Working traverse2

shaders/cube.wgsl

@@ -98,6 +98,99 @@ struct ColorTile
 @group(0) @binding(0) var<storage> structure_tiles : array<StructureTile>;
 @group(0) @binding(1) var<storage> color_tiles : array<ColorTile>;
 
+fn traverse2(max_depth: u32, root_color: vec4<f32>, root_subdivided: bool, eye_pos: vec3<f32>, ray_dir: vec3<f32>) -> vec4<f32>
+{
+    if max_depth == 0 || !root_subdivided
+    {
+        return inferno_quintic(0.);
+        return root_color;
+    }
+
+    var node_structs = structure_tiles[0];
+    var node_colors = color_tiles[0];
+    var node_tile_idx = u32(0);
+    //var stack = array<u32, 4>(0, 0, 0, 0);
+    var stack_struct = array<StructureTile, 4>(node_structs, node_structs, node_structs, node_structs);
+    var stack_colors = array<ColorTile, 4>(node_colors, node_colors, node_colors, node_colors);
+
+    var voxel_scale = N * N * N;
+    var scale_exp = 3;
+    let voxel_scale_lut = array<u32, 4>(
+        1,
+        N,
+        N * N,
+        N * N * N
+    );
+
+    let dist = 1. / ray_dir;
+    let origin = eye_pos * 256.;
+    var pos = origin; // In voxel space
+    let offset = - origin * dist;
+    let wall_offset = select(vec3(0.), vec3(1.), ray_dir > vec3(0.));
+    var voxel_pos = clamp(vec3<u32>(floor(pos)), vec3(0), vec3(N * N * N * N - 1));
+    let max = 20.;
+    for(var iter = 0; iter < 60; iter++) 
+    {
+        var voxel_pos = vec3<u32>(floor(pos));
+        var child_pos = voxel_pos >> vec3<u32>(scale_exp * 2);
+        var local_child_pos = child_pos & vec3<u32>(3);
+        var child_idx = local_child_pos.x + local_child_pos.y * N + local_child_pos.z * N * N;
+        
+        while (node_structs.children[child_idx] >> 31) != 0 && (u32(4 - scale_exp) <  max_depth)
+        {
+            //stack[scale_exp] = u32(node_tile_idx);
+            stack_struct[scale_exp] = node_structs;
+            stack_colors[scale_exp] = node_colors;
+            scale_exp -= 1;
+            voxel_scale = voxel_scale_lut[scale_exp];
+
+            node_tile_idx = node_structs.children[child_idx] & 0x3fffffff;
+            node_structs = structure_tiles[node_tile_idx];
+            node_colors = color_tiles[node_tile_idx];
+
+            child_pos = voxel_pos >> vec3<u32>(scale_exp * 2);
+            local_child_pos = child_pos & vec3<u32>(3);
+            child_idx = local_child_pos.x + local_child_pos.y * N + local_child_pos.z * N * N;
+        }
+
+        if node_colors.colors[child_idx].w != 0
+        {
+            return inferno_quintic(f32(iter) / max);
+            return node_colors.colors[child_idx];
+        }
+
+        // Compute intersection
+        let global_voxel = vec3<f32>(child_pos * voxel_scale);
+        let cell_max = global_voxel + vec3<f32>(voxel_scale) * wall_offset;
+
+        let t1 = fma(dist, cell_max, offset);
+
+        let t_far = min(t1.x, min(t1.y, t1.z)) + 0.001;
+        pos = origin + t_far * ray_dir;
+
+        let diff = vec3<u32>(pos + 256) ^ vec3<u32>(global_voxel + 256);
+        let diff_exp = (firstLeadingBit((diff.x | diff.y | diff.z)) >> 1);
+
+        if diff_exp > 3
+        {
+            return inferno_quintic(f32(iter) / max);
+            discard;
+        }
+
+        if diff_exp > u32(scale_exp)
+        {
+            scale_exp = i32(diff_exp);
+
+            voxel_scale = voxel_scale_lut[scale_exp];
+            //node_tile_idx = stack[scale_exp];
+            node_structs = stack_struct[scale_exp];
+            node_colors = stack_colors[scale_exp];
+        }     
+    }
+
+    return vec4(1., 0., 1., 1.);
+}
+
 //@fragment
 fn traverse(max_depth: u32, root_color: vec4<f32>, root_subdivided: bool, eye_pos: vec3<f32>, ray_dir: vec3<f32>) -> vec4<f32>
 {
@@ -215,6 +308,7 @@ fn traverse(max_depth: u32, root_color: vec4<f32>, root_subdivided: bool, eye_po
             let next_child = current_child_pos + advance_mask * step;
             if any(next_child < vec3(0)) || any(next_child >= vec3(i32(N)))
             {
+
                 let aligned_child = select(vec3(0), vec3(i32(N)), vec3(step) > vec3(0));
                 let masked_aligned = advance_mask * ((aligned_child * i32(current_child_size)) + current_node_offset);
                 let exiting_axis = masked_aligned.x + masked_aligned.y + masked_aligned.z + 256;
@@ -304,7 +398,7 @@ fn fragment_tree_main(in: VertexOutput) -> @location(0) vec4<f32>
         div *= 4;
     }
     overlay = 1.;
-    return overlay * traverse(4, in.root_color, in.root_subdivided != 0, in.eye_pos - in.cube_pos, dir);
+    return overlay * traverse2(4, in.root_color, in.root_subdivided != 0, hit_pos, dir);
 }
 
 @fragment
@@ -437,3 +531,14 @@ fn intersectAABB(rayOrigin: vec3<f32>, rayDir: vec3<f32>, boxMin: vec3<f32>, box
     let tFar = min(min(t2.x, t2.y), t2.z);
     return vec2(tNear, tFar);
 };
+
+fn inferno_quintic( xx: f32 ) -> vec4<f32>
+{
+	let x = saturate(xx);
+	let x1 = vec4( 1.0, x, x * x, x * x * x ); // 1 x x2 x3
+	let x2 = x1 * x1.w * x; // x4 x5 x6 x7
+	return vec4(saturate( vec3(
+		dot( x1.xyzw, vec4( -0.027780558, 1.228188385, 0.278906882, 3.892783760 ) ) + dot( x2.xy, vec2( -8.490712758, 4.069046086 ) ),
+		dot( x1.xyzw, vec4( 0.014065206, 0.015360518, 1.605395918, -4.821108251 ) ) + dot( x2.xy, vec2( 8.389314011, -4.193858954 ) ),
+		dot( x1.xyzw, vec4( -0.019628385, 3.122510347, -5.893222355, 2.798380308 ) ) + dot( x2.xy, vec2( -3.608884658, 4.324996022 ) ) ) ), 1.);
+}