render3d(vulkan): the ground reads the sun cascades itself - no separate terrain sun pass

tersun.frag rasterised every terrain patch a second time into a screen buffer, because on OpenGL the
cascade read inside terrain.frag fell off a driver cliff (about 6 ms a frame). Vulkan has no such
cliff: its terrain programs are the SUN_INLINE variant, which evaluates the same two tiers with the
same normal and cross-fade in the ground's own shader, and terrain_draw skips the pass - in the frame
and in the water reflection. OpenGL keeps the pass. R3D_SUN_PASS=1 keeps it on Vulkan, for comparing.

Mac Vulkan town 1759 -> 1692 draws; frames within 2/255 of the pass (15331 px, float rounding). PC camp
1984 -> 1882 draws, 3.9 s for 400 frames either way, self-tests 59/59, validation 0. OpenGL frames
byte-identical.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

packages/ludic.render3d/shaders/spv/manifest.txt

@@ -2324,6 +2324,210 @@ T f4382918 u_snow_d 15
 T f4382918 u_sunshadow 16
 T f4382918 u_tershadow 17
 T f4382918 u_ts_height 18
+P f2952361 terrain.vert terrain.frag #define SUN_INLINE;
+B f2952361 vert 0 204
+U f2952361 vert u_half 0 float 1 0
+U f2952361 vert u_view 16 mat4 1 0
+U f2952361 vert u_proj 80 mat4 1 0
+U f2952361 vert u_origin 144 vec2 1 0
+U f2952361 vert u_cam_pos 160 vec3 1 0
+U f2952361 vert u_node 176 vec3 1 0
+U f2952361 vert u_morph 192 vec2 1 0
+U f2952361 vert u_grid 200 float 1 0
+I f2952361 a_xz 0 vec2
+B f2952361 frag 1 872
+U f2952361 frag u_cascade_vp 0 mat4 5 64
+U f2952361 frag u_cascade_split 320 float 5 16
+U f2952361 frag u_cascade_range 400 float 5 16
+U f2952361 frag u_cascade_texel 480 float 5 16
+U f2952361 frag u_sun_dir 560 vec3 1 0
+U f2952361 frag u_sun_color 576 vec3 1 0
+U f2952361 frag u_cam_pos 592 vec3 1 0
+U f2952361 frag u_prefilter_levels 604 float 1 0
+U f2952361 frag u_fog_density 608 float 1 0
+U f2952361 frag u_fog_height_falloff 612 float 1 0
+U f2952361 frag u_fog_base 616 float 1 0
+U f2952361 frag u_clip_y 620 float 1 0
+U f2952361 frag u_spec_scale 624 float 1 0
+U f2952361 frag u_sky_rot 632 vec2 1 0
+U f2952361 frag u_ibl_scale 640 vec3 1 0
+U f2952361 frag u_daylight 652 float 1 0
+U f2952361 frag u_fire_pos 656 vec3 1 0
+U f2952361 frag u_fire_color 672 vec3 1 0
+U f2952361 frag u_hand_pos 688 vec3 1 0
+U f2952361 frag u_hand_color 704 vec3 1 0
+U f2952361 frag u_hand_dir 720 vec3 1 0
+U f2952361 frag u_hand_cone 732 float 1 0
+U f2952361 frag u_ts_origin 736 vec2 1 0
+U f2952361 frag u_ts_half 744 float 1 0
+U f2952361 frag u_ts_on 748 float 1 0
+U f2952361 frag u_force_cascade 752 int 1 0
+U f2952361 frag u_cloud_shadow 756 float 1 0
+U f2952361 frag u_time 760 float 1 0
+U f2952361 frag u_half 764 float 1 0
+U f2952361 frag u_texel 768 float 1 0
+U f2952361 frag u_ortho_on 772 float 1 0
+U f2952361 frag u_carpet_on 776 float 1 0
+U f2952361 frag u_snow_line 780 float 1 0
+U f2952361 frag u_lake_level 784 float 1 0
+U f2952361 frag u_sea_level 788 float 1 0
+U f2952361 frag u_origin 792 vec2 1 0
+U f2952361 frag u_view 800 mat4 1 0
+U f2952361 frag u_far_split 864 float 1 0
+U f2952361 frag u_far_band 868 float 1 0
+T f2952361 u_brdf 2
+T f2952361 u_carpet 3
+T f2952361 u_grass_a 4
+T f2952361 u_grass_d 5
+T f2952361 u_grass_n 6
+T f2952361 u_height 7
+T f2952361 u_irradiance 8
+T f2952361 u_ortho 9
+T f2952361 u_prefilter 10
+T f2952361 u_rock_a 11
+T f2952361 u_rock_d 12
+T f2952361 u_rock_n 13
+T f2952361 u_shadow 14
+T f2952361 u_snow_d 15
+T f2952361 u_sunshadow 16
+T f2952361 u_tershadow 17
+T f2952361 u_ts_height 18
+P a0f2dbac terrain.vert terrain.frag #define SUN_INLINE;#define FAR_ONLY;
+B a0f2dbac vert 0 204
+U a0f2dbac vert u_half 0 float 1 0
+U a0f2dbac vert u_view 16 mat4 1 0
+U a0f2dbac vert u_proj 80 mat4 1 0
+U a0f2dbac vert u_origin 144 vec2 1 0
+U a0f2dbac vert u_cam_pos 160 vec3 1 0
+U a0f2dbac vert u_node 176 vec3 1 0
+U a0f2dbac vert u_morph 192 vec2 1 0
+U a0f2dbac vert u_grid 200 float 1 0
+I a0f2dbac a_xz 0 vec2
+B a0f2dbac frag 1 872
+U a0f2dbac frag u_cascade_vp 0 mat4 5 64
+U a0f2dbac frag u_cascade_split 320 float 5 16
+U a0f2dbac frag u_cascade_range 400 float 5 16
+U a0f2dbac frag u_cascade_texel 480 float 5 16
+U a0f2dbac frag u_sun_dir 560 vec3 1 0
+U a0f2dbac frag u_sun_color 576 vec3 1 0
+U a0f2dbac frag u_cam_pos 592 vec3 1 0
+U a0f2dbac frag u_prefilter_levels 604 float 1 0
+U a0f2dbac frag u_fog_density 608 float 1 0
+U a0f2dbac frag u_fog_height_falloff 612 float 1 0
+U a0f2dbac frag u_fog_base 616 float 1 0
+U a0f2dbac frag u_clip_y 620 float 1 0
+U a0f2dbac frag u_spec_scale 624 float 1 0
+U a0f2dbac frag u_sky_rot 632 vec2 1 0
+U a0f2dbac frag u_ibl_scale 640 vec3 1 0
+U a0f2dbac frag u_daylight 652 float 1 0
+U a0f2dbac frag u_fire_pos 656 vec3 1 0
+U a0f2dbac frag u_fire_color 672 vec3 1 0
+U a0f2dbac frag u_hand_pos 688 vec3 1 0
+U a0f2dbac frag u_hand_color 704 vec3 1 0
+U a0f2dbac frag u_hand_dir 720 vec3 1 0
+U a0f2dbac frag u_hand_cone 732 float 1 0
+U a0f2dbac frag u_ts_origin 736 vec2 1 0
+U a0f2dbac frag u_ts_half 744 float 1 0
+U a0f2dbac frag u_ts_on 748 float 1 0
+U a0f2dbac frag u_force_cascade 752 int 1 0
+U a0f2dbac frag u_cloud_shadow 756 float 1 0
+U a0f2dbac frag u_time 760 float 1 0
+U a0f2dbac frag u_half 764 float 1 0
+U a0f2dbac frag u_texel 768 float 1 0
+U a0f2dbac frag u_ortho_on 772 float 1 0
+U a0f2dbac frag u_carpet_on 776 float 1 0
+U a0f2dbac frag u_snow_line 780 float 1 0
+U a0f2dbac frag u_lake_level 784 float 1 0
+U a0f2dbac frag u_sea_level 788 float 1 0
+U a0f2dbac frag u_origin 792 vec2 1 0
+U a0f2dbac frag u_view 800 mat4 1 0
+U a0f2dbac frag u_far_split 864 float 1 0
+U a0f2dbac frag u_far_band 868 float 1 0
+T a0f2dbac u_brdf 2
+T a0f2dbac u_carpet 3
+T a0f2dbac u_grass_a 4
+T a0f2dbac u_grass_d 5
+T a0f2dbac u_grass_n 6
+T a0f2dbac u_height 7
+T a0f2dbac u_irradiance 8
+T a0f2dbac u_ortho 9
+T a0f2dbac u_prefilter 10
+T a0f2dbac u_rock_a 11
+T a0f2dbac u_rock_d 12
+T a0f2dbac u_rock_n 13
+T a0f2dbac u_shadow 14
+T a0f2dbac u_snow_d 15
+T a0f2dbac u_sunshadow 16
+T a0f2dbac u_tershadow 17
+T a0f2dbac u_ts_height 18
+P 5e4589a1 terrain.vert terrain.frag #define SUN_INLINE;#define NEAR_ONLY;
+B 5e4589a1 vert 0 204
+U 5e4589a1 vert u_half 0 float 1 0
+U 5e4589a1 vert u_view 16 mat4 1 0
+U 5e4589a1 vert u_proj 80 mat4 1 0
+U 5e4589a1 vert u_origin 144 vec2 1 0
+U 5e4589a1 vert u_cam_pos 160 vec3 1 0
+U 5e4589a1 vert u_node 176 vec3 1 0
+U 5e4589a1 vert u_morph 192 vec2 1 0
+U 5e4589a1 vert u_grid 200 float 1 0
+I 5e4589a1 a_xz 0 vec2
+B 5e4589a1 frag 1 872
+U 5e4589a1 frag u_cascade_vp 0 mat4 5 64
+U 5e4589a1 frag u_cascade_split 320 float 5 16
+U 5e4589a1 frag u_cascade_range 400 float 5 16
+U 5e4589a1 frag u_cascade_texel 480 float 5 16
+U 5e4589a1 frag u_sun_dir 560 vec3 1 0
+U 5e4589a1 frag u_sun_color 576 vec3 1 0
+U 5e4589a1 frag u_cam_pos 592 vec3 1 0
+U 5e4589a1 frag u_prefilter_levels 604 float 1 0
+U 5e4589a1 frag u_fog_density 608 float 1 0
+U 5e4589a1 frag u_fog_height_falloff 612 float 1 0
+U 5e4589a1 frag u_fog_base 616 float 1 0
+U 5e4589a1 frag u_clip_y 620 float 1 0
+U 5e4589a1 frag u_spec_scale 624 float 1 0
+U 5e4589a1 frag u_sky_rot 632 vec2 1 0
+U 5e4589a1 frag u_ibl_scale 640 vec3 1 0
+U 5e4589a1 frag u_daylight 652 float 1 0
+U 5e4589a1 frag u_fire_pos 656 vec3 1 0
+U 5e4589a1 frag u_fire_color 672 vec3 1 0
+U 5e4589a1 frag u_hand_pos 688 vec3 1 0
+U 5e4589a1 frag u_hand_color 704 vec3 1 0
+U 5e4589a1 frag u_hand_dir 720 vec3 1 0
+U 5e4589a1 frag u_hand_cone 732 float 1 0
+U 5e4589a1 frag u_ts_origin 736 vec2 1 0
+U 5e4589a1 frag u_ts_half 744 float 1 0
+U 5e4589a1 frag u_ts_on 748 float 1 0
+U 5e4589a1 frag u_force_cascade 752 int 1 0
+U 5e4589a1 frag u_cloud_shadow 756 float 1 0
+U 5e4589a1 frag u_time 760 float 1 0
+U 5e4589a1 frag u_half 764 float 1 0
+U 5e4589a1 frag u_texel 768 float 1 0
+U 5e4589a1 frag u_ortho_on 772 float 1 0
+U 5e4589a1 frag u_carpet_on 776 float 1 0
+U 5e4589a1 frag u_snow_line 780 float 1 0
+U 5e4589a1 frag u_lake_level 784 float 1 0
+U 5e4589a1 frag u_sea_level 788 float 1 0
+U 5e4589a1 frag u_origin 792 vec2 1 0
+U 5e4589a1 frag u_view 800 mat4 1 0
+U 5e4589a1 frag u_far_split 864 float 1 0
+U 5e4589a1 frag u_far_band 868 float 1 0
+T 5e4589a1 u_brdf 2
+T 5e4589a1 u_carpet 3
+T 5e4589a1 u_grass_a 4
+T 5e4589a1 u_grass_d 5
+T 5e4589a1 u_grass_n 6
+T 5e4589a1 u_height 7
+T 5e4589a1 u_irradiance 8
+T 5e4589a1 u_ortho 9
+T 5e4589a1 u_prefilter 10
+T 5e4589a1 u_rock_a 11
+T 5e4589a1 u_rock_d 12
+T 5e4589a1 u_rock_n 13
+T 5e4589a1 u_shadow 14
+T 5e4589a1 u_snow_d 15
+T 5e4589a1 u_sunshadow 16
+T 5e4589a1 u_tershadow 17
+T 5e4589a1 u_ts_height 18
 P bf3151c8 terrain.vert tersun.frag 
 B bf3151c8 vert 0 204
 U bf3151c8 vert u_half 0 float 1 0

packages/ludic.render3d/shaders/terrain.frag

@@ -200,6 +200,20 @@ void triC(bool cheap, sampler2D d, sampler2D nm, sampler2D am, vec3 p, vec3 dpx,
   if (cheap) { samplePlain(d, nm, am, p.xz * scale, dpx.xz * scale, dpy.xz * scale, alb, nrm, arm); nrm = normalize(vec3(nrm.x, 1.0, nrm.y) + vec3(0.0, 1e-3, 0.0)); }
   else sampleTri(d, nm, am, p, dpx, dpy, n, scale, alb, nrm, arm);
 }
+#ifdef SUN_INLINE
+// Vulkan: the cascades are read here, in the ground's own shader, and tersun.frag's pass - a second
+// rasterisation of every patch - is not drawn. The cliff tersun.frag describes is OpenGL's driver, and
+// OpenGL keeps that pass. The same normal, the same two tiers and the same cross-fade as tersun.frag.
+uniform float u_far_split;
+uniform float u_far_band;
+float sunInline(vec3 p, float dist, float viewDepth) {
+  vec3 Ng = normalize(texture(u_height, v_huv).gba);
+  if (dist > u_far_split + u_far_band) return sunShadowCheap(p, Ng, viewDepth);
+  if (dist < u_far_split - u_far_band) return sunShadow(p, Ng, viewDepth);
+  return mix(sunShadow(p, Ng, viewDepth), sunShadowCheap(p, Ng, viewDepth),
+             smoothstep(u_far_split - u_far_band, u_far_split + u_far_band, dist));
+}
+#endif
 vec3 groundShade(vec3 p, vec3 N, float slope, float dist, float viewDepth, bool cheap) {
   // ---- material weights ----
   float macro = fbmC(cheap, p.xz * 0.02, 2);
@@ -411,9 +425,11 @@ vec3 groundShade(vec3 p, vec3 N, float slope, float dist, float viewDepth, bool
   // ground's own relief is in the baked height-field shadow that sunShadow() applies
 #ifdef TFAST_1
   float shadow = 1.0;
+#elif defined(SUN_INLINE)
+  float shadow = sunInline(p, dist, viewDepth);
 #else
   // tersun.frag computed this for exactly this pixel; see the note there for why the
-  // cascade read cannot happen in here.
+  // cascade read cannot happen in here (on OpenGL).
   float shadow = texelFetch(u_sunshadow, ivec2(gl_FragCoord.xy), 0).r;
 #endif
   vec3 col = shade(p, n, alb, rough, metal, ao, shadow, viewDepth);
@@ -421,8 +437,10 @@ vec3 groundShade(vec3 p, vec3 N, float slope, float dist, float viewDepth, bool
   return col;
 }
 
+#ifndef SUN_INLINE
 uniform float u_far_split;
 uniform float u_far_band;
+#endif
 void main() {
   vec3 p = v_wpos;
   if (p.y < u_clip_y) discard;

packages/ludic.render3d/shaders/variants.list

@@ -46,5 +46,8 @@ skin.vert|shadow.frag|#define SHADOW_PASS;#define ALPHA_TEST;
 terrain.vert|terrain.frag|
 terrain.vert|terrain.frag|#define FAR_ONLY;
 terrain.vert|terrain.frag|#define NEAR_ONLY;
+terrain.vert|terrain.frag|#define SUN_INLINE;
+terrain.vert|terrain.frag|#define SUN_INLINE;#define FAR_ONLY;
+terrain.vert|terrain.frag|#define SUN_INLINE;#define NEAR_ONLY;
 terrain.vert|tersun.frag|
 water.vert|water.frag|

packages/ludic.render3d/terrain.ludic

@@ -32,6 +32,7 @@ var ter_prog: int = 0
 var ter_prog_far: int = 0
 var ter_prog_near: int = 0           # the detailed tier alone (NEAR_ONLY)
 var ter_sun_prog: int = 0            # tersun.frag: sun visibility into a screen buffer
+var ter_sun_inline: bool = false     # the ground's shader reads the cascades itself (Vulkan); no sun pass
 var ter_sun_tgt: Target = null       # that buffer, at the frame's size
 var ter_sun_refl: Target = null      # and at the reflection's, which is smaller
 var ter_sun_dumped: bool = false
@@ -423,6 +424,11 @@ function terrain_init_finish() -> void {
   # draw every patch with one tier's program (what each tier costs over a whole frame);
   # R3D_NOSPLIT goes back to the single program that holds both tiers.
   if Os.has_env("R3D_TFAST") { defs = defs + "#define TFAST_" + Os.env("R3D_TFAST") + "\n" }
+  # Vulkan reads the cascades inside the ground's own shader and draws no separate sun pass
+  # (terrain.frag, SUN_INLINE): one rasterisation of the patches instead of two, in the frame and in
+  # the reflection. R3D_SUN_PASS=1 keeps the pass there too, for comparing.
+  ter_sun_inline = gpu_kind == GPU_VK and not Os.has_env("R3D_SUN_PASS")
+  if ter_sun_inline { defs = defs + "#define SUN_INLINE\n" }
   ter_prog = r3d_program("terrain.vert", "terrain.frag", defs)
   ter_prog_far = r3d_program("terrain.vert", "terrain.frag", defs + "#define FAR_ONLY\n")
   ter_prog_near = r3d_program("terrain.vert", "terrain.frag", defs + "#define NEAR_ONLY\n")
@@ -487,9 +493,12 @@ function terrain_bind_prog(p: int) -> void {
   # The ground reads its sun visibility out of the buffer tersun.frag filled, and has no
   # use for the cascade array shadow_bind just put on this unit; leaving both bound under
   # one unit is undefined ground, so the array comes off first.
-  gpu_tex_unit(15)
-  gpu_tex_bind(GPU_TEX2D_ARRAY, 0)
-  r3d_bind_2d(p, "u_sunshadow", 15, ter_sun_tex)
+  # (With the read inline, the array shadow_bind put there is exactly what the ground wants.)
+  if not ter_sun_inline {
+    gpu_tex_unit(15)
+    gpu_tex_bind(GPU_TEX2D_ARRAY, 0)
+    r3d_bind_2d(p, "u_sunshadow", 15, ter_sun_tex)
+  }
 }
 
 # The visibility buffer for the size being drawn into. The reflection is rendered at its
@@ -559,6 +568,7 @@ function terrain_sun_pass(w: int, h: int, depth: int) -> Target {
 }
 
 function terrain_sun_prepare() -> void {
+  if ter_sun_inline { return }
   ter_sun_tex = terrain_sun_pass(post_w, post_h, post_hdr.depth).color
   ter_sun_done = true
 }
@@ -571,7 +581,7 @@ function terrain_draw() -> void {
   var vw = post_w
   var vh = post_h
   if ter_reflect { vw = water_refl.w;  vh = water_refl.h }
-  if not ter_sun_done {
+  if not ter_sun_done and not ter_sun_inline {
     var dep = post_hdr.depth
     if ter_reflect { dep = water_refl.depth }
     ter_sun_tex = terrain_sun_pass(vw, vh, dep).color