hyperverse 0.0.1__py3-none-any.whl

hyperdrone/_vendor/rl-tools/include/rl_tools/rendering/raytracing/backends/metal/operations_cpu.h

@@ -0,0 +1,1140 @@
+#include "../../../../version.h"
+#include "../../../../rl_tools.h"
+#if (defined(RL_TOOLS_DISABLE_INCLUDE_GUARDS) || !defined(RL_TOOLS_RENDERING_RAYTRACING_BACKENDS_METAL_OPERATIONS_CPU_H)) && (RL_TOOLS_USE_THIS_VERSION == 1)
+#pragma once
+#define RL_TOOLS_RENDERING_RAYTRACING_BACKENDS_METAL_OPERATIONS_CPU_H
+
+#include "../../renderer.h"
+#include "../../operations_cpu_common.h"
+#include "../specialization.h"
+#include "context.h"
+#include "device_source.h"
+
+#include <vector>
+#include <cstring>
+#include <string>
+
+RL_TOOLS_NAMESPACE_WRAPPER_START
+namespace rl_tools {
+    namespace rendering::raytracing::backends {
+        template <typename SPEC>
+        struct RendererState<rendering::raytracing::backends::Metal, SPEC>: metal::Context {};
+
+        template <typename SPEC>
+        struct LibraryState<rendering::raytracing::backends::Metal, SPEC> {};
+
+        template <typename SPEC>
+        struct SceneState<rendering::raytracing::backends::Metal, SPEC> {};
+    }
+
+    namespace rendering::raytracing::backends::metal{
+        template <typename SPEC>
+        Context& context(rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+            return *renderer.backend;
+        }
+
+        inline NS::SharedPtr<MTL::Texture> make_texture(MTL::Device* device, const uint8_t* pixels, int width, int height, bool srgb){
+            auto descriptor = MTL::TextureDescriptor::texture2DDescriptor(srgb ? MTL::PixelFormatRGBA8Unorm_sRGB : MTL::PixelFormatRGBA8Unorm, width, height, false);
+            descriptor->setUsage(MTL::TextureUsageShaderRead);
+            descriptor->setStorageMode(MTL::StorageModeShared);
+            auto texture = NS::TransferPtr(device->newTexture(descriptor));
+            texture->replaceRegion(MTL::Region::Make2D(0, 0, width, height), 0, pixels, (NS::UInteger)width * 4);
+            return texture;
+        }
+
+        inline void wait_in_flight(Context& ctx){
+            if(ctx.in_flight.get() != nullptr){
+                ctx.in_flight->waitUntilCompleted();
+                ctx.in_flight.reset();
+            }
+            if(ctx.in_flight_collision.get() != nullptr){
+                ctx.in_flight_collision->waitUntilCompleted();
+                ctx.in_flight_collision.reset();
+            }
+            if(ctx.in_flight_update.get() != nullptr){
+                ctx.in_flight_update->waitUntilCompleted();
+                ctx.in_flight_update.reset();
+            }
+        }
+
+        // secondary_output: the flow passes' status buffer next to their flow OUTPUT
+        template <typename SPEC>
+        void encode_fullscreen_pass(Context& ctx, MTL::CommandBuffer* command_buffer, MTL::ComputePipelineState* pipeline, MTL::Buffer* output, float shutter_t = 0.f, MTL::Buffer* secondary_output = nullptr){
+            MTL::ComputeCommandEncoder* encoder = command_buffer->computeCommandEncoder();
+            encoder->setComputePipelineState(pipeline);
+            encoder->setBuffer(ctx.launch_params.get(), 0, bindings::LAUNCH_PARAMS);
+            encoder->setBuffer(ctx.cameras.get(), 0, bindings::CAMERAS_CLOSE);
+            encoder->setBuffer(ctx.cameras_open.get() != nullptr ? ctx.cameras_open.get() : ctx.cameras.get(), 0, bindings::CAMERAS_OPEN);
+            encoder->setBuffer(ctx.cameras_previous.get() != nullptr ? ctx.cameras_previous.get() : ctx.cameras.get(), 0, bindings::CAMERAS_PREVIOUS);
+            encoder->setBuffer(output, 0, bindings::OUTPUT);
+            if(secondary_output != nullptr){
+                encoder->setBuffer(secondary_output, 0, bindings::FLOW_STATUS_OUTPUT);
+            }
+            if constexpr (SPEC::ENABLE_DYNAMIC_MOTION_BLUR){
+                if(ctx.rgb_accumulator.get() != nullptr){
+                    encoder->setBuffer(ctx.rgb_accumulator.get(), 0, bindings::RGB_ACCUMULATOR);
+                }
+                if(ctx.depth_accumulator.get() != nullptr){
+                    encoder->setBuffer(ctx.depth_accumulator.get(), 0, bindings::DEPTH_ACCUMULATOR);
+                }
+                encoder->setBytes(&shutter_t, sizeof(float), bindings::SHUTTER);
+            }
+            encoder->setBuffer(ctx.mesh_records.get(), 0, bindings::MESH_RECORDS);
+            if(ctx.scene_lights.get() != nullptr){
+                encoder->setBuffer(ctx.scene_lights.get(), 0, bindings::SCENE_LIGHTS);
+            }
+            encoder->setAccelerationStructure(ctx.acceleration_structure.get(), bindings::ACCELERATION_STRUCTURE);
+            encoder->setBuffer(ctx.instance_record_base.get(), 0, bindings::INSTANCE_RECORD_BASE);
+            encoder->setBuffer(ctx.instance_data.get(), 0, bindings::INSTANCE_DATA);
+            encoder->setBuffer(ctx.overlay_structures.get(), 0, bindings::OVERLAY_STRUCTURES);
+            encoder->setBuffer(ctx.overlay_structures_previous.get() != nullptr ? ctx.overlay_structures_previous.get() : ctx.overlay_structures.get(), 0, bindings::OVERLAY_STRUCTURES_PREVIOUS);
+            encoder->setBuffer(ctx.overlay_attachments.get(), 0, bindings::OVERLAY_ATTACHMENTS);
+            encoder->setBuffer(ctx.instance_classes.get(), 0, bindings::INSTANCE_CLASSES);
+            if(ctx.observation.get() != nullptr){
+                encoder->setBuffer(ctx.observation.get(), 0, bindings::OBSERVATION);
+            }
+            if(ctx.flow_deltas.get() != nullptr){
+                encoder->setBuffer(ctx.flow_deltas.get(), 0, bindings::FLOW_DELTAS);
+            }
+            for(auto& object_acceleration_structure : ctx.object_acceleration_structures){
+                encoder->useResource(object_acceleration_structure.get(), MTL::ResourceUsageRead);
+            }
+            for(auto& overlay_acceleration_structure : ctx.overlay_acceleration_structures){
+                encoder->useResource(overlay_acceleration_structure.get(), MTL::ResourceUsageRead);
+            }
+            for(auto& overlay_acceleration_structure : ctx.overlay_acceleration_structures_previous){
+                encoder->useResource(overlay_acceleration_structure.get(), MTL::ResourceUsageRead);
+            }
+            for(auto& buffer : ctx.mesh_buffers){
+                encoder->useResource(buffer.get(), MTL::ResourceUsageRead);
+            }
+            for(auto& texture : ctx.mesh_textures){
+                encoder->useResource(texture.get(), MTL::ResourceUsageRead);
+            }
+            if(ctx.dummy_texture.get() != nullptr){
+                encoder->useResource(ctx.dummy_texture.get(), MTL::ResourceUsageRead);
+            }
+            encoder->dispatchThreads(MTL::Size::Make(SPEC::FB_WIDTH, SPEC::FB_HEIGHT, 1), MTL::Size::Make(8, 8, 1));
+            encoder->endEncoding();
+        }
+
+        template <typename SPEC>
+        void encode_resolve_pass(Context& ctx, MTL::CommandBuffer* command_buffer){
+            MTL::ComputeCommandEncoder* encoder = command_buffer->computeCommandEncoder();
+            encoder->setComputePipelineState(ctx.resolve_pipeline.get());
+            encoder->setBuffer(ctx.launch_params.get(), 0, bindings::LAUNCH_PARAMS);
+            if constexpr (SPEC::HAS_RGB){
+                encoder->setBuffer(ctx.frame_buffer.get(), 0, bindings::OUTPUT);
+                encoder->setBuffer(ctx.rgb_accumulator.get(), 0, bindings::RGB_ACCUMULATOR);
+            }
+            if constexpr (SPEC::HAS_OBSERVATION){
+                encoder->setBuffer(ctx.observation.get(), 0, bindings::OBSERVATION);
+            }
+            if constexpr (SPEC::HAS_DEPTH){
+                encoder->setBuffer(ctx.depth_accumulator.get(), 0, bindings::DEPTH_ACCUMULATOR);
+                encoder->setBuffer(ctx.depth_buffer.get(), 0, bindings::DEPTH_OUTPUT);
+            }
+            encoder->dispatchThreads(MTL::Size::Make(SPEC::FB_WIDTH, SPEC::FB_HEIGHT, 1), MTL::Size::Make(8, 8, 1));
+            encoder->endEncoding();
+        }
+
+        template <typename SPEC>
+        void encode_collision_pass(Context& ctx, MTL::CommandBuffer* command_buffer){
+            MTL::ComputeCommandEncoder* encoder = command_buffer->computeCommandEncoder();
+            encoder->setComputePipelineState(ctx.collision_pipeline.get());
+            encoder->setBuffer(ctx.launch_params.get(), 0, bindings::LAUNCH_PARAMS);
+            encoder->setBuffer(ctx.cameras.get(), 0, bindings::CAMERAS_CLOSE);
+            encoder->setBuffer(ctx.probe_directions.get(), 0, bindings::PROBE_DIRECTIONS);
+            encoder->setBuffer(ctx.collision_results.get(), 0, bindings::COLLISION_RESULTS);
+            encoder->setAccelerationStructure(ctx.acceleration_structure.get(), bindings::ACCELERATION_STRUCTURE);
+            encoder->setBuffer(ctx.overlay_structures.get(), 0, bindings::OVERLAY_STRUCTURES);
+            encoder->setBuffer(ctx.overlay_attachments.get(), 0, bindings::OVERLAY_ATTACHMENTS);
+            for(auto& object_acceleration_structure : ctx.object_acceleration_structures){
+                encoder->useResource(object_acceleration_structure.get(), MTL::ResourceUsageRead);
+            }
+            for(auto& overlay_acceleration_structure : ctx.overlay_acceleration_structures){
+                encoder->useResource(overlay_acceleration_structure.get(), MTL::ResourceUsageRead);
+            }
+            encoder->dispatchThreads(MTL::Size::Make(SPEC::NUM_CAMERAS, SPEC::NUM_PROBES, 1), MTL::Size::Make(8, 8, 1));
+            encoder->endEncoding();
+        }
+    }
+
+    template <typename DEVICE, typename SPEC>
+    void malloc(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+        namespace metal = rendering::raytracing::backends::metal;
+        static_assert(utils::typing::is_same_v<typename SPEC::T, float>, "The Metal raytracing backend requires T = float");
+
+        if constexpr (SPEC::ENABLE_OVERLAYS) {
+            malloc(device, renderer.transforms);
+            rendering::raytracing::detail::malloc_overlay_staging(renderer);
+        }
+        if constexpr (SPEC::HAS_FLOW_OVERLAYS) {
+            malloc(device, renderer.transforms_previous);
+            std::memset(data(renderer.transforms_previous), 0, decltype(renderer.transforms_previous)::SPEC::SIZE_BYTES);
+        }
+        if constexpr (SPEC::ENABLE_DYNAMIC_MOTION_BLUR) {
+            malloc(device, renderer.transforms_motion);
+            std::memset(data(renderer.transforms_motion), 0, decltype(renderer.transforms_motion)::SPEC::SIZE_BYTES);
+        }
+
+        renderer.backend = new rendering::raytracing::backends::RendererState<rendering::raytracing::backends::Metal, SPEC>{};
+        renderer.device.context = renderer.backend;
+        auto* ctx = renderer.backend;
+        ctx->device = NS::TransferPtr(MTL::CreateSystemDefaultDevice());
+        utils::assert_exit(device, ctx->device.get() != nullptr, "Metal: no default device available");
+        utils::assert_exit(device, ctx->device->supportsRaytracing(), "Metal: device does not support raytracing");
+        ctx->queue = NS::TransferPtr(ctx->device->newCommandQueue());
+
+        auto compile_options = NS::TransferPtr(MTL::CompileOptions::alloc()->init());
+        compile_options->setLanguageVersion(MTL::LanguageVersion3_0);
+        auto source = NS::TransferPtr(NS::String::string(rendering::raytracing::backends::metal::device_source(), NS::UTF8StringEncoding));
+        NS::Error* error = nullptr;
+        ctx->library = NS::TransferPtr(ctx->device->newLibrary(source.get(), compile_options.get(), &error));
+        if(ctx->library.get() == nullptr){
+            RL_TOOLS_RENDERING_RAYTRACING_LOG_ERR("Metal device source compilation failed: " << (error != nullptr ? error->localizedDescription()->utf8String() : "unknown error"));
+            utils::assert_exit(device, false, "Metal device source compilation failed");
+        }
+
+        constexpr typename SPEC::TI cam_pixels = SPEC::CAM_PIXELS;
+        // the output tensors alias the shared MTLBuffers the kernels write (see frame_buffer/
+        // depth_buffer/segmentation_buffer/collision_results/observation accessors)
+        if constexpr (SPEC::HAS_RGB) {
+            ctx->frame_buffer = NS::TransferPtr(ctx->device->newBuffer((size_t)SPEC::NUM_CAMERAS * cam_pixels * sizeof(uint32_t), MTL::ResourceStorageModeShared));
+            renderer.frame_buffer._data = (uint32_t*)ctx->frame_buffer->contents();
+        }
+        if constexpr (SPEC::HAS_DEPTH) {
+            ctx->depth_buffer = NS::TransferPtr(ctx->device->newBuffer((size_t)SPEC::NUM_CAMERAS * cam_pixels * sizeof(float), MTL::ResourceStorageModeShared));
+            renderer.depth_buffer._data = (float*)ctx->depth_buffer->contents();
+        }
+        if constexpr (SPEC::HAS_SEGMENTATION) {
+            ctx->segmentation_buffer = NS::TransferPtr(ctx->device->newBuffer((size_t)SPEC::NUM_CAMERAS * cam_pixels * sizeof(uint32_t), MTL::ResourceStorageModeShared));
+            renderer.segmentation_buffer._data = (uint32_t*)ctx->segmentation_buffer->contents();
+        }
+        if constexpr (SPEC::HAS_NORMALS) {
+            ctx->normals_buffer = NS::TransferPtr(ctx->device->newBuffer((size_t)SPEC::NUM_CAMERAS * cam_pixels * 3 * sizeof(float), MTL::ResourceStorageModeShared));
+            renderer.normals_buffer._data = (float*)ctx->normals_buffer->contents();
+        }
+        if constexpr (SPEC::HAS_FLOW) {
+            ctx->flow_buffer = NS::TransferPtr(ctx->device->newBuffer((size_t)SPEC::NUM_CAMERAS * cam_pixels * 2 * sizeof(float), MTL::ResourceStorageModeShared));
+            renderer.flow_buffer._data = (float*)ctx->flow_buffer->contents();
+            ctx->flow_status_buffer = NS::TransferPtr(ctx->device->newBuffer((size_t)SPEC::NUM_CAMERAS * cam_pixels * sizeof(uint32_t), MTL::ResourceStorageModeShared));
+            renderer.flow_status_buffer._data = (uint32_t*)ctx->flow_status_buffer->contents();
+            if constexpr (SPEC::HAS_FLOW_OVERLAYS) {
+                ctx->flow_deltas = NS::TransferPtr(ctx->device->newBuffer((size_t)2 * SPEC::NUM_OVERLAYS * SPEC::MAX_OVERLAY_INSTANCES * 12 * sizeof(float), MTL::ResourceStorageModeShared));
+            }
+        }
+        if constexpr (SPEC::HAS_FLOW_FORWARD) {
+            ctx->flow_forward_buffer = NS::TransferPtr(ctx->device->newBuffer((size_t)SPEC::NUM_CAMERAS * cam_pixels * 2 * sizeof(float), MTL::ResourceStorageModeShared));
+            renderer.flow_forward_buffer._data = (float*)ctx->flow_forward_buffer->contents();
+            ctx->flow_forward_status_buffer = NS::TransferPtr(ctx->device->newBuffer((size_t)SPEC::NUM_CAMERAS * cam_pixels * sizeof(uint32_t), MTL::ResourceStorageModeShared));
+            renderer.flow_forward_status_buffer._data = (uint32_t*)ctx->flow_forward_status_buffer->contents();
+        }
+        if constexpr (SPEC::HAS_OBSERVATION) {
+            static_assert(utils::typing::is_same_v<typename SPEC::OBSERVATION_T, float>, "The Metal raytracing backend requires OBSERVATION_T = float");
+            ctx->observation = NS::TransferPtr(ctx->device->newBuffer((size_t)SPEC::NUM_CAMERAS * cam_pixels * SPEC::OBSERVATION_CHANNELS * sizeof(float), MTL::ResourceStorageModeShared));
+            renderer.observation._data = (float*)ctx->observation->contents();
+        }
+        if constexpr (SPEC::ENABLE_DYNAMIC_MOTION_BLUR) {
+            // linear-radiance accumulators for the launch-level motion loop, resolved into the
+            // packed frame buffer / depth buffer by resolve_outputs after the last sample pass
+            if constexpr (SPEC::HAS_RGB) {
+                ctx->rgb_accumulator = NS::TransferPtr(ctx->device->newBuffer((size_t)SPEC::NUM_CAMERAS * cam_pixels * 3 * sizeof(float), MTL::ResourceStorageModeShared));
+                renderer.rgb_accumulator._data = (float*)ctx->rgb_accumulator->contents();
+            }
+            if constexpr (SPEC::HAS_DEPTH) {
+                ctx->depth_accumulator = NS::TransferPtr(ctx->device->newBuffer((size_t)SPEC::NUM_CAMERAS * cam_pixels * sizeof(float), MTL::ResourceStorageModeShared));
+                renderer.depth_accumulator._data = (float*)ctx->depth_accumulator->contents();
+            }
+        }
+#if !RL_TOOLS_RENDERING_RAYTRACING_DISABLE_PROBE_RAYS
+        ctx->collision_results = NS::TransferPtr(ctx->device->newBuffer((size_t)SPEC::NUM_CAMERAS * SPEC::NUM_PROBES * sizeof(rendering::raytracing::CollisionResult), MTL::ResourceStorageModeShared));
+        renderer.collision_results._data = (rendering::raytracing::CollisionResult*)ctx->collision_results->contents();
+#endif
+        ctx->launch_params = NS::TransferPtr(ctx->device->newBuffer(sizeof(metal::LaunchParams), MTL::ResourceStorageModeShared));
+
+        // camera tensors alias the shared MTLBuffers the encoders bind (see cameras(device,
+        // renderer)): host writes are consumed by the next launch with no staging copy
+        constexpr size_t camera_bytes = (size_t)SPEC::NUM_CAMERAS * sizeof(rendering::raytracing::Camera<typename SPEC::T>);
+        ctx->cameras = NS::TransferPtr(ctx->device->newBuffer(camera_bytes, MTL::ResourceStorageModeShared));
+        renderer.cameras._data = (rendering::raytracing::Camera<typename SPEC::T>*)ctx->cameras->contents();
+        if constexpr (SPEC::HAS_CAMERA_PAIR) {
+            ctx->cameras_open = NS::TransferPtr(ctx->device->newBuffer(camera_bytes, MTL::ResourceStorageModeShared));
+            renderer.cameras_open._data = (rendering::raytracing::Camera<typename SPEC::T>*)ctx->cameras_open->contents();
+        }
+        if constexpr (SPEC::HAS_FLOW) {
+            ctx->cameras_previous = NS::TransferPtr(ctx->device->newBuffer(camera_bytes, MTL::ResourceStorageModeShared));
+            renderer.cameras_previous._data = (rendering::raytracing::Camera<typename SPEC::T>*)ctx->cameras_previous->contents();
+        }
+        rendering::raytracing::detail::announce_backend(renderer);
+    }
+
+    template <typename DEVICE, typename SPEC, typename METADATA_T>
+    void init(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer, const rendering::raytracing::Scene& scene, const rendering::raytracing::AssetPool& pool, const rendering::SceneMetadata<METADATA_T>& metadata){
+        validate_semantic_segmentation<SPEC>(device, scene, pool);
+        renderer.max_ray_length = (typename SPEC::T)metadata.max_ray_length;
+        rendering::raytracing::detail::announce_configuration<SPEC>();
+        rendering::raytracing::detail::resolve_environment<SPEC>(scene.environment, renderer.ambient_color, renderer.miss_color_0, renderer.miss_color_1);
+        namespace metal = rendering::raytracing::backends::metal;
+        auto& ctx = metal::context(renderer);
+        NS::AutoreleasePool* autorelease_pool = NS::AutoreleasePool::alloc()->init();
+
+        RL_TOOLS_RENDERING_RAYTRACING_LOG("building " << scene.objects.size() << " object(s), " << scene.instances.size() << " instance(s) ...");
+
+        // scene-dependent resources from a previous init are released here; the compiled library
+        // and pipelines below are spec-dependent and survive re-init
+        ctx.mesh_buffers.clear();
+        ctx.mesh_textures.clear();
+        ctx.object_acceleration_structures.clear();
+        ctx.scene_lights = NS::SharedPtr<MTL::Buffer>{};
+
+        const uint8_t dummy_pixel[4] = {255, 255, 255, 255};
+        ctx.dummy_texture = metal::make_texture(ctx.device.get(), dummy_pixel, 1, 1, false);
+        const MTL::ResourceID dummy_texture_id = ctx.dummy_texture->gpuResourceID();
+
+        std::vector<const rendering::raytracing::Object*> all_objects;
+        for(const auto& object : scene.objects){
+            all_objects.push_back(&object);
+        }
+        if constexpr (SPEC::ENABLE_OVERLAYS){
+            rendering::raytracing::detail::register_pool_assets(device, renderer, pool, all_objects);
+        }
+
+        size_t total_meshes = 0;
+        for(const auto* object : all_objects){
+            total_meshes += object->meshes.size();
+        }
+        std::vector<metal::MeshRecord> mesh_records(total_meshes);
+        std::vector<uint32_t> object_record_base;
+        std::vector<std::vector<NS::Object*>> object_geometry_descriptors(all_objects.size());
+
+        size_t m = 0;
+        for(size_t object_i = 0; object_i < all_objects.size(); object_i++){
+            object_record_base.push_back((uint32_t)m);
+            for(const auto& md : all_objects[object_i]->meshes){
+            auto& geometry_descriptors = object_geometry_descriptors[object_i];
+            metal::MeshRecord& record = mesh_records[m];
+            record = {};
+            record.texture = dummy_texture_id;
+            record.normal_map = dummy_texture_id;
+            record.metallic_roughness_map = dummy_texture_id;
+            record.emissive_map = dummy_texture_id;
+            record.occlusion_map = dummy_texture_id;
+
+            size_t num_triangles = md.indices.size() / 3;
+
+            auto vertex_buffer = NS::TransferPtr(ctx.device->newBuffer(md.vertices.data(), md.vertices.size() * sizeof(float), MTL::ResourceStorageModeShared));
+            auto index_buffer = NS::TransferPtr(ctx.device->newBuffer(md.indices.data(), md.indices.size() * sizeof(int), MTL::ResourceStorageModeShared));
+            record.vertices = vertex_buffer->gpuAddress();
+            record.index = index_buffer->gpuAddress();
+
+            MTL::AccelerationStructureTriangleGeometryDescriptor* geometry = MTL::AccelerationStructureTriangleGeometryDescriptor::descriptor();
+            geometry->setVertexBuffer(vertex_buffer.get());
+            geometry->setVertexStride(3 * sizeof(float));
+            geometry->setVertexFormat(MTL::AttributeFormatFloat3);
+            geometry->setIndexBuffer(index_buffer.get());
+            geometry->setIndexType(MTL::IndexTypeUInt32);
+            geometry->setTriangleCount(num_triangles);
+            geometry->setOpaque(true);
+            geometry_descriptors.push_back(geometry);
+
+            ctx.mesh_buffers.push_back(vertex_buffer);
+            ctx.mesh_buffers.push_back(index_buffer);
+
+            if(!md.tex_coords.empty()){
+                auto tex_coord_buffer = NS::TransferPtr(ctx.device->newBuffer(md.tex_coords.data(), md.tex_coords.size() * sizeof(float), MTL::ResourceStorageModeShared));
+                record.tex_coord = tex_coord_buffer->gpuAddress();
+                ctx.mesh_buffers.push_back(tex_coord_buffer);
+            }
+            if(!md.normals.empty()){
+                auto normal_buffer = NS::TransferPtr(ctx.device->newBuffer(md.normals.data(), md.normals.size() * sizeof(float), MTL::ResourceStorageModeShared));
+                record.normal = normal_buffer->gpuAddress();
+                ctx.mesh_buffers.push_back(normal_buffer);
+            }
+
+            record.color[0] = md.color[0];
+            record.color[1] = md.color[1];
+            record.color[2] = md.color[2];
+            record.metallic = md.metallic;
+            record.roughness = md.roughness;
+            record.opacity = md.opacity;
+            record.emissive[0] = md.emissive[0];
+            record.emissive[1] = md.emissive[1];
+            record.emissive[2] = md.emissive[2];
+            record.alpha_cutoff = md.alpha_cutoff;
+            record.alpha_mode = md.alpha_mode;
+            record.unlit = md.unlit;
+
+            if(md.texture.present()){
+                auto texture = metal::make_texture(ctx.device.get(), md.texture.pixels.data(), md.texture.width, md.texture.height, true);
+                record.texture = texture->gpuResourceID();
+                record.has_texture = 1;
+                ctx.mesh_textures.push_back(texture);
+            }
+            if(md.normal_map.present()){
+                auto texture = metal::make_texture(ctx.device.get(), md.normal_map.pixels.data(), md.normal_map.width, md.normal_map.height, false);
+                record.normal_map = texture->gpuResourceID();
+                record.has_normal_map = 1;
+                ctx.mesh_textures.push_back(texture);
+            }
+            if(md.metallic_roughness_map.present()){
+                auto texture = metal::make_texture(ctx.device.get(), md.metallic_roughness_map.pixels.data(), md.metallic_roughness_map.width, md.metallic_roughness_map.height, false);
+                record.metallic_roughness_map = texture->gpuResourceID();
+                record.has_metallic_roughness_map = 1;
+                ctx.mesh_textures.push_back(texture);
+            }
+            if(md.emissive_map.present()){
+                auto texture = metal::make_texture(ctx.device.get(), md.emissive_map.pixels.data(), md.emissive_map.width, md.emissive_map.height, true);
+                record.emissive_map = texture->gpuResourceID();
+                record.has_emissive_map = 1;
+                ctx.mesh_textures.push_back(texture);
+            }
+            if(md.occlusion_map.present()){
+                auto texture = metal::make_texture(ctx.device.get(), md.occlusion_map.pixels.data(), md.occlusion_map.width, md.occlusion_map.height, false);
+                record.occlusion_map = texture->gpuResourceID();
+                record.has_occlusion_map = 1;
+                ctx.mesh_textures.push_back(texture);
+            }
+            m++;
+            }
+        }
+
+        // one primitive (bottom-level) acceleration structure per object, shared by its instances
+        {
+            MTL::CommandBuffer* command_buffer = ctx.queue->commandBuffer();
+            MTL::AccelerationStructureCommandEncoder* encoder = command_buffer->accelerationStructureCommandEncoder();
+            std::vector<NS::SharedPtr<MTL::Buffer>> scratch_buffers;
+            for(size_t object_i = 0; object_i < all_objects.size(); object_i++){
+                auto& geometry_descriptors = object_geometry_descriptors[object_i];
+                NS::Array* geometry_array = NS::Array::array((const NS::Object* const*)geometry_descriptors.data(), geometry_descriptors.size());
+                MTL::PrimitiveAccelerationStructureDescriptor* accel_descriptor = MTL::PrimitiveAccelerationStructureDescriptor::descriptor();
+                accel_descriptor->setGeometryDescriptors(geometry_array);
+                MTL::AccelerationStructureSizes sizes = ctx.device->accelerationStructureSizes(accel_descriptor);
+                auto object_acceleration_structure = NS::TransferPtr(ctx.device->newAccelerationStructure(sizes.accelerationStructureSize));
+                auto scratch_buffer = NS::TransferPtr(ctx.device->newBuffer(sizes.buildScratchBufferSize, MTL::ResourceStorageModePrivate));
+                encoder->buildAccelerationStructure(object_acceleration_structure.get(), accel_descriptor, scratch_buffer.get(), 0);
+                scratch_buffers.push_back(scratch_buffer);
+                ctx.object_acceleration_structures.push_back(object_acceleration_structure);
+            }
+            encoder->endEncoding();
+            command_buffer->commit();
+            command_buffer->waitUntilCompleted();
+        }
+
+        // instance (top-level) acceleration structure over the placed objects; overlay slots get
+        // their own tiny instance structures but share the global instance-id-indexed data arrays
+        ctx.object_record_base = object_record_base;
+        ctx.object_classes.clear();
+        for(const auto* object_pointer : all_objects){
+            ctx.object_classes.push_back(object_pointer->segmentation_class);
+        }
+        ctx.num_scene_instances = (uint32_t)scene.instances.size();
+        {
+            size_t total_instances = scene.instances.size();
+            if constexpr (SPEC::ENABLE_OVERLAYS){
+                // dynamic motion blur appends one extra overlay id range per motion sample:
+                // sample s's instances carry shifted user ids (num_scene + num_overlay_slots *
+                // (1 + s) + slot) so each pass's hits index a private slice of the instance side
+                // tables with no shader changes; the contract ids [num_scene, num_scene +
+                // num_overlay_slots) stay exclusive to the shutter-close state segmentation reads
+                // flow appends one more range after the blur samples for the previous instant's
+                // structures (visibility rays and forward primaries only: no side tables read)
+                constexpr size_t num_overlay_slots = (size_t)SPEC::NUM_OVERLAYS * SPEC::MAX_OVERLAY_INSTANCES;
+                constexpr size_t overlay_id_ranges = (SPEC::ENABLE_DYNAMIC_MOTION_BLUR ? (size_t)1 + SPEC::MOTION_BLUR_SAMPLES : 1) + (SPEC::HAS_FLOW_OVERLAYS ? 1 : 0);
+                total_instances += num_overlay_slots * overlay_id_ranges;
+            }
+            std::vector<MTL::AccelerationStructureUserIDInstanceDescriptor> instance_descriptors(scene.instances.size());
+            std::vector<metal::InstanceData> instance_data(total_instances, metal::InstanceData{});
+            std::vector<uint32_t> instance_record_base(total_instances, 0);
+            for(size_t instance_i = 0; instance_i < scene.instances.size(); instance_i++){
+                const auto& instance = scene.instances[instance_i];
+                auto& descriptor = instance_descriptors[instance_i];
+                descriptor = {};
+                for(int column = 0; column < 4; column++){
+                    descriptor.transformationMatrix.columns[column] = MTL::PackedFloat3{instance.transform[0 + column], instance.transform[4 + column], instance.transform[8 + column]};
+                }
+                descriptor.options = MTL::AccelerationStructureInstanceOptionOpaque;
+                descriptor.mask = 0xFFFFFFFFu;
+                descriptor.intersectionFunctionTableOffset = 0;
+                descriptor.accelerationStructureIndex = (uint32_t)instance.object;
+                descriptor.userID = (uint32_t)instance_i; // global instance id (contract: segmentation_object, operations_cpu_common.h)
+
+                auto& data = instance_data[instance_i];
+                data = {};
+                for(int element = 0; element < 12; element++){
+                    data.object_to_world[element] = instance.transform[element];
+                }
+                if(instance.identity){
+                    std::memcpy(data.world_to_object, data.object_to_world, sizeof(data.world_to_object));
+                }
+                else{
+                    rendering::raytracing::detail::invert_transform(instance.transform, data.world_to_object);
+                }
+                data.identity = instance.identity ? 1 : 0;
+                instance_record_base[instance_i] = object_record_base[instance.object];
+            }
+            ctx.instance_descriptors = NS::TransferPtr(ctx.device->newBuffer(instance_descriptors.data(), instance_descriptors.size() * sizeof(MTL::AccelerationStructureUserIDInstanceDescriptor), MTL::ResourceStorageModeShared));
+            ctx.instance_data = NS::TransferPtr(ctx.device->newBuffer(instance_data.data(), instance_data.size() * sizeof(metal::InstanceData), MTL::ResourceStorageModeShared));
+            ctx.instance_record_base = NS::TransferPtr(ctx.device->newBuffer(instance_record_base.data(), instance_record_base.size() * sizeof(uint32_t), MTL::ResourceStorageModeShared));
+            std::vector<uint32_t> instance_classes(total_instances > 0 ? total_instances : 1, 0);
+            for(size_t instance_i = 0; instance_i < scene.instances.size(); instance_i++){
+                instance_classes[instance_i] = ctx.object_classes[scene.instances[instance_i].object];
+            }
+            ctx.instance_classes = NS::TransferPtr(ctx.device->newBuffer(instance_classes.data(), instance_classes.size() * sizeof(uint32_t), MTL::ResourceStorageModeShared));
+
+            std::vector<NS::Object*> object_acceleration_structure_pointers;
+            for(auto& object_acceleration_structure : ctx.object_acceleration_structures){
+                object_acceleration_structure_pointers.push_back(object_acceleration_structure.get());
+            }
+            NS::Array* object_array = NS::Array::array((const NS::Object* const*)object_acceleration_structure_pointers.data(), object_acceleration_structure_pointers.size());
+            MTL::InstanceAccelerationStructureDescriptor* instance_accel_descriptor = MTL::InstanceAccelerationStructureDescriptor::descriptor();
+            instance_accel_descriptor->setInstancedAccelerationStructures(object_array);
+            instance_accel_descriptor->setInstanceCount(scene.instances.size());
+            instance_accel_descriptor->setInstanceDescriptorBuffer(ctx.instance_descriptors.get());
+            instance_accel_descriptor->setInstanceDescriptorType(MTL::AccelerationStructureInstanceDescriptorTypeUserID);
+            MTL::AccelerationStructureSizes sizes = ctx.device->accelerationStructureSizes(instance_accel_descriptor);
+            ctx.acceleration_structure = NS::TransferPtr(ctx.device->newAccelerationStructure(sizes.accelerationStructureSize));
+            auto scratch_buffer = NS::TransferPtr(ctx.device->newBuffer(sizes.buildScratchBufferSize, MTL::ResourceStorageModePrivate));
+            MTL::CommandBuffer* command_buffer = ctx.queue->commandBuffer();
+            MTL::AccelerationStructureCommandEncoder* encoder = command_buffer->accelerationStructureCommandEncoder();
+            encoder->buildAccelerationStructure(ctx.acceleration_structure.get(), instance_accel_descriptor, scratch_buffer.get(), 0);
+            encoder->endEncoding();
+            command_buffer->commit();
+            command_buffer->waitUntilCompleted();
+        }
+        ctx.overlay_acceleration_structures.clear();
+        ctx.overlay_instance_descriptors.clear();
+        ctx.overlay_sample_instance_descriptors.clear();
+        ctx.overlay_scratch_buffers.clear();
+        ctx.overlay_acceleration_structures_previous.clear();
+        ctx.overlay_instance_descriptors_previous.clear();
+        ctx.overlay_scratch_buffers_previous.clear();
+        ctx.overlay_structures_previous = NS::SharedPtr<MTL::Buffer>{};
+        if constexpr (SPEC::ENABLE_OVERLAYS){
+            std::vector<NS::Object*> object_acceleration_structure_pointers;
+            for(auto& object_acceleration_structure : ctx.object_acceleration_structures){
+                object_acceleration_structure_pointers.push_back(object_acceleration_structure.get());
+            }
+            NS::Array* object_array = NS::Array::array((const NS::Object* const*)object_acceleration_structure_pointers.data(), object_acceleration_structure_pointers.size());
+            // one overlay structure set per instant: the current one, plus the previous one under flow
+            const auto create_overlay_structure_set = [&](std::vector<NS::SharedPtr<MTL::Buffer>>& instance_descriptors, std::vector<NS::SharedPtr<MTL::AccelerationStructure>>& acceleration_structures, std::vector<NS::SharedPtr<MTL::Buffer>>& scratch_buffers) -> NS::SharedPtr<MTL::Buffer> {
+                std::vector<metal::OverlayStructureEntry> overlay_entries(SPEC::NUM_OVERLAYS, metal::OverlayStructureEntry{});
+                for(size_t overlay = 0; overlay < SPEC::NUM_OVERLAYS; overlay++){
+                    auto descriptor_buffer = NS::TransferPtr(ctx.device->newBuffer((size_t)SPEC::MAX_OVERLAY_INSTANCES * sizeof(MTL::AccelerationStructureUserIDInstanceDescriptor), MTL::ResourceStorageModeShared));
+                    MTL::InstanceAccelerationStructureDescriptor* overlay_descriptor = MTL::InstanceAccelerationStructureDescriptor::descriptor();
+                    overlay_descriptor->setInstancedAccelerationStructures(object_array);
+                    overlay_descriptor->setInstanceCount(SPEC::MAX_OVERLAY_INSTANCES);
+                    overlay_descriptor->setInstanceDescriptorBuffer(descriptor_buffer.get());
+                    overlay_descriptor->setInstanceDescriptorType(MTL::AccelerationStructureInstanceDescriptorTypeUserID);
+                    MTL::AccelerationStructureSizes overlay_sizes = ctx.device->accelerationStructureSizes(overlay_descriptor);
+                    auto overlay_acceleration_structure = NS::TransferPtr(ctx.device->newAccelerationStructure(overlay_sizes.accelerationStructureSize));
+                    auto overlay_scratch = NS::TransferPtr(ctx.device->newBuffer(overlay_sizes.buildScratchBufferSize, MTL::ResourceStorageModePrivate));
+                    instance_descriptors.push_back(descriptor_buffer);
+                    acceleration_structures.push_back(overlay_acceleration_structure);
+                    scratch_buffers.push_back(overlay_scratch);
+                    overlay_entries[overlay].structure = overlay_acceleration_structure->gpuResourceID();
+                    overlay_entries[overlay].num_active = 0;
+                }
+                return NS::TransferPtr(ctx.device->newBuffer(overlay_entries.data(), overlay_entries.size() * sizeof(metal::OverlayStructureEntry), MTL::ResourceStorageModeShared));
+            };
+            ctx.overlay_structures = create_overlay_structure_set(ctx.overlay_instance_descriptors, ctx.overlay_acceleration_structures, ctx.overlay_scratch_buffers);
+            if constexpr (SPEC::HAS_FLOW_OVERLAYS){
+                ctx.overlay_structures_previous = create_overlay_structure_set(ctx.overlay_instance_descriptors_previous, ctx.overlay_acceleration_structures_previous, ctx.overlay_scratch_buffers_previous);
+            }
+            if constexpr (SPEC::ENABLE_DYNAMIC_MOTION_BLUR){
+                for(size_t sample = 0; sample < SPEC::MOTION_BLUR_SAMPLES; sample++){
+                    for(size_t overlay = 0; overlay < SPEC::NUM_OVERLAYS; overlay++){
+                        ctx.overlay_sample_instance_descriptors.push_back(NS::TransferPtr(ctx.device->newBuffer((size_t)SPEC::MAX_OVERLAY_INSTANCES * sizeof(MTL::AccelerationStructureUserIDInstanceDescriptor), MTL::ResourceStorageModeShared)));
+                    }
+                }
+            }
+            std::vector<uint32_t> attachments_init((size_t)SPEC::NUM_CAMERAS * SPEC::MAX_OVERLAYS_PER_CAMERA, 0xFFFFFFFFu);
+            ctx.overlay_attachments = NS::TransferPtr(ctx.device->newBuffer(attachments_init.data(), attachments_init.size() * sizeof(uint32_t), MTL::ResourceStorageModeShared));
+            rendering::raytracing::detail::reset_overlay_state(renderer);
+        }
+        else if(ctx.overlay_structures.get() == nullptr){
+            // never dereferenced (fc_overlay_count == 0) but keeps the kernel bindings valid
+            ctx.overlay_structures = NS::TransferPtr(ctx.device->newBuffer(sizeof(metal::OverlayStructureEntry), MTL::ResourceStorageModeShared));
+            ctx.overlay_attachments = NS::TransferPtr(ctx.device->newBuffer(sizeof(uint32_t), MTL::ResourceStorageModeShared));
+        }
+
+        ctx.mesh_records = NS::TransferPtr(ctx.device->newBuffer(mesh_records.data(), mesh_records.size() * sizeof(metal::MeshRecord), MTL::ResourceStorageModeShared));
+
+        const auto scene_lights = rendering::raytracing::detail::effective_scene_lights<SPEC::HAS_RGB && SPEC::SHADING::PBR_SHADING>(scene);
+        if(!scene_lights.empty()){
+            ctx.scene_lights = NS::TransferPtr(ctx.device->newBuffer(scene_lights.data(), scene_lights.size() * sizeof(rendering::raytracing::SceneLight), MTL::ResourceStorageModeShared));
+        }
+
+        auto* params = (metal::LaunchParams*)ctx.launch_params->contents();
+        *params = {};
+        params->fb_width = SPEC::FB_WIDTH;
+        params->fb_height = SPEC::FB_HEIGHT;
+        params->cam_width = SPEC::CAM_WIDTH;
+        params->cam_height = SPEC::CAM_HEIGHT;
+        params->grid_cols = SPEC::GRID_COLS;
+        params->num_cameras = SPEC::NUM_CAMERAS;
+        params->num_probes = SPEC::NUM_PROBES;
+        params->num_scene_lights = (uint32_t)scene_lights.size();
+        params->max_depth = renderer.max_ray_length > 0 ? renderer.max_ray_length : 1e30f;
+        params->max_dist = renderer.max_ray_length;
+        for(int component = 0; component < 3; component++){
+            params->ambient_color[component] = (float)renderer.ambient_color[component];
+            params->miss_color_0[component] = (float)renderer.miss_color_0[component];
+            params->miss_color_1[component] = (float)renderer.miss_color_1[component];
+        }
+        params->first_overlay_instance = (uint32_t)scene.instances.size();
+        params->num_overlay_slots = SPEC::ENABLE_OVERLAYS ? (uint32_t)SPEC::NUM_OVERLAYS * SPEC::MAX_OVERLAY_INSTANCES : 0u;
+
+        if(!ctx.pipelines_built){
+            auto constants = NS::TransferPtr(MTL::FunctionConstantValues::alloc()->init());
+            using CONSTANTS = rendering::raytracing::backends::SpecializationConstants<SPEC>;
+            bool srgb_output = CONSTANTS::SRGB_OUTPUT;
+            bool motion_blur = CONSTANTS::MOTION_BLUR;
+            int motion_samples = CONSTANTS::MOTION_SAMPLES;
+            int aa_grid = CONSTANTS::AA_GRID;
+            bool checker_background = CONSTANTS::CHECKER_BACKGROUND;
+            bool load_textures = CONSTANTS::LOAD_TEXTURES;
+            bool normal_shading = CONSTANTS::NORMAL_SHADING;
+            bool metallic_reflections = CONSTANTS::METALLIC_REFLECTIONS;
+            bool pbr_shading = CONSTANTS::PBR_SHADING;
+            bool punctual_light_shadows = CONSTANTS::PUNCTUAL_LIGHT_SHADOWS;
+            int overlay_count = CONSTANTS::OVERLAY_COUNT;
+            bool semantic_segmentation = CONSTANTS::SEMANTIC_SEGMENTATION;
+            bool has_observation = CONSTANTS::HAS_OBSERVATION;
+            bool dynamic_motion_blur = CONSTANTS::DYNAMIC_MOTION_BLUR;
+            bool resolve_rgb = CONSTANTS::RESOLVE_RGB;
+            bool resolve_depth = CONSTANTS::RESOLVE_DEPTH;
+            constants->setConstantValue(&srgb_output, MTL::DataTypeBool, (NS::UInteger)metal::function_constants::SRGB_OUTPUT);
+            constants->setConstantValue(&motion_blur, MTL::DataTypeBool, (NS::UInteger)metal::function_constants::MOTION_BLUR);
+            constants->setConstantValue(&motion_samples, MTL::DataTypeInt, (NS::UInteger)metal::function_constants::MOTION_SAMPLES);
+            constants->setConstantValue(&aa_grid, MTL::DataTypeInt, (NS::UInteger)metal::function_constants::AA_GRID);
+            constants->setConstantValue(&checker_background, MTL::DataTypeBool, (NS::UInteger)metal::function_constants::CHECKER_BACKGROUND);
+            constants->setConstantValue(&load_textures, MTL::DataTypeBool, (NS::UInteger)metal::function_constants::LOAD_TEXTURES);
+            constants->setConstantValue(&normal_shading, MTL::DataTypeBool, (NS::UInteger)metal::function_constants::NORMAL_SHADING);
+            constants->setConstantValue(&metallic_reflections, MTL::DataTypeBool, (NS::UInteger)metal::function_constants::METALLIC_REFLECTIONS);
+            constants->setConstantValue(&pbr_shading, MTL::DataTypeBool, (NS::UInteger)metal::function_constants::PBR_SHADING);
+            constants->setConstantValue(&punctual_light_shadows, MTL::DataTypeBool, (NS::UInteger)metal::function_constants::PUNCTUAL_LIGHT_SHADOWS);
+            constants->setConstantValue(&overlay_count, MTL::DataTypeInt, (NS::UInteger)metal::function_constants::OVERLAY_COUNT);
+            constants->setConstantValue(&semantic_segmentation, MTL::DataTypeBool, (NS::UInteger)metal::function_constants::SEMANTIC_SEGMENTATION);
+            constants->setConstantValue(&has_observation, MTL::DataTypeBool, (NS::UInteger)metal::function_constants::HAS_OBSERVATION);
+            constants->setConstantValue(&dynamic_motion_blur, MTL::DataTypeBool, (NS::UInteger)metal::function_constants::DYNAMIC_MOTION_BLUR);
+            constants->setConstantValue(&resolve_rgb, MTL::DataTypeBool, (NS::UInteger)metal::function_constants::RESOLVE_RGB);
+            constants->setConstantValue(&resolve_depth, MTL::DataTypeBool, (NS::UInteger)metal::function_constants::RESOLVE_DEPTH);
+
+            auto make_pipeline = [&](const char* name) -> NS::SharedPtr<MTL::ComputePipelineState> {
+                NS::Error* error = nullptr;
+                auto function = NS::TransferPtr(ctx.library->newFunction(NS::String::string(name, NS::UTF8StringEncoding), constants.get(), &error));
+                if(function.get() == nullptr){
+                    RL_TOOLS_RENDERING_RAYTRACING_LOG_ERR("Metal function specialization failed for " << name << ": " << (error != nullptr ? error->localizedDescription()->utf8String() : "unknown error"));
+                    utils::assert_exit(device, false, "Metal function specialization failed");
+                }
+                error = nullptr;
+                auto pipeline = NS::TransferPtr(ctx.device->newComputePipelineState(function.get(), &error));
+                if(pipeline.get() == nullptr){
+                    RL_TOOLS_RENDERING_RAYTRACING_LOG_ERR("Metal pipeline creation failed for " << name << ": " << (error != nullptr ? error->localizedDescription()->utf8String() : "unknown error"));
+                    utils::assert_exit(device, false, "Metal pipeline creation failed");
+                }
+                return pipeline;
+            };
+
+            if constexpr (SPEC::HAS_RGB) {
+                ctx.rgb_pipeline = make_pipeline("render_rgb");
+            }
+            if constexpr (SPEC::HAS_DEPTH) {
+                ctx.depth_pipeline = make_pipeline("render_depth");
+            }
+            if constexpr (SPEC::HAS_SEGMENTATION) {
+                ctx.segmentation_pipeline = make_pipeline("render_segmentation");
+            }
+            if constexpr (SPEC::HAS_NORMALS) {
+                ctx.normals_pipeline = make_pipeline("render_normals");
+            }
+            if constexpr (SPEC::HAS_FLOW) {
+                ctx.flow_pipeline = make_pipeline("render_flow");
+            }
+            if constexpr (SPEC::HAS_FLOW_FORWARD) {
+                ctx.flow_forward_pipeline = make_pipeline("render_flow_forward");
+            }
+            if constexpr (SPEC::ENABLE_DYNAMIC_MOTION_BLUR) {
+                ctx.resolve_pipeline = make_pipeline("resolve_outputs");
+            }
+#if !RL_TOOLS_RENDERING_RAYTRACING_DISABLE_PROBE_RAYS
+            ctx.collision_pipeline = make_pipeline("render_collision");
+#endif
+            ctx.pipelines_built = true;
+        }
+
+        if constexpr (SPEC::ENABLE_OVERLAYS){
+            update(device, renderer); // publish the (empty) overlays and the attachment table
+        }
+
+        autorelease_pool->release();
+    }
+
+
+    // the host must not mutate the shared instance/descriptor buffers while renders or builds
+    // are in flight, hence the wait at the top; the commit is not waited on — command buffers on
+    // one queue execute in commit order and Metal's hazard tracking orders the acceleration
+    // structure writes before any subsequent render pass that reads them
+    template <typename DEVICE, typename SPEC>
+    void update_launch(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+        static_assert(SPEC::ENABLE_OVERLAYS, "update requires an overlay-enabled renderer specification");
+        namespace metal = rendering::raytracing::backends::metal;
+        using TI = typename SPEC::TI;
+        auto& ctx = metal::context(renderer);
+        metal::wait_in_flight(ctx);
+        NS::AutoreleasePool* autorelease_pool = NS::AutoreleasePool::alloc()->init();
+
+        auto* overlay_entries = (metal::OverlayStructureEntry*)ctx.overlay_structures->contents();
+        auto* instance_data = (metal::InstanceData*)ctx.instance_data->contents();
+        auto* instance_record_base = (uint32_t*)ctx.instance_record_base->contents();
+
+        MTL::CommandBuffer* command_buffer = nullptr;
+        MTL::AccelerationStructureCommandEncoder* encoder = nullptr;
+        std::vector<NS::Object*> object_acceleration_structure_pointers;
+        for(auto& object_acceleration_structure : ctx.object_acceleration_structures){
+            object_acceleration_structure_pointers.push_back(object_acceleration_structure.get());
+        }
+        NS::Array* object_array = NS::Array::array((const NS::Object* const*)object_acceleration_structure_pointers.data(), object_acceleration_structure_pointers.size());
+
+        rendering::raytracing::detail::publish_overlay_transforms(renderer);
+        // rebuilt unconditionally: producers may write the transforms tensors directly, which
+        // leaves no host-observable pending row
+        for(TI overlay = 0; overlay < SPEC::NUM_OVERLAYS; overlay++){
+            auto& overlay_state = renderer.overlays[overlay];
+            const size_t base = (size_t)ctx.num_scene_instances + (size_t)overlay * SPEC::MAX_OVERLAY_INSTANCES;
+            auto* descriptors = (MTL::AccelerationStructureUserIDInstanceDescriptor*)ctx.overlay_instance_descriptors[overlay]->contents();
+            uint32_t num_active = 0;
+            for(TI slot = 0; slot < SPEC::MAX_OVERLAY_INSTANCES; slot++){
+                const auto& host_slot = overlay_state.slots[slot];
+                if(!host_slot.active) continue;
+                const size_t global = base + slot;
+                const float* world = data(renderer.transforms) + ((size_t)overlay * SPEC::MAX_OVERLAY_INSTANCES + slot) * 12;
+                auto& descriptor = descriptors[num_active];
+                descriptor = {};
+                for(int column = 0; column < 4; column++){
+                    descriptor.transformationMatrix.columns[column] = MTL::PackedFloat3{world[0 + column], world[4 + column], world[8 + column]};
+                }
+                descriptor.options = MTL::AccelerationStructureInstanceOptionOpaque;
+                descriptor.mask = 0xFFFFFFFFu;
+                descriptor.intersectionFunctionTableOffset = 0;
+                descriptor.accelerationStructureIndex = (uint32_t)host_slot.object;
+                descriptor.userID = (uint32_t)global; // global instance id (contract: segmentation_object, operations_cpu_common.h)
+
+                auto& data = instance_data[global];
+                data = {};
+                for(int element = 0; element < 12; element++){
+                    data.object_to_world[element] = world[element];
+                }
+                const bool identity = rendering::raytracing::detail::transform_is_identity(world);
+                if(identity){
+                    std::memcpy(data.world_to_object, data.object_to_world, sizeof(data.world_to_object));
+                }
+                else{
+                    rendering::raytracing::detail::invert_transform(world, data.world_to_object);
+                }
+                data.identity = identity ? 1 : 0;
+                instance_record_base[global] = ctx.object_record_base[host_slot.object];
+                ((uint32_t*)ctx.instance_classes->contents())[global] = ctx.object_classes[host_slot.object];
+                num_active++;
+            }
+            overlay_entries[overlay].num_active = num_active;
+            if(num_active > 0){
+                if(command_buffer == nullptr){
+                    command_buffer = ctx.queue->commandBuffer();
+                    encoder = command_buffer->accelerationStructureCommandEncoder();
+                }
+                MTL::InstanceAccelerationStructureDescriptor* overlay_descriptor = MTL::InstanceAccelerationStructureDescriptor::descriptor();
+                overlay_descriptor->setInstancedAccelerationStructures(object_array);
+                overlay_descriptor->setInstanceCount(num_active);
+                overlay_descriptor->setInstanceDescriptorBuffer(ctx.overlay_instance_descriptors[overlay].get());
+                overlay_descriptor->setInstanceDescriptorType(MTL::AccelerationStructureInstanceDescriptorTypeUserID);
+                encoder->buildAccelerationStructure(ctx.overlay_acceleration_structures[overlay].get(), overlay_descriptor, ctx.overlay_scratch_buffers[overlay].get(), 0);
+            }
+        }
+        if constexpr (SPEC::HAS_FLOW_OVERLAYS){
+            // the previous instant: descriptors from transforms_previous over the shifted id range
+            // after the blur samples (no side tables: the flow passes only need hit distances and
+            // ids), built into the resident second structure set; plus the per-slot delta table
+            constexpr size_t num_overlay_slots = (size_t)SPEC::NUM_OVERLAYS * SPEC::MAX_OVERLAY_INSTANCES;
+            constexpr size_t previous_range = 1 + (SPEC::ENABLE_DYNAMIC_MOTION_BLUR ? (size_t)SPEC::MOTION_BLUR_SAMPLES : 0);
+            auto* overlay_entries_previous = (metal::OverlayStructureEntry*)ctx.overlay_structures_previous->contents();
+            for(TI overlay = 0; overlay < SPEC::NUM_OVERLAYS; overlay++){
+                auto& overlay_state = renderer.overlays[overlay];
+                const size_t base = (size_t)ctx.num_scene_instances + num_overlay_slots * previous_range + (size_t)overlay * SPEC::MAX_OVERLAY_INSTANCES;
+                auto* descriptors = (MTL::AccelerationStructureUserIDInstanceDescriptor*)ctx.overlay_instance_descriptors_previous[overlay]->contents();
+                uint32_t num_active = 0;
+                for(TI slot = 0; slot < SPEC::MAX_OVERLAY_INSTANCES; slot++){
+                    const auto& host_slot = overlay_state.slots[slot];
+                    if(!host_slot.active) continue;
+                    const float* world = data(renderer.transforms_previous) + ((size_t)overlay * SPEC::MAX_OVERLAY_INSTANCES + slot) * 12;
+                    auto& descriptor = descriptors[num_active];
+                    descriptor = {};
+                    for(int column = 0; column < 4; column++){
+                        descriptor.transformationMatrix.columns[column] = MTL::PackedFloat3{world[0 + column], world[4 + column], world[8 + column]};
+                    }
+                    descriptor.options = MTL::AccelerationStructureInstanceOptionOpaque;
+                    descriptor.mask = 0xFFFFFFFFu;
+                    descriptor.intersectionFunctionTableOffset = 0;
+                    descriptor.accelerationStructureIndex = (uint32_t)host_slot.object;
+                    descriptor.userID = (uint32_t)(base + slot);
+                    num_active++;
+                }
+                overlay_entries_previous[overlay].num_active = num_active;
+                if(num_active > 0){
+                    if(command_buffer == nullptr){
+                        command_buffer = ctx.queue->commandBuffer();
+                        encoder = command_buffer->accelerationStructureCommandEncoder();
+                    }
+                    MTL::InstanceAccelerationStructureDescriptor* overlay_descriptor = MTL::InstanceAccelerationStructureDescriptor::descriptor();
+                    overlay_descriptor->setInstancedAccelerationStructures(object_array);
+                    overlay_descriptor->setInstanceCount(num_active);
+                    overlay_descriptor->setInstanceDescriptorBuffer(ctx.overlay_instance_descriptors_previous[overlay].get());
+                    overlay_descriptor->setInstanceDescriptorType(MTL::AccelerationStructureInstanceDescriptorTypeUserID);
+                    encoder->buildAccelerationStructure(ctx.overlay_acceleration_structures_previous[overlay].get(), overlay_descriptor, ctx.overlay_scratch_buffers_previous[overlay].get(), 0);
+                }
+            }
+            rendering::raytracing::detail::compose_flow_deltas(renderer, data(renderer.transforms), data(renderer.transforms_previous), (float*)ctx.flow_deltas->contents());
+        }
+        if constexpr (SPEC::ENABLE_DYNAMIC_MOTION_BLUR){
+            // per-sample overlay instance descriptors + shifted-id side-table slices consumed by
+            // render_launch's per-sample builds; layout mirrors the shutter-close path above
+            constexpr size_t SLAB = (size_t)SPEC::NUM_OVERLAYS * SPEC::MAX_OVERLAY_INSTANCES * 12;
+            constexpr size_t num_overlay_slots = (size_t)SPEC::NUM_OVERLAYS * SPEC::MAX_OVERLAY_INSTANCES;
+            for(TI sample = 0; sample < SPEC::MOTION_BLUR_SAMPLES; sample++){
+                const float* transforms_base = data(renderer.transforms_motion) + sample * SLAB;
+                for(TI overlay = 0; overlay < SPEC::NUM_OVERLAYS; overlay++){
+                    auto& overlay_state = renderer.overlays[overlay];
+                    auto* descriptors = (MTL::AccelerationStructureUserIDInstanceDescriptor*)ctx.overlay_sample_instance_descriptors[(size_t)sample * SPEC::NUM_OVERLAYS + overlay]->contents();
+                    uint32_t num_active = 0;
+                    for(TI slot = 0; slot < SPEC::MAX_OVERLAY_INSTANCES; slot++){
+                        const auto& host_slot = overlay_state.slots[slot];
+                        if(!host_slot.active) continue;
+                        const size_t global = (size_t)ctx.num_scene_instances + num_overlay_slots * (1 + (size_t)sample) + (size_t)overlay * SPEC::MAX_OVERLAY_INSTANCES + slot;
+                        const float* world = transforms_base + ((size_t)overlay * SPEC::MAX_OVERLAY_INSTANCES + slot) * 12;
+                        auto& descriptor = descriptors[num_active];
+                        descriptor = {};
+                        for(int column = 0; column < 4; column++){
+                            descriptor.transformationMatrix.columns[column] = MTL::PackedFloat3{world[0 + column], world[4 + column], world[8 + column]};
+                        }
+                        descriptor.options = MTL::AccelerationStructureInstanceOptionOpaque;
+                        descriptor.mask = 0xFFFFFFFFu;
+                        descriptor.intersectionFunctionTableOffset = 0;
+                        descriptor.accelerationStructureIndex = (uint32_t)host_slot.object;
+                        descriptor.userID = (uint32_t)global;
+
+                        auto& data_entry = instance_data[global];
+                        data_entry = {};
+                        for(int element = 0; element < 12; element++){
+                            data_entry.object_to_world[element] = world[element];
+                        }
+                        const bool identity = rendering::raytracing::detail::transform_is_identity(world);
+                        if(identity){
+                            std::memcpy(data_entry.world_to_object, data_entry.object_to_world, sizeof(data_entry.world_to_object));
+                        }
+                        else{
+                            rendering::raytracing::detail::invert_transform(world, data_entry.world_to_object);
+                        }
+                        data_entry.identity = identity ? 1 : 0;
+                        instance_record_base[global] = ctx.object_record_base[host_slot.object];
+                        ((uint32_t*)ctx.instance_classes->contents())[global] = ctx.object_classes[host_slot.object];
+                        num_active++;
+                    }
+                }
+            }
+        }
+        if(renderer.attachments_dirty){
+            auto* attachments = (uint32_t*)ctx.overlay_attachments->contents();
+            for(size_t index = 0; index < (size_t)SPEC::NUM_CAMERAS * SPEC::MAX_OVERLAYS_PER_CAMERA; index++){
+                attachments[index] = (uint32_t)renderer.attachments[index];
+            }
+            renderer.attachments_dirty = false;
+        }
+        if(command_buffer != nullptr){
+            encoder->endEncoding();
+            command_buffer->commit();
+            ctx.in_flight_update = NS::RetainPtr(command_buffer);
+        }
+        autorelease_pool->release();
+    }
+
+    template <typename DEVICE, typename SPEC>
+    void update_sync(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+        static_assert(SPEC::ENABLE_OVERLAYS, "update requires an overlay-enabled renderer specification");
+        namespace metal = rendering::raytracing::backends::metal;
+        auto& ctx = metal::context(renderer);
+        if(ctx.in_flight_update.get() != nullptr){
+            ctx.in_flight_update->waitUntilCompleted();
+            ctx.in_flight_update.reset();
+        }
+    }
+
+    // the previous instant becomes the current one: shared-storage buffers and host tensors, so
+    // plain copies after the in-flight wait; rows the host has not yet published reach
+    // transforms_previous through their staging mirror at the next update
+    template <typename DEVICE, typename SPEC>
+    void advance_flow_launch(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+        static_assert(SPEC::HAS_FLOW, "advance_flow requires a flow-capable renderer specification");
+        namespace metal = rendering::raytracing::backends::metal;
+        metal::wait_in_flight(metal::context(renderer));
+        std::memcpy(data(renderer.cameras_previous), data(renderer.cameras), decltype(renderer.cameras)::SPEC::SIZE_BYTES);
+        if constexpr (SPEC::HAS_FLOW_OVERLAYS){
+            rendering::raytracing::detail::advance_flow_staging(renderer);
+            std::memcpy(data(renderer.transforms_previous), data(renderer.transforms), decltype(renderer.transforms)::SPEC::SIZE_BYTES);
+        }
+    }
+
+    template <typename DEVICE, typename SPEC>
+    void advance_flow_sync(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+        static_assert(SPEC::HAS_FLOW, "advance_flow requires a flow-capable renderer specification");
+    }
+
+    template <typename DEVICE, typename SPEC>
+    void advance_flow(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+        advance_flow_launch(device, renderer);
+        advance_flow_sync(device, renderer);
+    }
+
+
+    // renderer memory-domain copies: shared-storage buffers are host-addressable after the
+    // in-flight wait, so the transfer delegates to the host-device tensor copy
+    template <typename TO_DEVICE, typename FROM_SPEC, typename TO_SPEC>
+    void copy(rendering::raytracing::backends::Device<rendering::raytracing::backends::Metal>& from_device, TO_DEVICE& to_device, const Tensor<FROM_SPEC>& from, Tensor<TO_SPEC>& to){
+        rendering::raytracing::backends::metal::wait_in_flight(*from_device.context);
+        copy(to_device, to_device, from, to);
+    }
+    template <typename FROM_DEVICE, typename FROM_SPEC, typename TO_SPEC>
+    void copy(FROM_DEVICE& from_device, rendering::raytracing::backends::Device<rendering::raytracing::backends::Metal>& to_device, const Tensor<FROM_SPEC>& from, Tensor<TO_SPEC>& to){
+        rendering::raytracing::backends::metal::wait_in_flight(*to_device.context);
+        copy(from_device, from_device, from, to);
+    }
+
+    template <typename DEVICE, typename SPEC>
+    void generate_probe_directions(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+#if RL_TOOLS_RENDERING_RAYTRACING_DISABLE_PROBE_RAYS
+        RL_TOOLS_RENDERING_RAYTRACING_LOG("Probe rays disabled (RL_TOOLS_RENDERING_RAYTRACING_DISABLE_PROBE_RAYS=1)");
+        return;
+#else
+        namespace metal = rendering::raytracing::backends::metal;
+        auto& ctx = metal::context(renderer);
+        std::vector<float> dirs = rendering::raytracing::detail::generate_probe_direction_vectors<SPEC>();
+        ctx.probe_directions = NS::TransferPtr(ctx.device->newBuffer(dirs.data(), dirs.size() * sizeof(float), MTL::ResourceStorageModeShared));
+#endif
+    }
+
+    // render produces the image outputs the spec declares; the collision-probe pass is the
+    // separate probe verb so it can be scheduled independently (e.g. alongside update)
+    template <typename DEVICE, typename SPEC>
+    void render_launch(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+        namespace metal = rendering::raytracing::backends::metal;
+        using TI = typename SPEC::TI;
+        auto& ctx = metal::context(renderer);
+        NS::AutoreleasePool* pool = NS::AutoreleasePool::alloc()->init();
+        MTL::CommandBuffer* command_buffer = ctx.queue->commandBuffer();
+        if constexpr (SPEC::ENABLE_DYNAMIC_MOTION_BLUR){
+            // one command buffer per frame: zero the accumulators, then per motion sample build
+            // the overlay structures in place from that sample's descriptors and run the
+            // accumulate dispatches, then restore the shutter-close state (segmentation, probes,
+            // steady state) and resolve — encoder order plus Metal hazard tracking on the
+            // acceleration structures and accumulators serializes builds against dispatches
+            uint32_t num_active_per_overlay[SPEC::NUM_OVERLAYS];
+            uint32_t total_active = 0;
+            for(TI overlay = 0; overlay < SPEC::NUM_OVERLAYS; overlay++){
+                uint32_t num_active = 0;
+                for(TI slot = 0; slot < SPEC::MAX_OVERLAY_INSTANCES; slot++){
+                    num_active += renderer.overlays[overlay].slots[slot].active ? 1 : 0;
+                }
+                num_active_per_overlay[overlay] = num_active;
+                total_active += num_active;
+            }
+            {
+                MTL::BlitCommandEncoder* blit = command_buffer->blitCommandEncoder();
+                if constexpr (SPEC::HAS_RGB){
+                    blit->fillBuffer(ctx.rgb_accumulator.get(), NS::Range::Make(0, ctx.rgb_accumulator->length()), 0);
+                }
+                if constexpr (SPEC::HAS_DEPTH){
+                    blit->fillBuffer(ctx.depth_accumulator.get(), NS::Range::Make(0, ctx.depth_accumulator->length()), 0);
+                }
+                blit->endEncoding();
+            }
+            std::vector<NS::Object*> object_acceleration_structure_pointers;
+            for(auto& object_acceleration_structure : ctx.object_acceleration_structures){
+                object_acceleration_structure_pointers.push_back(object_acceleration_structure.get());
+            }
+            NS::Array* object_array = NS::Array::array((const NS::Object* const*)object_acceleration_structure_pointers.data(), object_acceleration_structure_pointers.size());
+            const auto encode_overlay_builds = [&](auto&& descriptor_buffer_for_overlay){
+                if(total_active == 0){
+                    return;
+                }
+                MTL::AccelerationStructureCommandEncoder* encoder = command_buffer->accelerationStructureCommandEncoder();
+                for(TI overlay = 0; overlay < SPEC::NUM_OVERLAYS; overlay++){
+                    if(num_active_per_overlay[overlay] == 0) continue;
+                    MTL::InstanceAccelerationStructureDescriptor* overlay_descriptor = MTL::InstanceAccelerationStructureDescriptor::descriptor();
+                    overlay_descriptor->setInstancedAccelerationStructures(object_array);
+                    overlay_descriptor->setInstanceCount(num_active_per_overlay[overlay]);
+                    overlay_descriptor->setInstanceDescriptorBuffer(descriptor_buffer_for_overlay(overlay));
+                    overlay_descriptor->setInstanceDescriptorType(MTL::AccelerationStructureInstanceDescriptorTypeUserID);
+                    encoder->buildAccelerationStructure(ctx.overlay_acceleration_structures[overlay].get(), overlay_descriptor, ctx.overlay_scratch_buffers[overlay].get(), 0);
+                }
+                encoder->endEncoding();
+            };
+            for(TI sample = 0; sample < SPEC::MOTION_BLUR_SAMPLES; sample++){
+                encode_overlay_builds([&](TI overlay){ return ctx.overlay_sample_instance_descriptors[(size_t)sample * SPEC::NUM_OVERLAYS + overlay].get(); });
+                const float shutter_t = ((float)sample + 0.5f) / (float)SPEC::MOTION_BLUR_SAMPLES;
+                if constexpr (SPEC::HAS_RGB) {
+                    metal::encode_fullscreen_pass<SPEC>(ctx, command_buffer, ctx.rgb_pipeline.get(), ctx.frame_buffer.get(), shutter_t);
+                }
+                if constexpr (SPEC::HAS_DEPTH) {
+                    metal::encode_fullscreen_pass<SPEC>(ctx, command_buffer, ctx.depth_pipeline.get(), ctx.depth_buffer.get(), shutter_t);
+                }
+            }
+            encode_overlay_builds([&](TI overlay){ return ctx.overlay_instance_descriptors[overlay].get(); });
+            if constexpr (SPEC::HAS_SEGMENTATION) {
+                metal::encode_fullscreen_pass<SPEC>(ctx, command_buffer, ctx.segmentation_pipeline.get(), ctx.segmentation_buffer.get());
+            }
+            if constexpr (SPEC::HAS_NORMALS) {
+                metal::encode_fullscreen_pass<SPEC>(ctx, command_buffer, ctx.normals_pipeline.get(), ctx.normals_buffer.get());
+            }
+            if constexpr (SPEC::HAS_FLOW) {
+                metal::encode_fullscreen_pass<SPEC>(ctx, command_buffer, ctx.flow_pipeline.get(), ctx.flow_buffer.get(), 0.f, ctx.flow_status_buffer.get());
+            }
+            if constexpr (SPEC::HAS_FLOW_FORWARD) {
+                metal::encode_fullscreen_pass<SPEC>(ctx, command_buffer, ctx.flow_forward_pipeline.get(), ctx.flow_forward_buffer.get(), 0.f, ctx.flow_forward_status_buffer.get());
+            }
+            metal::encode_resolve_pass<SPEC>(ctx, command_buffer);
+            command_buffer->commit();
+            ctx.in_flight = NS::RetainPtr(command_buffer);
+            pool->release();
+            return;
+        }
+        if constexpr (SPEC::HAS_RGB) {
+            metal::encode_fullscreen_pass<SPEC>(ctx, command_buffer, ctx.rgb_pipeline.get(), ctx.frame_buffer.get());
+        }
+        if constexpr (SPEC::HAS_DEPTH) {
+            metal::encode_fullscreen_pass<SPEC>(ctx, command_buffer, ctx.depth_pipeline.get(), ctx.depth_buffer.get());
+        }
+        if constexpr (SPEC::HAS_SEGMENTATION) {
+            metal::encode_fullscreen_pass<SPEC>(ctx, command_buffer, ctx.segmentation_pipeline.get(), ctx.segmentation_buffer.get());
+        }
+        if constexpr (SPEC::HAS_NORMALS) {
+            metal::encode_fullscreen_pass<SPEC>(ctx, command_buffer, ctx.normals_pipeline.get(), ctx.normals_buffer.get());
+        }
+        if constexpr (SPEC::HAS_FLOW) {
+            metal::encode_fullscreen_pass<SPEC>(ctx, command_buffer, ctx.flow_pipeline.get(), ctx.flow_buffer.get(), 0.f, ctx.flow_status_buffer.get());
+        }
+        if constexpr (SPEC::HAS_FLOW_FORWARD) {
+            metal::encode_fullscreen_pass<SPEC>(ctx, command_buffer, ctx.flow_forward_pipeline.get(), ctx.flow_forward_buffer.get(), 0.f, ctx.flow_forward_status_buffer.get());
+        }
+        command_buffer->commit();
+        ctx.in_flight = NS::RetainPtr(command_buffer);
+        pool->release();
+    }
+
+    template <typename DEVICE, typename SPEC>
+    void render_sync(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+        namespace metal = rendering::raytracing::backends::metal;
+        auto& ctx = metal::context(renderer);
+        if(ctx.in_flight.get() != nullptr){
+            ctx.in_flight->waitUntilCompleted();
+            ctx.in_flight.reset();
+        }
+    }
+
+    template <typename DEVICE, typename SPEC>
+    void probe_launch(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+        namespace metal = rendering::raytracing::backends::metal;
+        auto& ctx = metal::context(renderer);
+        if(ctx.collision_pipeline.get() != nullptr){
+            NS::AutoreleasePool* pool = NS::AutoreleasePool::alloc()->init();
+            MTL::CommandBuffer* command_buffer = ctx.queue->commandBuffer();
+            metal::encode_collision_pass<SPEC>(ctx, command_buffer);
+            command_buffer->commit();
+            ctx.in_flight_collision = NS::RetainPtr(command_buffer);
+            pool->release();
+        }
+    }
+
+    template <typename DEVICE, typename SPEC>
+    void probe_sync(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+        namespace metal = rendering::raytracing::backends::metal;
+        auto& ctx = metal::context(renderer);
+        if(ctx.in_flight_collision.get() != nullptr){
+            ctx.in_flight_collision->waitUntilCompleted();
+            ctx.in_flight_collision.reset();
+        }
+    }
+
+    template <typename DEVICE, typename SPEC>
+    std::string device_name(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+        return renderer.backend->device->name()->utf8String();
+    }
+
+    template <typename DEVICE, typename SPEC>
+    void synchronize(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+        namespace metal = rendering::raytracing::backends::metal;
+        metal::wait_in_flight(metal::context(renderer));
+    }
+
+    template <typename DEVICE, typename SPEC>
+    void free(DEVICE& device, rendering::raytracing::Renderer<SPEC, rendering::raytracing::backends::Metal>& renderer){
+        namespace metal = rendering::raytracing::backends::metal;
+        if(renderer.backend != nullptr){
+            metal::wait_in_flight(metal::context(renderer));
+            delete renderer.backend;
+            renderer.backend = nullptr;
+            renderer.device.context = nullptr;
+        }
+        // the input and output tensors alias shared MTLBuffers destroyed with the context
+        renderer.cameras._data = nullptr;
+        if constexpr (SPEC::HAS_CAMERA_PAIR) {
+            renderer.cameras_open._data = nullptr;
+        }
+        if constexpr (SPEC::HAS_RGB) {
+            renderer.frame_buffer._data = nullptr;
+        }
+        if constexpr (SPEC::HAS_DEPTH) {
+            renderer.depth_buffer._data = nullptr;
+        }
+        if constexpr (SPEC::HAS_SEGMENTATION) {
+            renderer.segmentation_buffer._data = nullptr;
+        }
+        if constexpr (SPEC::HAS_NORMALS) {
+            renderer.normals_buffer._data = nullptr;
+        }
+        if constexpr (SPEC::HAS_FLOW) {
+            renderer.cameras_previous._data = nullptr;
+            renderer.flow_buffer._data = nullptr;
+            renderer.flow_status_buffer._data = nullptr;
+        }
+        if constexpr (SPEC::HAS_FLOW_FORWARD) {
+            renderer.flow_forward_buffer._data = nullptr;
+            renderer.flow_forward_status_buffer._data = nullptr;
+        }
+        if constexpr (SPEC::HAS_OBSERVATION) {
+            renderer.observation._data = nullptr;
+        }
+        renderer.collision_results._data = nullptr;
+        if constexpr (SPEC::ENABLE_OVERLAYS) {
+            free(device, renderer.transforms);
+        }
+        if constexpr (SPEC::HAS_FLOW_OVERLAYS) {
+            free(device, renderer.transforms_previous);
+        }
+        if constexpr (SPEC::ENABLE_DYNAMIC_MOTION_BLUR) {
+            free(device, renderer.transforms_motion);
+            if constexpr (SPEC::HAS_RGB) {
+                renderer.rgb_accumulator._data = nullptr;
+            }
+            if constexpr (SPEC::HAS_DEPTH) {
+                renderer.depth_accumulator._data = nullptr;
+            }
+        }
+    }
+
+}
+RL_TOOLS_NAMESPACE_WRAPPER_END
+
+
+#include "../../operations_cpu_post.h"
+
+#endif