# ============================================================================
# shadow.ludic — cascaded shadow maps for the sun: four 2048^2 depth layers,
# each an orthographic light frustum fitted to the bounding sphere of a slice
# of the camera frustum and snapped to its own texel grid (no swimming).
# ============================================================================

# the size of each cascade's depth layer; shadow_set_res changes it at run time
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)
  gpu_fb_bind(render3d_st, render3d_st.sh_fbo)
  gpu_fb_no_color(render3d_st)
  gpu_fb_bind(render3d_st, 0)
  render3d_st.sh_vp = floats(16 * SHADOW_CASCADES)
  render3d_st.sh_vp_v = m4_views(render3d_st.sh_vp, SHADOW_CASCADES)
  render3d_st.sh_split = floats(SHADOW_CASCADES)
  render3d_st.sh_range = floats(SHADOW_CASCADES)
  render3d_st.sh_texel = floats(SHADOW_CASCADES)
  # the fourth slice keeps a tree-sized texel out to a kilometre; only the massif uses the last
  render3d_st.sh_split[0] = 16.0;  render3d_st.sh_split[1] = 60.0;  render3d_st.sh_split[2] = 250.0;  render3d_st.sh_split[3] = 1100.0;  render3d_st.sh_split[4] = 6000.0
  render3d_st.sh_tmp_proj = m4_new();  render3d_st.sh_tmp_vp = m4_new();  render3d_st.sh_tmp_inv = m4_new();  render3d_st.sh_tmp_view = m4_new()
  render3d_st.sh_corner = floats(3)
}

# A shadow resolution setting: 1024, 2048 or 4096 per cascade. The depth layers are made again at
# the new size; the pass attaches a layer per cascade every frame, and the lighting reads the
# texel size from the map itself, so nothing else has to follow.
function shadow_set_res(render3d_st: mut Render3dState, r: int) -> void {
  if r < 256 or r == render3d_st.shadow_res { return }
  let was = render3d_st.shadow_res
  render3d_st.shadow_res = r
  if render3d_st.sh_tex == 0 { return }
  gpu_tex_free(render3d_st, render3d_st.sh_tex)
  shadow_make_tex(render3d_st)
  # not enough video memory for that size: go back to the one that worked, and smaller again if even
  # that is refused now, rather than ending with no shadow map at all
  if not gpu_tex_ok(render3d_st, render3d_st.sh_tex) {
    render3d_st.shadow_refused = r
    print(`r3d: shadows: no memory for {r} x {r} cascades; keeping {was}`)
    var size = was
    render3d_st.shadow_res = size
    gpu_tex_free(render3d_st, render3d_st.sh_tex)
    shadow_make_tex(render3d_st)
    while not gpu_tex_ok(render3d_st, render3d_st.sh_tex) and size > 512 {
      size = size / 2
      render3d_st.shadow_res = size
      gpu_tex_free(render3d_st, render3d_st.sh_tex)
      shadow_make_tex(render3d_st)
    }
  }
}

function shadow_make_tex(render3d_st: mut Render3dState) -> void {
  render3d_st.sh_tex = gpu_tex_new(render3d_st)
  gpu_tex_bind(render3d_st, GPU_TEX2D_ARRAY, render3d_st.sh_tex)
  gpu_tex_image3d(render3d_st, GL_DEPTH_COMPONENT32F, render3d_st.shadow_res, render3d_st.shadow_res, SHADOW_CASCADES, GL_DEPTH_COMPONENT, GL_FLOAT, null)
  gpu_tex_param(render3d_st, GPU_TEX2D_ARRAY, GL_TEXTURE_MIN_FILTER, GL_LINEAR)
  gpu_tex_param(render3d_st, GPU_TEX2D_ARRAY, GL_TEXTURE_MAG_FILTER, GL_LINEAR)
  gpu_tex_param(render3d_st, GPU_TEX2D_ARRAY, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_BORDER)
  gpu_tex_param(render3d_st, GPU_TEX2D_ARRAY, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_BORDER)
  gpu_tex_param(render3d_st, GPU_TEX2D_ARRAY, GL_TEXTURE_COMPARE_MODE, GL_COMPARE_REF_TO_TEXTURE)
  gpu_tex_param(render3d_st, GPU_TEX2D_ARRAY, GL_TEXTURE_COMPARE_FUNC, GL_LEQUAL)
  let border = gl_floats(4)
  gl_put(border, 0, 1.0);  gl_put(border, 1, 1.0);  gl_put(border, 2, 1.0);  gl_put(border, 3, 1.0)
  gpu_tex_border(render3d_st, GPU_TEX2D_ARRAY, border)
  free(border)
}

# light view-projection for the camera-frustum slice [near, far]
function shadow_fit(render3d_st: mut Render3dState, c: int, near: float, far: float) -> void {
  # Fit the slice in VIEW space, not world space. The bounding sphere of a frustum
  # slice depends only on near/far/fov/aspect — never on where the camera is pointing —
  # so computing it here makes the radius a constant per cascade. Doing it in world
  # space (as this did) let the radius wobble as the camera turned, which changed the
  # texel size, which moved the grid the projection is snapped to, so the whole shadow
  # map resampled every frame: that is the crawl and flicker seen while moving.
  m4_perspective(render3d_st.sh_tmp_proj, render3d_st.cam_fov, render3d_st.cam_aspect, near, far)
  m4_inverse(render3d_st.sh_tmp_inv, render3d_st.sh_tmp_proj)          # NDC -> view space
  let cview = v3_new(0.0, 0.0, 0.0)
  let corners = floats(24)
  for i in 0 .. 8 {
    var x = -1.0;  var y = -1.0;  var z = -1.0
    if (i & 1) != 0 { x = 1.0 }
    if (i & 2) != 0 { y = 1.0 }
    if (i & 4) != 0 { z = 1.0 }
    let w = m4_xform_point(render3d_st.sh_corner, render3d_st.sh_tmp_inv, x, y, z)
    let iw = 1.0 / w
    corners[i * 3] = render3d_st.sh_corner[0] * iw;  corners[i * 3 + 1] = render3d_st.sh_corner[1] * iw;  corners[i * 3 + 2] = render3d_st.sh_corner[2] * iw
    cview[0] = cview[0] + corners[i * 3];  cview[1] = cview[1] + corners[i * 3 + 1];  cview[2] = cview[2] + corners[i * 3 + 2]
  }
  v3_scale(cview, cview, 1.0 / 8.0)
  var radius = 0.0
  for i in 0 .. 8 {
    v3_set(render3d_st.sh_corner, corners[i * 3], corners[i * 3 + 1], corners[i * 3 + 2])
    let d = v3_dist(render3d_st.sh_corner, cview)
    if d > radius { radius = d }
  }
  radius = radius * 1.05
  # the slice centre back into world space
  m4_inverse(render3d_st.sh_tmp_vp, render3d_st.cam_view)
  let center = floats(3)
  m4_xform_point(center, render3d_st.sh_tmp_vp, cview[0], cview[1], cview[2])
  free(cview)
  # light view: from far along the sun direction, looking at the centre
  let eye = floats(3)
  # casters up to ~900 m toward the sun (a mountain across the valley), and the
  # slice itself behind the centre: a tight depth range keeps the bias small
  let back = radius + 900.0
  v3_madd(eye, center, render3d_st.sun_dir, back)
  let up = v3_new(0.0, 1.0, 0.0)
  m4_look_at(render3d_st.sh_tmp_view, eye, center, up)
  # snap the ortho window to the shadow texel grid
  let texel = radius * 2.0 / float(render3d_st.shadow_res)
  m4_xform_point(render3d_st.sh_corner, render3d_st.sh_tmp_view, center[0], center[1], center[2])
  let ox = Math.floor(render3d_st.sh_corner[0] / texel) * texel - render3d_st.sh_corner[0]
  let oy = Math.floor(render3d_st.sh_corner[1] / texel) * texel - render3d_st.sh_corner[1]
  let nr = -radius
  let zfar = back + radius + 100.0
  m4_ortho(render3d_st.sh_tmp_proj, nr + ox, radius + ox, nr + oy, radius + oy, 1.0, zfar)
  render3d_st.sh_range[c] = zfar - 1.0
  render3d_st.sh_texel[c] = texel
  let out = floats(16)
  m4_mul(out, render3d_st.sh_tmp_proj, render3d_st.sh_tmp_view)
  for i in 0 .. 16 { render3d_st.sh_vp[c * 16 + i] = out[i] }
  free(out);  free(eye);  free(up);  free(center);  free(corners)
}

function shadow_cascade_vp(render3d_st: Render3dState, c: int) -> floats { return render3d_st.sh_vp_v[c] }


# render every cascade; `draw` happens through terrain_draw_shadow + the scene's casters
function shadow_pass(render3d_st: mut Render3dState) -> void {
  var near = render3d_st.cam_near
  gpu_fb_bind(render3d_st, render3d_st.sh_fbo)
  gpu_viewport(render3d_st, 0, 0, render3d_st.shadow_res, render3d_st.shadow_res)
  gpu_depth_test(render3d_st, true)
  gpu_depth_func(render3d_st, GL_LESS)
  gpu_depth_bias(render3d_st, 2.0, 4.0)
  gpu_cull(render3d_st, false)
  for c in 0 .. SHADOW_CASCADES {
    render3d_st.sh_cascade = c
    shadow_fit(render3d_st, c, near, render3d_st.sh_split[c])
    prof_cpu_mark(render3d_st, "shadow fit")
    near = render3d_st.sh_split[c]
    gpu_fb_depth_layer(render3d_st, render3d_st.sh_tex, c)
    if render3d_st.r3d_debug and c == 0 { let st = gpu_fb_status(render3d_st);  print(`shadow fbo status {st}`) }
    gpu_clear(render3d_st, GL_DEPTH_BUFFER_BIT)
    let vp = shadow_cascade_vp(render3d_st, c)
    # shadows off (a video setting): the cascades stay cleared, so everything reads lit
    if render3d_st.sh_enabled {
      if not render3d_st.sh_skip_terrain { terrain_draw_shadow(vp) }
      r3d_scene_casters(render3d_st, vp)
    }
  }
  gpu_depth_bias(render3d_st, 0.0, 0.0)
  gpu_fb_bind(render3d_st, 0)
  if render3d_st.r3d_debug_shadow { shadow_dump(render3d_st) }
  if render3d_st.r3d_debug_shadow and not render3d_st.sh_printed2 {
    render3d_st.sh_printed2 = true
    let q = floats(3)
    if render3d_st.sh_probe_x != 0.0 {
      let vp = shadow_cascade_vp(render3d_st, 2)
      m4_xform_point(q, vp, render3d_st.sh_probe_x, render3d_st.sh_probe_y, render3d_st.sh_probe_z)
      print(`probe base ndc {fixed(q[0])} {fixed(q[1])} {fixed(q[2])}`)
      m4_xform_point(q, vp, render3d_st.sh_probe_x, render3d_st.sh_probe_y + 15.0, render3d_st.sh_probe_z)
      print(`probe top  ndc {fixed(q[0])} {fixed(q[1])} {fixed(q[2])}  -> map texel {int((q[0] * 0.5 + 0.5) * float(render3d_st.shadow_res))} {int((q[1] * 0.5 + 0.5) * float(render3d_st.shadow_res))}`)
      # where the top's shadow lands on the ground: walk down the sun ray
      let gx = render3d_st.sh_probe_x + 0.0 - render3d_st.sun_dir[0] * (15.0 / render3d_st.sun_dir[1])
      let gz = render3d_st.sh_probe_z - render3d_st.sun_dir[2] * (15.0 / render3d_st.sun_dir[1])
      m4_xform_point(q, vp, gx, terrain_height(render3d_st, gx, gz), gz)
      print(`shadow-of-top ground ndc {fixed(q[0])} {fixed(q[1])} {fixed(q[2])} at {fixed(gx)} {fixed(gz)}`)
    }
    for c in 0 .. SHADOW_CASCADES {
      let vp = shadow_cascade_vp(render3d_st, c)
      # a point 5 m ahead of the camera on the ground
      let px = render3d_st.cam_pos[0] + render3d_st.cam_fwd[0] * 5.0;  let pz = render3d_st.cam_pos[2] + render3d_st.cam_fwd[2] * 5.0
      let w = m4_xform_point(q, vp, px, terrain_height(render3d_st, px, pz), pz)
      print(`cascade {c}: ndc {fixed(q[0])} {fixed(q[1])} {fixed(q[2])} w {fixed(w)}  m0 {fixed(vp[0])} m5 {fixed(vp[5])} m14 {fixed(vp[14])}`)
    }
    free(q)
  }
}

# Debug: cascade depths as grey PPMs (build/dbg_shadow_<c>.ppm)
@alloc_ok("asked for by the player or a tool, not by the frame")
function shadow_dump(render3d_st: mut Render3dState) -> void {
  let n = render3d_st.shadow_res * render3d_st.shadow_res
  let buf = floats(n * SHADOW_CASCADES)
  gpu_tex_bind(render3d_st, GPU_TEX2D_ARRAY, render3d_st.sh_tex)
  gpu_tex_param(render3d_st, GPU_TEX2D_ARRAY, GL_TEXTURE_COMPARE_MODE, GL_NONE)
  gpu_tex_read(render3d_st, GPU_TEX2D_ARRAY, GL_DEPTH_COMPONENT, GL_FLOAT, data_of(buf))
  gpu_tex_param(render3d_st, GPU_TEX2D_ARRAY, GL_TEXTURE_COMPARE_MODE, GL_COMPARE_REF_TO_TEXTURE)
  if render3d_st.sh_probe_x != 0.0 {
    let vp = shadow_cascade_vp(render3d_st, 2)
    let q = floats(3)
    m4_xform_point(q, vp, render3d_st.sh_probe_x, render3d_st.sh_probe_y + 12.0, render3d_st.sh_probe_z)
    let tx = int((q[0] * 0.5 + 0.5) * float(render3d_st.shadow_res))
    let ty = int((q[1] * 0.5 + 0.5) * float(render3d_st.shadow_res))
    let want = q[2] * 0.5 + 0.5
    print(`probe (12 m up) texel {tx} {ty} card depth {fixed(want * 1000.0)}/1000`)
    for dy in 0 .. 5 {
      let yy = ty - 40 + dy * 20
      print(`  row {yy}: {fixed(buf[2 * n + yy * render3d_st.shadow_res + tx - 20] * 1000.0)} {fixed(buf[2 * n + yy * render3d_st.shadow_res + tx] * 1000.0)} {fixed(buf[2 * n + yy * render3d_st.shadow_res + tx + 20] * 1000.0)} /1000`)
    }
    free(q)
  }
  let sm = 512
  let row = bytes(sm * 3)
  for c in 0 .. SHADOW_CASCADES {
    # stretch between the map's own min and max (ignoring the far plane)
    var lo = 1.0;  var hi = 0.0
    var i = 0
    while i < n { let d = buf[c * n + i];  if d < 0.999 { if d < lo { lo = d };  if d > hi { hi = d } };  i += 97 }
    print(`cascade {c} depth range {fixed(lo)} .. {fixed(hi)}`)
    let f = file_open(`build/dbg_shadow_{c}.ppm`, "wb")
    let hdr = `P6\n{sm} {sm}\n255\n`
    file_write(f, hdr, len(hdr))
    let st = render3d_st.shadow_res / sm
    for y in 0 .. sm {
      for x in 0 .. sm {
        let d = buf[c * n + (y * st) * render3d_st.shadow_res + x * st]
        let g = int(Math.clamp((d - lo) / Math.max(hi - lo, 0.0001), 0.0, 1.0) * 255.0)
        row[x * 3] = g;  row[x * 3 + 1] = g;  row[x * 3 + 2] = g
      }
      file_write(f, row, sm * 3)
    }
    file_close(f)
  }
  free(buf);  free(row)
}

# a uniform array's location: some drivers only answer to the "[0]" spelling
# an array uniform by its first element's name, else its bare one; both names are literals, because
# building `name + "[0]"` here made a string per program per pass per frame that was never freed
function sh_loc(render3d_st: Render3dState, prog: int, first: string, name: string) -> int {
  var loc = gpu_uniform(render3d_st, prog, first)
  if loc < 0 { loc = gpu_uniform(render3d_st, prog, name) }
  return loc
}
function shadow_bind(render3d_st: mut Render3dState, prog: int) -> void {
  r3d_bind_tex(render3d_st, prog, "u_shadow", 15, GPU_TEX2D_ARRAY, render3d_st.sh_tex)
  # the height-field shadow (terrain.ludic); a stand-in texture keeps the unit valid before the bake
  var ts = render3d_st.ter_shadow_tex
  var ts_on = 1.0
  if ts == 0 { ts = render3d_st.ter_height_tex;  ts_on = 0.0 }
  r3d_bind_2d(render3d_st, prog, "u_tershadow", 6, ts)
  terrain_bind_height(render3d_st, prog)
  u_f(render3d_st, gpu_uniform(render3d_st, prog, "u_ts_on"), ts_on)
  u_f(render3d_st, gpu_uniform(render3d_st, prog, "u_ts_half"), float(render3d_st.TERRAIN_HALF))
  u_f2(render3d_st, gpu_uniform(render3d_st, prog, "u_ts_origin"), render3d_st.ter_ox, render3d_st.ter_oz)
  var loc = gpu_uniform(render3d_st, prog, "u_cascade_vp[0]")
  if loc < 0 { loc = gpu_uniform(render3d_st, prog, "u_cascade_vp") }
  if render3d_st.r3d_debug_shadow and not render3d_st.sh_printed { render3d_st.sh_printed = true;  print(`cascade vp loc {loc} / {gpu_uniform(render3d_st, prog, "u_cascade_vp")}  split loc {gpu_uniform(render3d_st, prog, "u_cascade_split")} shadow loc {gpu_uniform(render3d_st, prog, "u_shadow")}`) }
  u_mat4n(render3d_st, loc, SHADOW_CASCADES, render3d_st.sh_vp)
  u_fv(render3d_st, sh_loc(render3d_st, prog, "u_cascade_split[0]", "u_cascade_split"), SHADOW_CASCADES, render3d_st.sh_split)
  if render3d_st.r3d_debug_shadow and not render3d_st.sh_printed3 { render3d_st.sh_printed3 = true;  print(`range {fixed(render3d_st.sh_range[0])} {fixed(render3d_st.sh_range[1])} {fixed(render3d_st.sh_range[2])} {fixed(render3d_st.sh_range[3])} texel*1000 {fixed(render3d_st.sh_texel[0] * 1000.0)} {fixed(render3d_st.sh_texel[1] * 1000.0)} {fixed(render3d_st.sh_texel[2] * 1000.0)} {fixed(render3d_st.sh_texel[3] * 1000.0)} locs {gpu_uniform(render3d_st, prog, "u_cascade_range")} {gpu_uniform(render3d_st, prog, "u_cascade_texel")}`) }
  u_fv(render3d_st, sh_loc(render3d_st, prog, "u_cascade_range[0]", "u_cascade_range"), SHADOW_CASCADES, render3d_st.sh_range)
  if r3d_env_has(render3d_st, "R3D_FORCE") { render3d_st.sh_force = Text.to_int(r3d_env(render3d_st, "R3D_FORCE")) }
  u_i(render3d_st, gpu_uniform(render3d_st, prog, "u_force_cascade"), render3d_st.sh_force)
  u_fv(render3d_st, sh_loc(render3d_st, prog, "u_cascade_texel[0]", "u_cascade_texel"), SHADOW_CASCADES, render3d_st.sh_texel)
}