feat(render3d): the meadow's blades are culled on the GPU - under 1.5 ms of a MoltenVK frame for all the grass

- grass_cull.comp decides each candidate blade once a frame (place, ground, density, water,
  slope, frustum, colour field) and writes survivors into three bands by distance, one indirect
  draw each; grass_inst.vert only bends and places the vertices. OpenGL keeps grass.vert's
  per-vertex path; R3D_GRASS_GPU=0 compares
- compute programs take sampled textures after their buffers (gpu_compute_tex / gpu_dispatch_tex),
  and every dispatch now records a compute-to-draw memory barrier
- the blades as a sward: 4 m cells, bands with five, three and one-quad blades, spacing doubling
  every 18 m to 70 m, never narrower than a pixel; lit facing the sun and leaning to the sky,
  shadow looked up above the ground (it read the terrain as its caster), a colour ramp that
  leaves only the sheath dark, clumps, dry patches and a tussock shade worked out per blade
- scatter layers flagged grass are skipped while the blades draw (and under R3D_NOGRASS);
  R3D_BLADES, R3D_NOBLADES win over the game's setting; R3D_GRASS_S0/D0 for measuring

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Orkun ÇAKILKAYA 2026-09-27 15:38:43 +03:00
parent f815f3e3cf
commit cc384efee0
34 changed files with 798 additions and 212 deletions

View file

@ -1,7 +1,7 @@
# ============================================================================
# grass.ludic — procedural GPU ground cover with continuous density (no rings).
#
# The world is cut into 16 m cells; blade j of a cell always stands in the same place
# The world is cut into 4 m cells; blade j of a cell always stands in the same place
# (shaders/grass.vert). Draws are per tile: the CPU walks tiles around the camera,
# frustum-culls them, and feeds each visible tile as many blade indices per cell as its
# NEAREST point could need; the vertex stage then keeps only the indices that exist at
@ -10,7 +10,7 @@
# draw count down — the cells and their hashes are the same in every tile size.
# ============================================================================
const GRASS_CELL: int = 16
const GRASS_CELL: int = 4
# A photographed blade, as an atlas of straightened blades side by side (the game sets
# this; the renderer does not name a game asset). 0 = the procedural gradient, which is
# what this was for a year: a two-tone ramp with a hard edge, and every blade in the
@ -89,11 +89,13 @@ function grass_blade_mesh(render3d_st: mut Render3dState, rows: int) -> Mesh {
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"
var defs = "#define FOLIAGE\n#define BLADE\n#define GBLADE\n"
if render3d_st.grass_merge {
defs = defs + "#define TILES\n"
render3d_st.grass_rec = words(GRASS_RECS * 5); render3d_st.grass_tv = floats(GRASS_RECS * 4); render3d_st.grass_chunk_tv = floats(GRASS_CHUNK * 4)
render3d_st.grass_band_start = words(4); render3d_st.grass_band_cells = words(4)
render3d_st.grass_band_mesh = new []Mesh
for b in 0 .. 4 { push(render3d_st.grass_band_mesh, null) }
render3d_st.grass_cmds = gpu_buffer_new(render3d_st)
}
render3d_st.grass_prog = r3d_program(render3d_st, "grass.vert", "model.frag", defs)
@ -101,6 +103,8 @@ function grass_init(render3d_st: mut Render3dState) -> void {
# five rows, four quads: the arch above needs somewhere to bend, and at four rows a
# blade that leans over is three straight segments and shows every join
render3d_st.grass_mesh = grass_blade_mesh(render3d_st, 5)
render3d_st.grass_mesh3 = grass_blade_mesh(render3d_st, 3)
render3d_st.grass_mesh2 = grass_blade_mesh(render3d_st, 2)
render3d_st.grass_wind = 2.4
# Matched to the blade's width: a 1 cm blade at 0.11 m spacing covers a third of what a
# 2.8 cm blade did, and the meadow goes bare. The game's graphics settings override this
@ -108,14 +112,17 @@ function grass_init(render3d_st: mut Render3dState) -> void {
# gets there, so the two have to agree or a shot shows something no player will see.
# That is exactly how the last change measured as "no effect": the render was identical
# because this line, not the settings, was deciding.
# The spacing doubles every 18 m and the blades stop at 70 m: at 45 m and 1600 m a 1 cm blade
# was under a pixel from 20 m on, costing a full blade's vertices to shimmer (5 ms of 9.7 on GL).
render3d_st.grass_s0 = 0.066
render3d_st.grass_d0 = 45.0
render3d_st.grass_radius = 1600.0
render3d_st.grass_d0 = 18.0
render3d_st.grass_radius = 70.0
if r3d_env_has(render3d_st, "R3D_NOBLADES") { render3d_st.grass_on = false }
if r3d_env_has(render3d_st, "R3D_GRASS_R") { render3d_st.grass_radius = float(Text.to_int(r3d_env(render3d_st, "R3D_GRASS_R"))) }
if r3d_env_has(render3d_st, "R3D_GRASS_S0") { render3d_st.grass_s0 = float(Text.to_int(r3d_env(render3d_st, "R3D_GRASS_S0"))) / 1000.0 }
if r3d_env_has(render3d_st, "R3D_GRASS_D0") { render3d_st.grass_d0 = float(Text.to_int(r3d_env(render3d_st, "R3D_GRASS_D0"))) }
if r3d_env_has(render3d_st, "R3D_GRASS_DBG") { render3d_st.grass_dbg = Text.to_int(r3d_env(render3d_st, "R3D_GRASS_DBG")) }
gg_init(render3d_st)
}
# indices per 16 m cell that could exist at distance d (the count the shader computes)
@ -126,11 +133,13 @@ function grass_count_at(render3d_st: Render3dState, d: float) -> int {
}
# one tile size over one distance band
function grass_tiles(render3d_st: mut Render3dState, size: int, d_min: float, d_max: float) -> void {
function grass_tiles(render3d_st: mut Render3dState, size: int, d_min: float, d_max: float, blade: Mesh) -> void {
let p = render3d_st.grass_prog
let cells = size / GRASS_CELL
if d_min >= render3d_st.grass_radius { return }
if render3d_st.grass_merge {
render3d_st.grass_band_start[render3d_st.grass_band_n] = render3d_st.grass_n; render3d_st.grass_band_cells[render3d_st.grass_band_n] = cells; render3d_st.grass_band_n += 1
render3d_st.grass_band_start[render3d_st.grass_band_n] = render3d_st.grass_n; render3d_st.grass_band_cells[render3d_st.grass_band_n] = cells
render3d_st.grass_band_mesh[render3d_st.grass_band_n] = blade; render3d_st.grass_band_n += 1
} else {
u_i(render3d_st, gpu_uniform(render3d_st, p, "u_tile_cells"), cells)
}
@ -161,7 +170,7 @@ function grass_tiles(render3d_st: mut Render3dState, size: int, d_min: float, d_
var inst = per * cells * cells
if inst > 65535 { inst = 65535 }
let r = render3d_st.grass_n * 5
render3d_st.grass_rec[r] = render3d_st.grass_mesh.count; render3d_st.grass_rec[r + 1] = inst; render3d_st.grass_rec[r + 2] = 0; render3d_st.grass_rec[r + 3] = 0
render3d_st.grass_rec[r] = blade.count; render3d_st.grass_rec[r + 1] = inst; render3d_st.grass_rec[r + 2] = 0; render3d_st.grass_rec[r + 3] = 0
render3d_st.grass_rec[r + 4] = ((render3d_st.grass_n - render3d_st.grass_band_start[render3d_st.grass_band_n - 1]) % GRASS_CHUNK) * 65536
let t = render3d_st.grass_n * 4
render3d_st.grass_tv[t] = ox; render3d_st.grass_tv[t + 1] = oz; render3d_st.grass_tv[t + 2] = float(per); render3d_st.grass_tv[t + 3] = 0.0
@ -170,7 +179,7 @@ function grass_tiles(render3d_st: mut Render3dState, size: int, d_min: float, d_
} else if per > 0 {
u_f2(render3d_st, gpu_uniform(render3d_st, p, "u_tile"), ox, oz)
u_i(render3d_st, gpu_uniform(render3d_st, p, "u_per_cell"), per)
mesh_draw_instanced(render3d_st, render3d_st.grass_mesh, per * cells * cells)
mesh_draw_instanced(render3d_st, blade, per * cells * cells)
render3d_st.grass_draws += 1
}
}
@ -180,11 +189,27 @@ function grass_tiles(render3d_st: mut Render3dState, size: int, d_min: float, d_
}
}
function grass_live(render3d_st: Render3dState) -> bool { return not render3d_st.grass_env_off and (render3d_st.grass_on or render3d_st.grass_force) and render3d_st.grass_prog != 0 }
function grass_draw(render3d_st: mut Render3dState) -> void {
if not (render3d_st.grass_on or render3d_st.grass_force) or render3d_st.ter_reflect or render3d_st.grass_prog == 0 { return }
if not grass_live(render3d_st) or render3d_st.ter_reflect { return }
if render3d_st.gg_on { gg_draw(render3d_st); return }
var p = render3d_st.grass_prog
if grass_mesh_live(render3d_st) { p = render3d_st.grass_mesh_prog }
gpu_use_program(render3d_st, p)
grass_bind(render3d_st, p)
gpu_cull(render3d_st, false)
render3d_st.grass_draws = 0
render3d_st.grass_n = 0
render3d_st.grass_band_n = 0
gpu_mesh_bind(render3d_st, render3d_st.grass_mesh)
grass_bands(render3d_st)
if render3d_st.grass_merge { grass_flush(render3d_st) }
gpu_cull(render3d_st, true)
}
# the blades' uniforms, shared by the per-vertex path and the culled one (grass_gpu.ludic)
function grass_bind(render3d_st: mut Render3dState, p: int) -> void {
u_mat4(render3d_st, gpu_uniform(render3d_st, p, "u_view"), render3d_st.cam_view)
u_mat4(render3d_st, gpu_uniform(render3d_st, p, "u_proj"), render3d_st.cam_proj)
u_mat4(render3d_st, gpu_uniform(render3d_st, p, "u_vp"), render3d_st.cam_vp_clean)
@ -207,6 +232,9 @@ function grass_draw(render3d_st: mut Render3dState) -> void {
u_f(render3d_st, gpu_uniform(render3d_st, p, "u_s0"), render3d_st.grass_s0)
u_f(render3d_st, gpu_uniform(render3d_st, p, "u_d0"), render3d_st.grass_d0)
u_f(render3d_st, gpu_uniform(render3d_st, p, "u_radius"), render3d_st.grass_radius)
var ph = render3d_st.post_h
if ph <= 0 { ph = gl_height() }
u_f(render3d_st, gpu_uniform(render3d_st, p, "u_px"), 2.0 * Math.tan(render3d_st.cam_fov * 0.5) / float(ph))
u_i(render3d_st, gpu_uniform(render3d_st, p, "u_dbg"), render3d_st.grass_dbg)
var orthotex = render3d_st.ter_ortho_tex
var oon = 1.0
@ -230,16 +258,16 @@ function grass_draw(render3d_st: mut Render3dState) -> void {
# does have a faint sheen, so this is small rather than nothing - the foliage layers have
# used 0.05 all along and never showed the fault.
u_f(render3d_st, gpu_uniform(render3d_st, p, "u_spec_scale"), 0.008)
gpu_cull(render3d_st, false)
render3d_st.grass_draws = 0
render3d_st.grass_n = 0
render3d_st.grass_band_n = 0
gpu_mesh_bind(render3d_st, render3d_st.grass_mesh)
grass_tiles(render3d_st, 16, 0.0, 300.0)
grass_tiles(render3d_st, 64, 300.0, 1200.0)
grass_tiles(render3d_st, 256, 1200.0, render3d_st.grass_radius)
if render3d_st.grass_merge { grass_flush(render3d_st) }
gpu_cull(render3d_st, true)
}
# the per-vertex path's tiles, in four bands
function grass_bands(render3d_st: mut Render3dState) -> void {
# four bands: the tile grows and the blade loses rows with distance (a band past the radius is
# skipped); a tile asks for what its NEAREST point needs, so a wide tile far out wastes the rest
grass_tiles(render3d_st, 4, 0.0, 6.0, render3d_st.grass_mesh)
grass_tiles(render3d_st, 4, 6.0, 16.0, render3d_st.grass_mesh3)
grass_tiles(render3d_st, 8, 16.0, 40.0, render3d_st.grass_mesh2)
grass_tiles(render3d_st, 16, 40.0, render3d_st.grass_radius, render3d_st.grass_mesh2)
}
# the records gathered this frame: uploaded once, then each band GRASS_CHUNK records a draw
@ -253,6 +281,8 @@ function grass_flush(render3d_st: mut Render3dState) -> void {
var e = render3d_st.grass_n
if b + 1 < render3d_st.grass_band_n { e = render3d_st.grass_band_start[b + 1] }
if e > s { u_i(render3d_st, gpu_uniform(render3d_st, p, "u_tile_cells"), render3d_st.grass_band_cells[b]) }
let blade = render3d_st.grass_band_mesh[b]
if e > s and not mesh { gpu_mesh_bind(render3d_st, blade) }
var k = s
while k < e {
var m = e - k
@ -275,7 +305,7 @@ function grass_flush(render3d_st: mut Render3dState) -> void {
}
} else {
u_f4v(render3d_st, gpu_uniform(render3d_st, p, "u_tiles"), m, render3d_st.grass_chunk_tv)
gpu_draw_mesh_indirect(render3d_st, render3d_st.grass_mesh, render3d_st.grass_cmds, k * 20, m, 0, 0)
gpu_draw_mesh_indirect(render3d_st, blade, render3d_st.grass_cmds, k * 20, m, 0, 0)
}
if not mesh { render3d_st.grass_draws += 1 }
k += m