1 m within 48 m, 2 m to 96 m, 4 m to 192 m ... 256 m at the root.
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function cdlod_init(render3d_st: mut Render3dState) -> void {
   let m = gpu_mesh_new(render3d_st)
   let n = CD_G + 1
@@ -624,6 +627,7 @@ function cdlod_init(render3d_st: mut Render3dState) -> void {
 }
 
 # min/max height per patch at every level, from the CPU copy of the height field
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function cdlod_bounds(render3d_st: mut Render3dState) -> void {
   render3d_st.cd_min = new []floats;  render3d_st.cd_max = new []floats
   let t = TERRAIN_RES / CD_LEAVES              # texels per leaf patch side

packages/ludic.render3d/texture.ludic

@@ -321,6 +321,7 @@ function tex_solid(render3d_st: mut Render3dState, r: int, g: int, b: int, a: in
 
 # ---- Radiance .hdr (RGBE, new-style RLE) -> RGB float bits -------------------------
 
+@alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play")
 function hdr_decode(render3d_st: mut Render3dState, path: pointer) -> floats {
   let d = r3d_read_file(render3d_st, path)
   if d == null { print(`hdr: cannot read {path}`);  return null }
@@ -437,6 +438,7 @@ function tex_target(render3d_st: mut Render3dState, w: int, h: int, ifmt: int, f
 }
 
 # Debug: the brightest texel of an RGBA float texture and where it is.
+@alloc_ok("a debug census and dumps, only under their switches")
 function tex_max(render3d_st: mut Render3dState, tex: int, w: int, h: int, tag: pointer) -> void {
   let buf = floats(w * h * 4)
   gpu_tex_bind(render3d_st, GPU_TEX2D, tex)