@woosh/meep-engine 3.28.0 → 3.29.0

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

@@ -1,90 +1,101 @@
-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 { 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";
-import { CodeChunk } from "../../../shader/compiler/CodeChunk.js";
-import { COMPUTE_SHADER_DEFAULT_GROUP_SIZE } from "../../../shader/COMPUTE_SHADER_DEFAULT_GROUP_SIZE.js";
-import { ComputeShader } from "../../../shader/ComputeShader.js";
-import { ShaderResourceSetDescriptor } from "../../../shader/resource/ShaderResourceSetDescriptor.js";
-import { GEOMETRY_METADATA_STRUCT } from "../../../geometry/GEOMETRY_METADATA_STRUCT.js";
-import { MESHLET_DEFINITION_STRUCT } from "../../../geometry/meshlet/encoding/MESHLET_DEFINITION_STRUCT.js";
-
-
-const resources = new ShaderResourceSetDescriptor();
-
-resources.createGroup()
-    .addStorageBuffer("prefix_sums", COLLECTION_U32_STRUCT_READ)
-
-resources.createGroup()
-    .addStorageBuffer("scene_database", WebGPUArray.from(u32))
-    .addStorageBuffer("geometries", WebGPUArray.from(GEOMETRY_METADATA_STRUCT))
-
-resources.createGroup()
-    .addStorageBuffer("mesh_index_list", COLLECTION_U32_STRUCT_READ)
-    .addStorageBuffer("output", MESHLET_COLLECTION_STRUCT, true)
-;
-
-const body = CodeChunk.from(
-// language=WGSL
-    `
-@compute @workgroup_size(${COMPUTE_SHADER_DEFAULT_GROUP_SIZE})
-fn unpack(@builtin(global_invocation_id) global_id : vec3<u32>){
-
-    let input_index = global_id.x;
-    
-    let input_mesh_count = prefix_sums.count;
-    
-    let mesh_search_index = lower_bound_branchless(&prefix_sums.elements, input_index + 1u, input_mesh_count);
-    
-    if(mesh_search_index >= input_mesh_count){
-        // overflow, we're done
-        return;
-    }
-    
-    var prefix_sum_prev = 0u;
-    
-    if (mesh_search_index > 0u) {
-        // NOTE: we can't use select here, as it would cause severe overflow
-        prefix_sum_prev = prefix_sums.elements[mesh_search_index - 1u];
-    }
-    
-    let local_offset = input_index - prefix_sum_prev;
-    
-    let mesh_index = mesh_index_list.elements[mesh_search_index];
-
-    let mesh = scene_read_mesh(&scene_database, mesh_index);
-
-    let geometry_index = mesh.geometry;
-
-    let geometry = geometries[geometry_index];
-    
-    let meshlet_count = geometry.meshlets_count;
-    
-    let lookup_offset = geometry.meshlets_address;
-    
-    // TODO the limit is not necessary, but it's a good sanity check
-    let meshlet_index = lookup_offset + min(local_offset, meshlet_count - 1);
-        
-    // we allow "append" operation by involving existing count
-    let output_index = input_index + output.count;
-    
-    output.elements[output_index] = ${MESHLET_DEFINITION_STRUCT.wgsl_ref}(
-        meshlet_index,
-        mesh_index,
-    );
-    
-
-}`, [
-        MESHLET_DEFINITION_STRUCT.declaration_chunk,
-        chunk_lower_bound_branchless,
-        chunk_scene_database_access,
-    ]);
-
-
-export const shader_mesh_expand_to_meshlets_prefix = ComputeShader.from({
-    label: "Mesh expand to meshlets via prefix",
-    body,
-    resources,
+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 { 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";
+import { CodeChunk } from "../../../shader/compiler/CodeChunk.js";
+import { COMPUTE_SHADER_DEFAULT_GROUP_SIZE } from "../../../shader/COMPUTE_SHADER_DEFAULT_GROUP_SIZE.js";
+import { ComputeShader } from "../../../shader/ComputeShader.js";
+import { ShaderResourceSetDescriptor } from "../../../shader/resource/ShaderResourceSetDescriptor.js";
+import { GEOMETRY_METADATA_STRUCT } from "../../../geometry/GEOMETRY_METADATA_STRUCT.js";
+import { MESHLET_DEFINITION_STRUCT } from "../../../geometry/meshlet/encoding/MESHLET_DEFINITION_STRUCT.js";
+
+
+const resources = new ShaderResourceSetDescriptor();
+
+resources.createGroup()
+    .addStorageBuffer("prefix_sums", COLLECTION_U32_STRUCT_READ)
+
+resources.createGroup()
+    .addStorageBuffer("scene_database", WebGPUArray.from(u32))
+    .addStorageBuffer("geometries", WebGPUArray.from(GEOMETRY_METADATA_STRUCT))
+
+resources.createGroup()
+    .addStorageBuffer("mesh_index_list", COLLECTION_U32_STRUCT_READ)
+    .addStorageBuffer("output", MESHLET_COLLECTION_STRUCT, true)
+;
+
+const body = CodeChunk.from(
+// language=WGSL
+    `
+@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,
+){
+
+    /*
+     Not global_invocation_id.x: this pass is dispatched from a folded command, because a million
+     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_mesh_count = prefix_sums.count;
+    
+    let mesh_search_index = lower_bound_branchless(&prefix_sums.elements, input_index + 1u, input_mesh_count);
+    
+    if(mesh_search_index >= input_mesh_count){
+        // overflow, we're done
+        return;
+    }
+    
+    var prefix_sum_prev = 0u;
+    
+    if (mesh_search_index > 0u) {
+        // NOTE: we can't use select here, as it would cause severe overflow
+        prefix_sum_prev = prefix_sums.elements[mesh_search_index - 1u];
+    }
+    
+    let local_offset = input_index - prefix_sum_prev;
+    
+    let mesh_index = mesh_index_list.elements[mesh_search_index];
+
+    let mesh = scene_read_mesh(&scene_database, mesh_index);
+
+    let geometry_index = mesh.geometry;
+
+    let geometry = geometries[geometry_index];
+    
+    let meshlet_count = geometry.meshlets_count;
+    
+    let lookup_offset = geometry.meshlets_address;
+    
+    // TODO the limit is not necessary, but it's a good sanity check
+    let meshlet_index = lookup_offset + min(local_offset, meshlet_count - 1);
+        
+    // we allow "append" operation by involving existing count
+    let output_index = input_index + output.count;
+    
+    output.elements[output_index] = ${MESHLET_DEFINITION_STRUCT.wgsl_ref}(
+        meshlet_index,
+        mesh_index,
+    );
+    
+
+}`, [
+        MESHLET_DEFINITION_STRUCT.declaration_chunk,
+        chunk_grid3d_to_index,
+        chunk_lower_bound_branchless,
+        chunk_scene_database_access,
+    ]);
+
+
+export const shader_mesh_expand_to_meshlets_prefix = ComputeShader.from({
+    label: "Mesh expand to meshlets via prefix",
+    body,
+    resources,
 })

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":"AAmCA,yDAIG;8BAlC2B,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":"AA2DA,yDAIG;8BAvD2B,kCAAkC"}

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

@@ -1,3 +1,6 @@
+import {
+    WEBGPU_GUARANTEED_MAX_COMPUTE_WORKGROUPS_PER_DIMENSION
+} from "../../../gpu_primitive/indirect/WEBGPU_GUARANTEED_MAX_COMPUTE_WORKGROUPS_PER_DIMENSION.js";
 import { STRUCT_DISPATCH_COMPUTE_COMMAND } from "../../../gpu_primitive/indirect/STRUCT_DISPATCH_COMPUTE_COMMAND.js";
 import { COLLECTION_U32_STRUCT_READ } from "../../../gpu_primitive/COLLECTION_U32_STRUCT_READ.js";
 import { chunk_divide_round_up_u32 } from "../../../shader/chunk/math/chunk_divide_round_up_u32.js";
@@ -21,12 +24,33 @@ fn main(){
     let total_count = select( 0u, prefix_sums.elements[ prefix_sums.count - 1u ], prefix_sums.count > 0u );
 
     const TARGET_GROUP_SIZE = ${COMPUTE_SHADER_DEFAULT_GROUP_SIZE};
-    
-    // TODO limit to something sensible, or better yet - use actual device limits
-    command.workgroupCountX = divide_round_up_u32(total_count, TARGET_GROUP_SIZE);
-    command.workgroupCountY = 1u;
-    command.workgroupCountZ = 1u;
-    
+
+    /*
+     Folded over three dimensions rather than written into x alone, because a command past
+     maxComputeWorkgroupsPerDimension does not run at all — not a short pass, no pass, and no error
+     either. The scan's total is meshlets, and a scene of a million ordinary mesh rows is past the
+     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
+     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.
+    */
+    const LIMIT = ${WEBGPU_GUARANTEED_MAX_COMPUTE_WORKGROUPS_PER_DIMENSION}u;
+
+    let desired_groups = divide_round_up_u32(total_count, TARGET_GROUP_SIZE);
+
+    let rows = divide_round_up_u32(desired_groups, LIMIT);
+    let slices = divide_round_up_u32(rows, LIMIT);
+
+    command.workgroupCountX = min( desired_groups, LIMIT );
+    command.workgroupCountY = min( max( rows, 1u ), LIMIT );
+    command.workgroupCountZ = max( slices, 1u );
+
 }
     `, [
         chunk_divide_round_up_u32
@@ -37,4 +61,4 @@ export const shader_prefix_sum_to_command = ComputeShader.from({
     label: "Prefix sum to command",
     resources,
     body,
-});
+});

package/src/shade/renderer/rasterize/expand/pass0/pass_mesh_cull.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"pass_mesh_cull.d.ts","sourceRoot":"","sources":["../../../../../../../src/shade/renderer/rasterize/expand/pass0/pass_mesh_cull.js"],"names":[],"mappings":"AAyBA;;;;;GAKG;AACH,qCAJW,qBAAqB,0EAmD/B;AAtED;IACI,yBAAqB;IACrB,4BAAwB;IAExB,YAAQ;IAER;;OAEG;IACH,yBAAQ;IAER,qBAAiB;IACjB,gBAAY;IAEZ,gBAAY;CACf"}
+{"version":3,"file":"pass_mesh_cull.d.ts","sourceRoot":"","sources":["../../../../../../../src/shade/renderer/rasterize/expand/pass0/pass_mesh_cull.js"],"names":[],"mappings":"AAwBA;;;;;GAKG;AACH,qCAJW,qBAAqB,0EA8C/B;AAjED;IACI,yBAAqB;IACrB,4BAAwB;IAExB,YAAQ;IAER;;OAEG;IACH,yBAAQ;IAER,qBAAiB;IACjB,gBAAY;IAEZ,gBAAY;CACf"}

package/src/shade/renderer/rasterize/expand/pass0/pass_mesh_cull.js

@@ -1,7 +1,6 @@
 import { assert } from "../../../../../core/assert.js";
 import { u32 } from "../../../shader/type/WEBGPU_WGSL_PRIMITIVE_TYPES.js";
 import { shader_hzb_mesh_filter_2way } from "../../cull/hzb/shader_hzb_mesh_filter_2way.js"
-import { U32_CONTAINER_STRUCT } from "../../../shader/type/U32_CONTAINER_STRUCT.js";
 import { COMPUTE_SHADER_DEFAULT_GROUP_SIZE } from "../../../shader/COMPUTE_SHADER_DEFAULT_GROUP_SIZE.js";
 import { build_indirect_compute_dispatch_command } from "../../../gpu_primitive/indirect/IndirectComputeCommandBuildPass.js";
 
@@ -57,15 +56,10 @@ export function pass_mesh_cull(data, resources, context) {
         0
     );
 
-    // TODO(chrome-153): transient uniform, not an immediate value — see the shader.
-    const u_bucket_index = encoder.allocateTransientValueBuffer(U32_CONTAINER_STRUCT, {
-        value: bucket_id,
-    });
-
     shader_hzb_mesh_filter_2way.dispatchIndirect({
         encoder,
         bindings: {
-            bucket_index: u_bucket_index,
+            bucket_index: { value: bucket_id },
             camera_previous: view_ctx.gpu_previous_camera_state.buffer,
             bucket_data,
             scene_database: gpu_scene.scene_database_buffer,

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":"AAqJA,uDAIG;8BA/H2B,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":"AA6JA,uDAIG;8BAtI2B,kCAAkC"}

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

@@ -1,154 +1,162 @@
-import { u32 } from "../../../shader/type/WEBGPU_WGSL_PRIMITIVE_TYPES.js";
-import { WebGPUArray } from "../../../shader/type/WebGPUArray.js";
-import { CAMERA_UNIFORM_STRUCT } from "../../../camera/CAMERA_UNIFORM_STRUCT.js";
-import { MESHLET_COLLECTION_STRUCT } from "../../MESHLET_COLLECTION_STRUCT.js";
-import { MESHLET_COLLECTION_WRITE_STRUCT } from "../../MESHLET_COLLECTION_WRITE_STRUCT.js";
-import { chunk_scene_database_access } from "../../../scene/chunk_scene_database_access.js";
-import {
-    chunk_query_depth_from_screen_space_bb
-} from "../../../shader/chunk/depth/chunk_query_depth_from_screen_space_bb.js";
-import { AABB3_STRUCT } from "../../../shader/chunk/geometry/aabb3/AABB3_STRUCT.js";
-import {
-    chunk_aabb3_intersects_frustum
-} from "../../../shader/chunk/geometry/aabb3/chunk_aabb3_intersects_frustum.js";
-import {
-    chunk_aabb3_intersects_frustum_degree
-} from "../../../shader/chunk/geometry/aabb3/chunk_aabb3_intersects_frustum_degree.js";
-import { chunk_aabb3_project } from "../../../shader/chunk/geometry/aabb3/chunk_aabb3_project.js";
-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_aabb2_clip_overlaps_texel_centers
-} from "../../../shader/chunk/texture/chunk_aabb2_clip_overlaps_texel_centers.js";
-import { CodeChunk } from "../../../shader/compiler/CodeChunk.js";
-import { COMPUTE_SHADER_DEFAULT_GROUP_SIZE } from "../../../shader/COMPUTE_SHADER_DEFAULT_GROUP_SIZE.js";
-import { ComputeShader } from "../../../shader/ComputeShader.js";
-import { ShaderResourceSetDescriptor } from "../../../shader/resource/ShaderResourceSetDescriptor.js";
-import { VIEW_UNIFORM_STRUCT } from "../../../view/VIEW_UNIFORM_STRUCT.js";
-import { MESHLET_METADATA_STRUCT } from "../../../geometry/meshlet/encoding/MESHLET_METADATA_STRUCT.js";
-
-const resources = new ShaderResourceSetDescriptor();
-resources.createGroup()
-    .addUniform("camera_current", CAMERA_UNIFORM_STRUCT)
-    .addUniform("camera_previous", CAMERA_UNIFORM_STRUCT)
-    .addUniform("view", VIEW_UNIFORM_STRUCT)
-    .addStorageBuffer("scene_database", WebGPUArray.from(u32))
-    .addStorageBuffer("meshlet_metadata", WebGPUArray.from(MESHLET_METADATA_STRUCT))
-    .addTexture("texture_hzb") // HZB of previous frame
-;
-
-resources.createGroup()
-    .addStorageBuffer("input", MESHLET_COLLECTION_STRUCT)
-    .addStorageBuffer("output_positive", MESHLET_COLLECTION_WRITE_STRUCT, true)
-    .addStorageBuffer("output_maybe", MESHLET_COLLECTION_WRITE_STRUCT, true)
-;
-
-
-const body = CodeChunk.from(
-// language=WGSL
-    `
-@compute @workgroup_size(${COMPUTE_SHADER_DEFAULT_GROUP_SIZE})
-fn unpack(@builtin(global_invocation_id) global_id : vec3<u32>){
-    
-    let input_index = global_id.x;
-    
-    if(input_index >= input.count){
-        return;
-    }
-    
-    let meshlet = input.elements[input_index];
-
-    let mesh = scene_read_mesh(&scene_database, meshlet.mesh);
-    let node = scene_read_node(&scene_database, mesh.node);
-
-    let bounds_array = meshlet_metadata[ meshlet.index ].bounds_box;
-
-
-    // check if cluster is too small to be visible
-
-    let meshlet_bounds = array_to_aabb3(bounds_array);
-
-    let world_bounds = aabb3_project(meshlet_bounds, node.global);
-    
-    /*
-        TODO do frustum cull using a sphere representation, which is faster
-    */
-    
-    if( !aabb3_intersects_frustum(world_bounds, camera_current.frustum) ){
-        // outside of view frustum
-        return ;
-    }
-    
-    {
-        // small primitive culling
-        var current_view_bounds:${AABB3_STRUCT.wgsl_ref};
-    
-        // TODO we can perform frustum culling using view bounds instead of actual frustum planes
-        
-        let valid = aabb3_project_perspective(&current_view_bounds, world_bounds, camera_current.view_projection_matrix);
-        
-        let resolution = vec2(view.width, view.height);
-        
-        let would_draw = aabb2_clip_overlaps_texel_centers(
-            current_view_bounds.min.xy,
-            current_view_bounds.max.xy,
-            resolution
-        );
-        
-        if(valid && !would_draw){
-            // cull, doesn't result in any draw
-            return;
-        }
-    }
-    
-    var previous_view_bounds:${AABB3_STRUCT.wgsl_ref};
-    
-    let valid_bounds = aabb3_project_perspective(&previous_view_bounds, world_bounds, camera_previous.view_projection_matrix);
-    
-    if(
-        !valid_bounds || query_depth_from_screen_space_bb(previous_view_bounds, texture_hzb) >= 0.0
-    ){
-    
-        let write_offset = atomicAdd(&output_positive.count, 1u);
-        
-        if(write_offset < arrayLength(&output_positive.elements)){
-            output_positive.elements[ write_offset ] = meshlet;
-        }
-    
-    }else{
-    
-        let write_offset = atomicAdd(&output_maybe.count, 1u);
-        
-        if(write_offset < arrayLength(&output_maybe.elements)){
-            output_maybe.elements[ write_offset ] = meshlet;
-        }
-    
-    }
-
-}`, [
-
-        AABB3_STRUCT.declaration_chunk,
-
-        chunk_array_to_aabb3,
-
-        chunk_aabb3_intersects_frustum_degree,
-
-        chunk_aabb3_intersects_frustum,
-
-        chunk_aabb3_project,
-
-        chunk_aabb3_project_perspective,
-        chunk_query_depth_from_screen_space_bb,
-
-        chunk_aabb2_clip_overlaps_texel_centers,
-
-        chunk_scene_database_access,
-    ]);
-
-export const shader_meshlet_filter_2way = ComputeShader.from({
-    label: "Filter Meshlets 2way",
-    resources,
-    body,
+import { u32 } from "../../../shader/type/WEBGPU_WGSL_PRIMITIVE_TYPES.js";
+import { WebGPUArray } from "../../../shader/type/WebGPUArray.js";
+import { CAMERA_UNIFORM_STRUCT } from "../../../camera/CAMERA_UNIFORM_STRUCT.js";
+import { MESHLET_COLLECTION_STRUCT } from "../../MESHLET_COLLECTION_STRUCT.js";
+import { MESHLET_COLLECTION_WRITE_STRUCT } from "../../MESHLET_COLLECTION_WRITE_STRUCT.js";
+import { chunk_scene_database_access } from "../../../scene/chunk_scene_database_access.js";
+import {
+    chunk_query_depth_from_screen_space_bb
+} from "../../../shader/chunk/depth/chunk_query_depth_from_screen_space_bb.js";
+import { AABB3_STRUCT } from "../../../shader/chunk/geometry/aabb3/AABB3_STRUCT.js";
+import {
+    chunk_aabb3_intersects_frustum
+} from "../../../shader/chunk/geometry/aabb3/chunk_aabb3_intersects_frustum.js";
+import {
+    chunk_aabb3_intersects_frustum_degree
+} from "../../../shader/chunk/geometry/aabb3/chunk_aabb3_intersects_frustum_degree.js";
+import { chunk_aabb3_project } from "../../../shader/chunk/geometry/aabb3/chunk_aabb3_project.js";
+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_aabb2_clip_overlaps_texel_centers
+} from "../../../shader/chunk/texture/chunk_aabb2_clip_overlaps_texel_centers.js";
+import { CodeChunk } from "../../../shader/compiler/CodeChunk.js";
+import { COMPUTE_SHADER_DEFAULT_GROUP_SIZE } from "../../../shader/COMPUTE_SHADER_DEFAULT_GROUP_SIZE.js";
+import { ComputeShader } from "../../../shader/ComputeShader.js";
+import { ShaderResourceSetDescriptor } from "../../../shader/resource/ShaderResourceSetDescriptor.js";
+import { VIEW_UNIFORM_STRUCT } from "../../../view/VIEW_UNIFORM_STRUCT.js";
+import { MESHLET_METADATA_STRUCT } from "../../../geometry/meshlet/encoding/MESHLET_METADATA_STRUCT.js";
+
+const resources = new ShaderResourceSetDescriptor();
+resources.createGroup()
+    .addUniform("camera_current", CAMERA_UNIFORM_STRUCT)
+    .addUniform("camera_previous", CAMERA_UNIFORM_STRUCT)
+    .addUniform("view", VIEW_UNIFORM_STRUCT)
+    .addStorageBuffer("scene_database", WebGPUArray.from(u32))
+    .addStorageBuffer("meshlet_metadata", WebGPUArray.from(MESHLET_METADATA_STRUCT))
+    .addTexture("texture_hzb") // HZB of previous frame
+;
+
+resources.createGroup()
+    .addStorageBuffer("input", MESHLET_COLLECTION_STRUCT)
+    .addStorageBuffer("output_positive", MESHLET_COLLECTION_WRITE_STRUCT, true)
+    .addStorageBuffer("output_maybe", MESHLET_COLLECTION_WRITE_STRUCT, true)
+;
+
+
+const body = CodeChunk.from(
+// language=WGSL
+    `
+@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) * ${COMPUTE_SHADER_DEFAULT_GROUP_SIZE}u + local_index;
+
+    if(input_index >= input.count){
+        return;
+    }
+    
+    let meshlet = input.elements[input_index];
+
+    let mesh = scene_read_mesh(&scene_database, meshlet.mesh);
+    let node = scene_read_node(&scene_database, mesh.node);
+
+    let bounds_array = meshlet_metadata[ meshlet.index ].bounds_box;
+
+
+    // check if cluster is too small to be visible
+
+    let meshlet_bounds = array_to_aabb3(bounds_array);
+
+    let world_bounds = aabb3_project(meshlet_bounds, node.global);
+    
+    /*
+        TODO do frustum cull using a sphere representation, which is faster
+    */
+    
+    if( !aabb3_intersects_frustum(world_bounds, camera_current.frustum) ){
+        // outside of view frustum
+        return ;
+    }
+    
+    {
+        // small primitive culling
+        var current_view_bounds:${AABB3_STRUCT.wgsl_ref};
+    
+        // TODO we can perform frustum culling using view bounds instead of actual frustum planes
+        
+        let valid = aabb3_project_perspective(&current_view_bounds, world_bounds, camera_current.view_projection_matrix);
+        
+        let resolution = vec2(view.width, view.height);
+        
+        let would_draw = aabb2_clip_overlaps_texel_centers(
+            current_view_bounds.min.xy,
+            current_view_bounds.max.xy,
+            resolution
+        );
+        
+        if(valid && !would_draw){
+            // cull, doesn't result in any draw
+            return;
+        }
+    }
+    
+    var previous_view_bounds:${AABB3_STRUCT.wgsl_ref};
+    
+    let valid_bounds = aabb3_project_perspective(&previous_view_bounds, world_bounds, camera_previous.view_projection_matrix);
+    
+    if(
+        !valid_bounds || query_depth_from_screen_space_bb(previous_view_bounds, texture_hzb) >= 0.0
+    ){
+    
+        let write_offset = atomicAdd(&output_positive.count, 1u);
+        
+        if(write_offset < arrayLength(&output_positive.elements)){
+            output_positive.elements[ write_offset ] = meshlet;
+        }
+    
+    }else{
+    
+        let write_offset = atomicAdd(&output_maybe.count, 1u);
+        
+        if(write_offset < arrayLength(&output_maybe.elements)){
+            output_maybe.elements[ write_offset ] = meshlet;
+        }
+    
+    }
+
+}`, [
+
+        AABB3_STRUCT.declaration_chunk,
+
+        chunk_grid3d_to_index,
+
+        chunk_array_to_aabb3,
+
+        chunk_aabb3_intersects_frustum_degree,
+
+        chunk_aabb3_intersects_frustum,
+
+        chunk_aabb3_project,
+
+        chunk_aabb3_project_perspective,
+        chunk_query_depth_from_screen_space_bb,
+
+        chunk_aabb2_clip_overlaps_texel_centers,
+
+        chunk_scene_database_access,
+    ]);
+
+export const shader_meshlet_filter_2way = ComputeShader.from({
+    label: "Filter Meshlets 2way",
+    resources,
+    body,
 });

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

@@ -1 +1 @@
-{"version":3,"file":"graph_extract_meshes_from_bucket.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/rasterize/fast/graph_extract_meshes_from_bucket.js"],"names":[],"mappings":"AAUA;;;;;;;;;;GAUG;AACH;IAN+B,KAAK;IACT,OAAO,EAAvB,MAAM;IACU,SAAS,EAAzB,MAAM;IACuB,MAAM;IACjC,MAAM,CAgElB"}
+{"version":3,"file":"graph_extract_meshes_from_bucket.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/rasterize/fast/graph_extract_meshes_from_bucket.js"],"names":[],"mappings":"AAUA;;;;;;;;;;GAUG;AACH;IAN+B,KAAK;IACT,OAAO,EAAvB,MAAM;IACU,SAAS,EAAzB,MAAM;IACuB,MAAM;IACjC,MAAM,CA2DlB"}

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

@@ -6,7 +6,7 @@ import {
     INDIRECT_COMPUTE_COMMAND_DESCRIPTOR
 } from "../../gpu_primitive/indirect/IndirectComputeCommandBuildPass.js";
 import { RasterizationCollectionType } from "../expand/RasterizationCollectionType.js";
-import { PUSH_CONSTANTS, shader_read_out_bucket } from "../bucket/shader_read_out_bucket.js";
+import { shader_read_out_bucket } from "../bucket/shader_read_out_bucket.js";
 
 /**
  * Extract a single bucket's mesh indices out of a bucket-sorted mesh buffer.
@@ -51,16 +51,11 @@ export function graph_extract_meshes_from_bucket({
             0
         );
 
-        // TODO(chrome-153): transient uniform, not an immediate value — see the shader.
-        const u_settings = encoder.allocateTransientValueBuffer(PUSH_CONSTANTS, {
-            bucket_id,
-        });
-
         shader_read_out_bucket.dispatchIndirect({
             encoder,
             command,
             bindings: {
-                settings: u_settings,
+                settings: { bucket_id },
                 input: buckets,
                 output,
             }