@woosh/meep-engine 3.30.0 → 3.32.0

package/src/shade/renderer/rasterize/expand/bucket/meshlet/shader_meshlets_count_by_rasterization_bucket.js

@@ -1,5 +1,5 @@
 import { assert } from "../../../../../../core/assert.js";
-import { chunk_grid3d_to_index } from "../../../../shader/chunk/grid/chunk_grid3d_to_index.js";
+import { chunk_folded_workgroup_index } from "../../../../shader/chunk/grid/chunk_folded_workgroup_index.js";
 import { atomic_u32, u32 } from "../../../../shader/type/WEBGPU_WGSL_PRIMITIVE_TYPES.js";
 import { WebGPUArray } from "../../../../shader/type/WebGPUArray.js";
 import { WebGPUStruct } from "../../../../shader/type/WebGPUStruct.js";
@@ -39,11 +39,10 @@ var<workgroup> wg_buckets: array<atomic<u32>, ${RASTERIZATION_BUCKET_LIMIT}>;
 @compute @workgroup_size(${GROUP_SIZE})
 fn main(
     @builtin(workgroup_id) workgroup_id : vec3<u32>,
-    @builtin(num_workgroups) group_count : vec3<u32>,
     @builtin(local_invocation_index) local_id : u32,
 ){
     // folded dispatch — see IndirectComputeCommandBuildPass
-    let thread_id = grid3d_to_index(workgroup_id, group_count.xy) * ${GROUP_SIZE}u + local_id;
+    let thread_id = folded_workgroup_index(workgroup_id) * ${GROUP_SIZE}u + local_id;
 
     let input_size = min(input.count, arrayLength(&input.elements));
 
@@ -82,7 +81,7 @@ fn main(
 }
     `,
     [
-        chunk_grid3d_to_index,
+        chunk_folded_workgroup_index,
         chunk_rasterization_material_bucket,
         chunk_scene_database_access,
     ]

package/src/shade/renderer/rasterize/expand/bucket/meshlet/shader_meshlets_extract_bucket.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"shader_meshlets_extract_bucket.d.ts","sourceRoot":"","sources":["../../../../../../../../src/shade/renderer/rasterize/expand/bucket/meshlet/shader_meshlets_extract_bucket.js"],"names":[],"mappings":"AAwEA,2DAIG;8BArE2B,qCAAqC"}
+{"version":3,"file":"shader_meshlets_extract_bucket.d.ts","sourceRoot":"","sources":["../../../../../../../../src/shade/renderer/rasterize/expand/bucket/meshlet/shader_meshlets_extract_bucket.js"],"names":[],"mappings":"AA0EA,2DAIG;8BAvE2B,qCAAqC"}

package/src/shade/renderer/rasterize/expand/bucket/meshlet/shader_meshlets_extract_bucket.js

@@ -1,7 +1,7 @@
 import { WebGPUArray } from "../../../../shader/type/WebGPUArray.js";
 import { WebGPUStruct } from "../../../../shader/type/WebGPUStruct.js";
 import { MESHLET_COLLECTION_STRUCT } from "../../../MESHLET_COLLECTION_STRUCT.js";
-import { chunk_grid3d_to_index } from "../../../../shader/chunk/grid/chunk_grid3d_to_index.js";
+import { chunk_folded_workgroup_index } from "../../../../shader/chunk/grid/chunk_folded_workgroup_index.js";
 import { U32_CONTAINER_STRUCT } from "../../../../shader/type/U32_CONTAINER_STRUCT.js";
 import { CodeChunk } from "../../../../shader/compiler/CodeChunk.js";
 import { COMPUTE_SHADER_DEFAULT_GROUP_SIZE } from "../../../../shader/COMPUTE_SHADER_DEFAULT_GROUP_SIZE.js";
@@ -31,16 +31,18 @@ const body = CodeChunk.from(
 @compute @workgroup_size(${COMPUTE_SHADER_DEFAULT_GROUP_SIZE})
 fn main(
     @builtin(workgroup_id) workgroup_id : vec3<u32>,
-    @builtin(num_workgroups) group_count : vec3<u32>,
     @builtin(local_invocation_index) local_index : u32,
 ){
     // folded dispatch — see IndirectComputeCommandBuildPass
-    let thread_id = grid3d_to_index(workgroup_id, group_count.xy) * ${COMPUTE_SHADER_DEFAULT_GROUP_SIZE}u + local_index;
+    let thread_id = folded_workgroup_index(workgroup_id) * ${COMPUTE_SHADER_DEFAULT_GROUP_SIZE}u + local_index;
     
     let bucket_id = bucket_index.value;
     
     let bucket_start = bucket_data.headers[bucket_id].offset; 
-    let bucket_size = bucket_data.headers[bucket_id].count;
+    
+    // clamped to what the output can hold, the way the input read below already is: the header is
+    // allowed to over-count, and the count published here is what the expansion downstream reads
+    let bucket_size = min(bucket_data.headers[bucket_id].count, arrayLength(&output.elements));
     
     if(thread_id == 0u){
         // copy collection size
@@ -65,7 +67,7 @@ fn main(
     
 }
     `, [
-        chunk_grid3d_to_index,
+        chunk_folded_workgroup_index,
         MESHLET_COLLECTION_STRUCT.declaration_chunk,
     ]
 )

package/src/shade/renderer/rasterize/expand/bucket/meshlet/shader_meshlets_rasterization_bucket_sort.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"shader_meshlets_rasterization_bucket_sort.d.ts","sourceRoot":"","sources":["../../../../../../../../src/shade/renderer/rasterize/expand/bucket/meshlet/shader_meshlets_rasterization_bucket_sort.js"],"names":[],"mappings":"AA4FA,sEAIE;8BAvF4B,qCAAqC"}
+{"version":3,"file":"shader_meshlets_rasterization_bucket_sort.d.ts","sourceRoot":"","sources":["../../../../../../../../src/shade/renderer/rasterize/expand/bucket/meshlet/shader_meshlets_rasterization_bucket_sort.js"],"names":[],"mappings":"AA0HA,sEAIE;8BApH4B,qCAAqC"}

package/src/shade/renderer/rasterize/expand/bucket/meshlet/shader_meshlets_rasterization_bucket_sort.js

@@ -1,8 +1,9 @@
+import { assert } from "../../../../../../core/assert.js";
 import { u32 } from "../../../../shader/type/WEBGPU_WGSL_PRIMITIVE_TYPES.js";
 import { WebGPUArray } from "../../../../shader/type/WebGPUArray.js";
 import { WebGPUStruct } from "../../../../shader/type/WebGPUStruct.js";
 import { MESHLET_COLLECTION_STRUCT } from "../../../MESHLET_COLLECTION_STRUCT.js";
-import { chunk_grid3d_to_index } from "../../../../shader/chunk/grid/chunk_grid3d_to_index.js";
+import { chunk_folded_workgroup_index } from "../../../../shader/chunk/grid/chunk_folded_workgroup_index.js";
 import { MATERIAL_METADATA_STRUCT } from "../../../../material/MATERIAL_METADATA_STRUCT.js";
 import { chunk_scene_database_access } from "../../../../scene/chunk_scene_database_access.js";
 import { CodeChunk } from "../../../../shader/compiler/CodeChunk.js";
@@ -33,58 +34,87 @@ resources.createGroup()
     .addStorageBuffer("output", RASTERIZATION_MESHLET_BUCKET_SORT_OUTPUT_STRUCT_WRITE, true)
 
 
+const GROUP_SIZE = COMPUTE_SHADER_DEFAULT_GROUP_SIZE;
+
+// one lane per bucket makes the workgroup's claims
+assert.lessThanOrEqual(RASTERIZATION_BUCKET_LIMIT, GROUP_SIZE);
+
+/**
+ * The meshlet twin of {@link shader_instances_rasterization_bucket_sort}: the same workgroup
+ * histogram, for the same reason.
+ */
 const body = CodeChunk.from(
     //language=WGSL
     `
-@compute @workgroup_size(${COMPUTE_SHADER_DEFAULT_GROUP_SIZE})
+var<workgroup> wg_count: array<atomic<u32>, ${RASTERIZATION_BUCKET_LIMIT}>;
+var<workgroup> wg_base: array<u32, ${RASTERIZATION_BUCKET_LIMIT}>;
+    
+@compute @workgroup_size(${GROUP_SIZE})
 fn main(
     @builtin(workgroup_id) workgroup_id : vec3<u32>,
-    @builtin(num_workgroups) group_count : vec3<u32>,
     @builtin(local_invocation_index) local_index : u32,
 ){
-
+    
     // folded dispatch — see IndirectComputeCommandBuildPass
-    let index = grid3d_to_index(workgroup_id, group_count.xy) * ${COMPUTE_SHADER_DEFAULT_GROUP_SIZE}u + local_index;
+    let index = folded_workgroup_index(workgroup_id) * ${GROUP_SIZE}u + local_index;
     
     let input_size = min(input.count, arrayLength(&input.elements));
     
-    if(index >= input_size){
-        return;
+    // a lane past the input does not return: every lane of the workgroup has to reach the barriers
+    let valid = index < input_size;
+    
+    var bucket_index = 0u;
+    var rank = 0u;
+    
+    if(valid){
+        let mesh_index = input.elements[index].mesh;
+    
+        let mesh = scene_read_mesh(&scene_database, mesh_index);
+    
+        let material = materials[mesh.material];
+    
+        bucket_index = rasterization_material_bucket(
+            material.transparency_mode,
+            material.draw_mode,
+            material.draw_side
+        );
+    
+        // this element's place in the workgroup's run of its bucket
+        rank = atomicAdd(&wg_count[bucket_index], 1u);
     }
     
-    let meshlet_def = input.elements[index];
-
-    let mesh_index = meshlet_def.mesh;
-
-    let mesh = scene_read_mesh(&scene_database, mesh_index);
-
-    let material = materials[mesh.material];
+    workgroupBarrier();
     
-    let bucket_index = rasterization_material_bucket(
-        material.transparency_mode,
-        material.draw_mode,
-        material.draw_side
-    );
+    // one lane per bucket claims the workgroup's whole run with one atomic
+    if(local_index < ${RASTERIZATION_BUCKET_LIMIT}u){
+        let run_length = atomicLoad(&wg_count[local_index]);
     
-    let element_offset = atomicAdd(&output.headers[bucket_index].count, 1u);
+        if(run_length > 0u){
+            let run_offset = atomicAdd(&output.headers[local_index].count, run_length);
     
-    let bucket_offset = offsets[bucket_index];
+            if(run_offset == 0u){
+                // first element in bucket, write offset as well
+                output.headers[local_index].offset = offsets[local_index];
+            }
     
-    if(element_offset == 0u){
-        // first element in bucket, write offset as well
-        output.headers[bucket_index].offset = bucket_offset;
+            wg_base[local_index] = run_offset;
+        }
     }
     
-    let output_offset = bucket_offset + element_offset;
+    workgroupBarrier();
+    
+    if(valid){
+        let output_offset = offsets[bucket_index] + wg_base[bucket_index] + rank;
     
-    if(output_offset < arrayLength(&output.elements)){
-        // guard against out-of-bounds access
-        output.elements[output_offset] = meshlet_def;
+        if(output_offset < arrayLength(&output.elements)){
+            // guard against out-of-bounds access
+            output.elements[output_offset] = input.elements[index];
+        }
     }
     
 }
     `, [
-        chunk_grid3d_to_index,
+        chunk_folded_workgroup_index,
         chunk_rasterization_material_bucket,
         chunk_scene_database_access,
     ]

package/src/shade/renderer/rasterize/expand/mesh/shader_mesh_expand_to_meshlets_prefix.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"shader_mesh_expand_to_meshlets_prefix.d.ts","sourceRoot":"","sources":["../../../../../../../src/shade/renderer/rasterize/expand/mesh/shader_mesh_expand_to_meshlets_prefix.js"],"names":[],"mappings":"AAgGA,kEAIE;8BA3F4B,kCAAkC"}
+{"version":3,"file":"shader_mesh_expand_to_meshlets_prefix.d.ts","sourceRoot":"","sources":["../../../../../../../src/shade/renderer/rasterize/expand/mesh/shader_mesh_expand_to_meshlets_prefix.js"],"names":[],"mappings":"AA+FA,kEAIE;8BA1F4B,kCAAkC"}

package/src/shade/renderer/rasterize/expand/mesh/shader_mesh_expand_to_meshlets_prefix.js

@@ -1,7 +1,7 @@
 import { u32 } from "../../../shader/type/WEBGPU_WGSL_PRIMITIVE_TYPES.js";
 import { WebGPUArray } from "../../../shader/type/WebGPUArray.js";
 import { COLLECTION_U32_STRUCT_READ } from "../../../gpu_primitive/COLLECTION_U32_STRUCT_READ.js";
-import { chunk_grid3d_to_index } from "../../../shader/chunk/grid/chunk_grid3d_to_index.js";
+import { chunk_folded_workgroup_index } from "../../../shader/chunk/grid/chunk_folded_workgroup_index.js";
 import { MESHLET_COLLECTION_STRUCT } from "../../MESHLET_COLLECTION_STRUCT.js";
 import { chunk_scene_database_access } from "../../../scene/chunk_scene_database_access.js";
 import { chunk_lower_bound_branchless } from "../../../shader/chunk/math/chunk_lower_bound_branchless.js";
@@ -33,7 +33,6 @@ const body = CodeChunk.from(
 @compute @workgroup_size(${COMPUTE_SHADER_DEFAULT_GROUP_SIZE})
 fn unpack(
     @builtin(workgroup_id) workgroup_id : vec3<u32>,
-    @builtin(num_workgroups) group_count : vec3<u32>,
     @builtin(local_invocation_index) local_index : u32,
 ){
 
@@ -42,7 +41,7 @@ fn unpack(
      ordinary mesh rows expand to more meshlets than one dispatch dimension reaches. See
      IndirectComputeCommandBuildPass. While y and z are 1 this is global_invocation_id.x exactly.
     */
-    let input_index = grid3d_to_index(workgroup_id, group_count.xy) * ${COMPUTE_SHADER_DEFAULT_GROUP_SIZE}u + local_index;
+    let input_index = folded_workgroup_index(workgroup_id) * ${COMPUTE_SHADER_DEFAULT_GROUP_SIZE}u + local_index;
 
     let input_mesh_count = prefix_sums.count;
     
@@ -88,7 +87,7 @@ fn unpack(
 
 }`, [
         MESHLET_DEFINITION_STRUCT.declaration_chunk,
-        chunk_grid3d_to_index,
+        chunk_folded_workgroup_index,
         chunk_lower_bound_branchless,
         chunk_scene_database_access,
     ]);

package/src/shade/renderer/rasterize/expand/mesh/shader_prefix_sum_to_command.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"shader_prefix_sum_to_command.d.ts","sourceRoot":"","sources":["../../../../../../../src/shade/renderer/rasterize/expand/mesh/shader_prefix_sum_to_command.js"],"names":[],"mappings":"AA2DA,yDAIG;8BAvD2B,kCAAkC"}
+{"version":3,"file":"shader_prefix_sum_to_command.d.ts","sourceRoot":"","sources":["../../../../../../../src/shade/renderer/rasterize/expand/mesh/shader_prefix_sum_to_command.js"],"names":[],"mappings":"AA8DA,yDAIG;8BA1D2B,kCAAkC"}

package/src/shade/renderer/rasterize/expand/mesh/shader_prefix_sum_to_command.js

@@ -32,13 +32,16 @@ fn main(){
      8,388,480 one dimension reaches, so every counter went on reporting a complete frame over a
      meshlet array the expansion had never written.
 
-     The shaders this drives read their index through grid3d_to_index, which is the same flattening
-     and is exactly global_invocation_id.x while y and z are 1. See
+     The mesh expansion this drives reads its index through folded_workgroup_index, which is exactly
+     global_invocation_id.x while y and z are 1; the particle emit it also sizes still flattens with
+     grid3d_to_index over num_workgroups, which comes to the same index. See
      IndirectComputeCommandBuildPass, whose own folded variant this mirrors — this one exists only
      because the count it needs sits at a data-dependent index rather than at a fixed offset.
 
      Clamped to the figure every WebGPU device is bound to allow rather than the one this device
      reports, which the shader has no way to be told; clamping low only moves work into y sooner.
+     It is also the stride folded_workgroup_index flattens by, so the two builders have to agree on
+     it: a row of x shorter than that would leave a gap in the index.
     */
     const LIMIT = ${WEBGPU_GUARANTEED_MAX_COMPUTE_WORKGROUPS_PER_DIMENSION}u;
 

package/src/shade/renderer/rasterize/expand/pass1/filter_meshlets.2-way.compute.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"filter_meshlets.2-way.compute.d.ts","sourceRoot":"","sources":["../../../../../../../src/shade/renderer/rasterize/expand/pass1/filter_meshlets.2-way.compute.js"],"names":[],"mappings":"AA6JA,uDAIG;8BAtI2B,kCAAkC"}
+{"version":3,"file":"filter_meshlets.2-way.compute.d.ts","sourceRoot":"","sources":["../../../../../../../src/shade/renderer/rasterize/expand/pass1/filter_meshlets.2-way.compute.js"],"names":[],"mappings":"AA4JA,uDAIG;8BArI2B,kCAAkC"}

package/src/shade/renderer/rasterize/expand/pass1/filter_meshlets.2-way.compute.js

@@ -19,7 +19,7 @@ import {
     chunk_aabb3_project_perspective
 } from "../../../shader/chunk/geometry/aabb3/chunk_aabb3_project_perspective.js";
 import { chunk_array_to_aabb3 } from "../../../shader/chunk/geometry/aabb3/chunk_array_to_aabb3.js";
-import { chunk_grid3d_to_index } from "../../../shader/chunk/grid/chunk_grid3d_to_index.js";
+import { chunk_folded_workgroup_index } from "../../../shader/chunk/grid/chunk_folded_workgroup_index.js";
 import {
     chunk_aabb2_clip_overlaps_texel_centers
 } from "../../../shader/chunk/texture/chunk_aabb2_clip_overlaps_texel_centers.js";
@@ -53,12 +53,11 @@ const body = CodeChunk.from(
 @compute @workgroup_size(${COMPUTE_SHADER_DEFAULT_GROUP_SIZE})
 fn unpack(
     @builtin(workgroup_id) workgroup_id : vec3<u32>,
-    @builtin(num_workgroups) group_count : vec3<u32>,
     @builtin(local_invocation_index) local_index : u32,
 ){
 
     // folded dispatch — see IndirectComputeCommandBuildPass
-    let input_index = grid3d_to_index(workgroup_id, group_count.xy) * ${COMPUTE_SHADER_DEFAULT_GROUP_SIZE}u + local_index;
+    let input_index = folded_workgroup_index(workgroup_id) * ${COMPUTE_SHADER_DEFAULT_GROUP_SIZE}u + local_index;
 
     if(input_index >= input.count){
         return;
@@ -137,7 +136,7 @@ fn unpack(
 
         AABB3_STRUCT.declaration_chunk,
 
-        chunk_grid3d_to_index,
+        chunk_folded_workgroup_index,
 
         chunk_array_to_aabb3,
 

package/src/shade/renderer/rasterize/fast/graph_rasterize_scene_fast.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"graph_rasterize_scene_fast.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/rasterize/fast/graph_rasterize_scene_fast.js"],"names":[],"mappings":"AAoBA;;;;;;;;;;GAUG;AACH;;;;;;;IAFa;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,aAAa,EAAE,MAAM,EAAE,CAAA;CAAC,CA0J3D;AAED;;;;;;;;;;;;;;;;;;;GAmBG;AACH;IAJ+B,KAAK;IACC,GAAG;IACA,MAAM;SA8C7C"}
+{"version":3,"file":"graph_rasterize_scene_fast.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/rasterize/fast/graph_rasterize_scene_fast.js"],"names":[],"mappings":"AAoBA;;;;;;;;;;GAUG;AACH;;;;;;;IAFa;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,aAAa,EAAE,MAAM,EAAE,CAAA;CAAC,CAkK3D;AAED;;;;;;;;;;;;;;;;;;;GAmBG;AACH;IAJ+B,KAAK;IACC,GAAG;IACA,MAAM;SA8C7C"}

package/src/shade/renderer/rasterize/fast/graph_rasterize_scene_fast.js

@@ -13,8 +13,8 @@ import { graph_vgeo_expand_if_enabled } from "../../geometry/virtual/runtime/gra
 import { RasterizationCollectionType } from "../expand/RasterizationCollectionType.js";
 import { construct_primitive_state } from "../bucket/construct_primitive_state.js";
 import { rasterization_opaque_buckets } from "../bucket/rasterization_opaque_buckets.js";
-import { rasterization_has_opaque_bucket } from "../bucket/rasterization_has_opaque_bucket.js";
 import { graph_mesh_frustum_cull } from "../compute/graph_mesh_frustum_cull.js";
+import { graph_clear_raster_targets } from "../graph_clear_raster_targets.js";
 import { graph_rasterize_meshes_alpha_tested } from "../native/graph_rasterize_meshes_alpha_tested.js";
 import { graph_extract_meshes_from_bucket } from "./graph_extract_meshes_from_bucket.js";
 
@@ -122,29 +122,39 @@ export function graph_rasterize_scene_fast_raw({
         is_first_draw = false;
     }
 
-    // alpha-tested: reuse the existing native pass but skip HZB filtering by passing hzb=-1
-    if (job.alpha_tested_pass !== undefined) {
-        /**
-         * The opaque path's first draw is the only thing that clears the visibility-buffer colour
-         * attachments and the depth attachment — the clear values live on the opaque pass descriptor
-         * and reach the targets through its `loadOp`, and every pass after it is recorded with
-         * `load`. When the scene has no opaque bucket that draw never happens, so the first pass that
-         * does run owns the clear. The alpha-tested descriptor carries the same clear values
-         * (`[0,0,0,0]`, VIZ_BUFFER_INVALID_MESH_VALUE, DEPTH_BACKGROUND_VALUE), so handing it over
-         * changes nothing about what the targets are cleared to.
-         *
-         * Wherever an opaque bucket exists — nearly every scene — this starts true and every call
-         * below asks for `false`, exactly as before.
-         */
-        let targets_cleared = rasterization_has_opaque_bucket(active_buckets);
+    if (is_first_draw) {
+        /*
+         No opaque bucket, so no draw above cleared the target, and the rectangle holds whatever was
+         there before: in the atlas, the shadow of whatever was in the scene then; in a point light's
+         cube, which is a pooled transient, whatever the pool's last user left. The clear goes in the
+         first draw's place, and everything after this loads.
+        */
+        const cleared = graph_clear_raster_targets({
+            graph,
+            depth: target_depth,
+            color_targets: targets_color,
+            pass_descriptor: job.opaque_pass,
+            viewport,
+        });
 
+        target_depth = cleared.depth;
+        targets_color = cleared.color_targets;
+    }
 
+    // alpha-tested: reuse the existing native pass but skip HZB filtering by passing hzb=-1
+    if (job.alpha_tested_pass !== undefined) {
+        /*
+         Both alpha-tested passes load, never clear: the targets were cleared above, by the first
+         opaque draw or in its place. The clear used to be handed to the first alpha-tested draw, which
+         left a scene with neither kind of draw — nothing in it, or only transparent surfaces — with
+         nothing clearing at all.
+        */
         const alpha_tested_front = graph_rasterize_meshes_alpha_tested({
             graph,
             buckets: gr_render_buckets,
             view_ctx: view_ctx,
             render_pass_descriptor: job.alpha_tested_pass,
-            clear_target: !targets_cleared,
+            clear_target: false,
             side: ShadeDrawSide.Front,
             cull_policy: job.cull_policy,
             limits,
@@ -157,14 +167,13 @@ export function graph_rasterize_scene_fast_raw({
 
         target_depth = alpha_tested_front.depth;
         targets_color = alpha_tested_front.color_targets;
-        targets_cleared = targets_cleared || alpha_tested_front.executed;
 
         const alpha_tested_double = graph_rasterize_meshes_alpha_tested({
             graph,
             buckets: gr_render_buckets,
             view_ctx: view_ctx,
             render_pass_descriptor: job.alpha_tested_pass,
-            clear_target: !targets_cleared,
+            clear_target: false,
             side: ShadeDrawSide.Double,
             cull_policy: job.cull_policy,
             limits,
@@ -177,7 +186,6 @@ export function graph_rasterize_scene_fast_raw({
 
         target_depth = alpha_tested_double.depth;
         targets_color = alpha_tested_double.color_targets;
-        targets_cleared = targets_cleared || alpha_tested_double.executed;
     }
 
     return {target_depth, targets_color};

package/src/shade/renderer/rasterize/forward/SHADING_SCENE_RESOURCE_GROUP.d.ts

@@ -1,3 +1,23 @@
+/**
+ * Group 1 of every transparent pipeline, whatever the template: what the forward vertex stage pulls
+ * vertices from.
+ *
+ * `materials` is the metadata table: the vertex stage reads the side a material draws from it, per
+ * instance, the way the opaque and alpha-tested rasterizers do — which is what takes group 0 out of
+ * the vertex stage, so the material's own group 0, fragment-only, binds here as it does in the
+ * G-buffer pass.
+ *
+ * Vertex-only: no fragment stage of the transparent passes reads any of it.
+ *
+ * @type {ShaderResourceGroupDescriptor}
+ */
 export const SHADING_SCENE_RESOURCE_GROUP: ShaderResourceGroupDescriptor;
+/**
+ * The layout the transparent passes build their group-1 bind group against, and every transparent
+ * pipeline declares.
+ *
+ * @type {BindGroupLayoutDescriptor}
+ */
+export const SHADING_SCENE_LAYOUT: BindGroupLayoutDescriptor;
 import { ShaderResourceGroupDescriptor } from "../../shader/resource/ShaderResourceGroupDescriptor.js";
 //# sourceMappingURL=SHADING_SCENE_RESOURCE_GROUP.d.ts.map

package/src/shade/renderer/rasterize/forward/SHADING_SCENE_RESOURCE_GROUP.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"SHADING_SCENE_RESOURCE_GROUP.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/rasterize/forward/SHADING_SCENE_RESOURCE_GROUP.js"],"names":[],"mappings":"AAQA,yEAAgF;8CANlC,wDAAwD"}
+{"version":3,"file":"SHADING_SCENE_RESOURCE_GROUP.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/rasterize/forward/SHADING_SCENE_RESOURCE_GROUP.js"],"names":[],"mappings":"AAUA;;;;;;;;;;;;GAYG;AACH,2CAFU,6BAA6B,CAEyC;AAYhF;;;;;GAKG;AACH,6DAAqG;8CAtCvD,wDAAwD"}

package/src/shade/renderer/rasterize/forward/SHADING_SCENE_RESOURCE_GROUP.js

@@ -1,3 +1,4 @@
+import { GPUShaderStage } from "../../../descriptor/GPUShaderStage.js";
 import { u32 } from "../../shader/type/WEBGPU_WGSL_PRIMITIVE_TYPES.js";
 import { WebGPUArray } from "../../shader/type/WebGPUArray.js";
 import { ShaderResourceGroupDescriptor } from "../../shader/resource/ShaderResourceGroupDescriptor.js";
@@ -5,7 +6,21 @@ import { GEOMETRY_ATTRIBUTE_DATA_TYPE } from "../../geometry/GEOMETRY_ATTRIBUTE_
 import { GEOMETRY_METADATA_STRUCT } from "../../geometry/GEOMETRY_METADATA_STRUCT.js";
 import { MESHLET_DEFINITION_STRUCT } from "../../geometry/meshlet/encoding/MESHLET_DEFINITION_STRUCT.js";
 import { MESHLET_METADATA_STRUCT } from "../../geometry/meshlet/encoding/MESHLET_METADATA_STRUCT.js";
+import { MATERIAL_METADATA_STRUCT } from "../../material/MATERIAL_METADATA_STRUCT.js";
 
+/**
+ * Group 1 of every transparent pipeline, whatever the template: what the forward vertex stage pulls
+ * vertices from.
+ *
+ * `materials` is the metadata table: the vertex stage reads the side a material draws from it, per
+ * instance, the way the opaque and alpha-tested rasterizers do — which is what takes group 0 out of
+ * the vertex stage, so the material's own group 0, fragment-only, binds here as it does in the
+ * G-buffer pass.
+ *
+ * Vertex-only: no fragment stage of the transparent passes reads any of it.
+ *
+ * @type {ShaderResourceGroupDescriptor}
+ */
 export const SHADING_SCENE_RESOURCE_GROUP = new ShaderResourceGroupDescriptor();
 
 SHADING_SCENE_RESOURCE_GROUP
@@ -14,4 +29,14 @@ SHADING_SCENE_RESOURCE_GROUP
     .addStorageBuffer("meshlet_data", GEOMETRY_ATTRIBUTE_DATA_TYPE)
     .addStorageBuffer("scene_database", WebGPUArray.from(u32))
     .addStorageBuffer("geometries", WebGPUArray.from(GEOMETRY_METADATA_STRUCT))
+    .addStorageBuffer("materials", WebGPUArray.from(MATERIAL_METADATA_STRUCT))
 
+SHADING_SCENE_RESOURCE_GROUP.addVisibility(GPUShaderStage.VERTEX);
+
+/**
+ * The layout the transparent passes build their group-1 bind group against, and every transparent
+ * pipeline declares.
+ *
+ * @type {BindGroupLayoutDescriptor}
+ */
+export const SHADING_SCENE_LAYOUT = SHADING_SCENE_RESOURCE_GROUP.generateBindGroupLayoutDescriptor();

package/src/shade/renderer/rasterize/forward/chunk_forward_shade_standard_fragment_brick4.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"chunk_forward_shade_standard_fragment_brick4.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/rasterize/forward/chunk_forward_shade_standard_fragment_brick4.js"],"names":[],"mappings":"AA6BA,qEA2JC;0BA9JyB,oCAAoC"}
+{"version":3,"file":"chunk_forward_shade_standard_fragment_brick4.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/rasterize/forward/chunk_forward_shade_standard_fragment_brick4.js"],"names":[],"mappings":"AA6BA,qEA4JC;0BA/JyB,oCAAoC"}

package/src/shade/renderer/rasterize/forward/chunk_forward_shade_standard_fragment_brick4.js

@@ -62,7 +62,8 @@ fn forward_shade_standard_fragment(
     let t_diffuse = textureSample(texture_diffuse, texture_diffuse_sampler, vert.uv);
     let t_emissive = textureSample(texture_emissive, texture_emissive_sampler, vert.uv);
 
-    var diffuse = t_diffuse.rgb * vert.color;
+    // the vertex stage hands on the vertex's own colour; the material's is applied here
+    var diffuse = t_diffuse.rgb * vert.color * material_info.albedo_color.rgb;
 
     // undo premultiplication (alpha is premultiplied in the texture)
     diffuse /= max( t_diffuse.a, 1e-7);// prevent division by zero

package/src/shade/renderer/rasterize/forward/chunk_forward_shade_standard_fragment_ibl.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"chunk_forward_shade_standard_fragment_ibl.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/rasterize/forward/chunk_forward_shade_standard_fragment_ibl.js"],"names":[],"mappings":"AA4BA;;;GAGG;AACH,kEAgJC;0BAvJyB,oCAAoC"}
+{"version":3,"file":"chunk_forward_shade_standard_fragment_ibl.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/rasterize/forward/chunk_forward_shade_standard_fragment_ibl.js"],"names":[],"mappings":"AA4BA;;;GAGG;AACH,kEAiJC;0BAxJyB,oCAAoC"}

package/src/shade/renderer/rasterize/forward/chunk_forward_shade_standard_fragment_ibl.js

@@ -65,7 +65,8 @@ fn forward_shade_standard_fragment(
     let t_diffuse = textureSample(texture_diffuse, texture_diffuse_sampler, vert.uv);
     let t_emissive = textureSample(texture_emissive, texture_emissive_sampler, vert.uv);
 
-    var diffuse = t_diffuse.rgb * vert.color;
+    // the vertex stage hands on the vertex's own colour; the material's is applied here
+    var diffuse = t_diffuse.rgb * vert.color * material_info.albedo_color.rgb;
 
     // undo premultiplication (alpha is premultiplied in the texture)
     diffuse /= max( t_diffuse.a, 1e-7);// prevent division by zero

package/src/shade/renderer/rasterize/forward/forward_template_pipeline.d.ts

@@ -0,0 +1,40 @@
+/**
+ * What a transparent pass's fragment stage declares for a template.
+ *
+ * @param {ShadeMaterialTemplate} template
+ * @param {ShaderResourceGroupDescriptor[]} pass_groups groups 2 and up
+ * @returns {ShaderResourceSetDescriptor}
+ */
+export function forward_fragment_resources(template: ShadeMaterialTemplate, pass_groups: ShaderResourceGroupDescriptor[]): ShaderResourceSetDescriptor;
+/**
+ * What the shared forward vertex stage declares: nothing at group 0, which is the template's and
+ * fragment-only.
+ *
+ * @param {ShaderResourceGroupDescriptor[]} pass_groups groups 2 and up, as far as the vertex stage
+ *        reads them
+ * @returns {ShaderResourceSetDescriptor}
+ */
+export function forward_vertex_resources(pass_groups: ShaderResourceGroupDescriptor[]): ShaderResourceSetDescriptor;
+/**
+ * @param {ShadeMaterialTemplate} template
+ * @param {BindGroupLayoutDescriptor[]} pass_layouts groups 2 and up, as the pass binds them
+ * @returns {PipelineLayoutDescriptor}
+ */
+export function forward_pipeline_layout(template: ShadeMaterialTemplate, pass_layouts: BindGroupLayoutDescriptor[]): PipelineLayoutDescriptor;
+/**
+ * @typedef {import("../../../descriptor/binding/BindGroupLayoutDescriptor.js").BindGroupLayoutDescriptor} BindGroupLayoutDescriptor
+ * @typedef {import("../../shader/resource/ShaderResourceGroupDescriptor.js").ShaderResourceGroupDescriptor} ShaderResourceGroupDescriptor
+ * @typedef {import("../../material/template/ShadeMaterialTemplate.js").ShadeMaterialTemplate} ShadeMaterialTemplate
+ */
+/**
+ * The lights, as the passes that shade with them lay the group out and bind it: fragment-only.
+ *
+ * @type {BindGroupLayoutDescriptor}
+ */
+export const FORWARD_LIGHT_LAYOUT: BindGroupLayoutDescriptor;
+export type BindGroupLayoutDescriptor = import("../../../descriptor/binding/BindGroupLayoutDescriptor.js").BindGroupLayoutDescriptor;
+export type ShaderResourceGroupDescriptor = import("../../shader/resource/ShaderResourceGroupDescriptor.js").ShaderResourceGroupDescriptor;
+export type ShadeMaterialTemplate = import("../../material/template/ShadeMaterialTemplate.js").ShadeMaterialTemplate;
+import { ShaderResourceSetDescriptor } from "../../shader/resource/ShaderResourceSetDescriptor.js";
+import { PipelineLayoutDescriptor } from "../../../descriptor/pipeline/PipelineLayoutDescriptor.js";
+//# sourceMappingURL=forward_template_pipeline.d.ts.map

package/src/shade/renderer/rasterize/forward/forward_template_pipeline.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"forward_template_pipeline.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/rasterize/forward/forward_template_pipeline.js"],"names":[],"mappings":"AA+BA;;;;;;GAMG;AACH,qDAJW,qBAAqB,eACrB,6BAA6B,EAAE,GAC7B,2BAA2B,CAavC;AAED;;;;;;;GAOG;AACH,sDAJW,6BAA6B,EAAE,GAE7B,2BAA2B,CAavC;AAED;;;;GAIG;AACH,kDAJW,qBAAqB,gBACrB,yBAAyB,EAAE,GACzB,wBAAwB,CAUpC;AA9ED;;;;GAIG;AAaH;;;;GAIG;AACH,mCAFU,yBAAyB,CAEyF;wCArB/G,OAAO,0DAA0D,EAAE,yBAAyB;4CAC5F,OAAO,wDAAwD,EAAE,6BAA6B;oCAC9F,OAAO,kDAAkD,EAAE,qBAAqB;4CANjD,sDAAsD;yCAFzD,0DAA0D"}