// This file is part of the 64k demo project.
// It implements the Renderer3D class.

#include "3d/renderer.h"
#include "generated/assets.h"
#include "gpu/effects/shader_composer.h"
#include "util/asset_manager.h"
#include <algorithm>
#include <cassert>
#include <cstring>
#include <iostream>

#if !defined(STRIP_ALL)
bool Renderer3D::s_debug_enabled_ = false;
#endif

void Renderer3D::init(WGPUDevice device, WGPUQueue queue,
                      WGPUTextureFormat format) {
  device_ = device;
  queue_ = queue;
  format_ = format;

  WGPUSamplerDescriptor sampler_desc = {};
  sampler_desc.addressModeU = WGPUAddressMode_Repeat;
  sampler_desc.addressModeV = WGPUAddressMode_Repeat;
  sampler_desc.magFilter = WGPUFilterMode_Linear;
  sampler_desc.minFilter = WGPUFilterMode_Linear;
  sampler_desc.maxAnisotropy = 1;
  default_sampler_ = wgpuDeviceCreateSampler(device_, &sampler_desc);

  create_default_resources();
  create_pipeline();
  create_skybox_pipeline();

#if !defined(STRIP_ALL)
  visual_debug_.init(device_, format_);
#endif
}

void Renderer3D::create_skybox_pipeline() {
  WGPUBindGroupLayoutEntry entries[3] = {};
  entries[0].binding = 0;
  entries[0].visibility = WGPUShaderStage_Fragment;
  entries[0].texture.sampleType = WGPUTextureSampleType_Float;
  entries[0].texture.viewDimension = WGPUTextureViewDimension_2D;

  entries[1].binding = 1;
  entries[1].visibility = WGPUShaderStage_Fragment;
  entries[1].sampler.type = WGPUSamplerBindingType_Filtering;

  entries[2].binding = 2;
  entries[2].visibility = WGPUShaderStage_Fragment;
  entries[2].buffer.type = WGPUBufferBindingType_Uniform;
  entries[2].buffer.minBindingSize = sizeof(GlobalUniforms);

  WGPUBindGroupLayoutDescriptor bgl_desc = {};
  bgl_desc.entryCount = 3;
  bgl_desc.entries = entries;
  WGPUBindGroupLayout bgl = wgpuDeviceCreateBindGroupLayout(device_, &bgl_desc);

  WGPUPipelineLayoutDescriptor pl_desc = {};
  pl_desc.bindGroupLayoutCount = 1;
  pl_desc.bindGroupLayouts = &bgl;
  WGPUPipelineLayout pipeline_layout =
      wgpuDeviceCreatePipelineLayout(device_, &pl_desc);

  const uint8_t* shader_code_asset =
      GetAsset(AssetId::ASSET_SHADER_SKYBOX, nullptr);
  std::string shader_source =
      ShaderComposer::Get().Compose({}, (const char*)shader_code_asset);

#if defined(DEMO_CROSS_COMPILE_WIN32)
  WGPUShaderModuleWGSLDescriptor wgsl_desc = {};
  wgsl_desc.chain.sType = WGPUSType_ShaderModuleWGSLDescriptor;
  wgsl_desc.code = shader_source.c_str();
  WGPUShaderModuleDescriptor shader_desc = {};
  shader_desc.nextInChain = (const WGPUChainedStruct*)&wgsl_desc.chain;
#else
  WGPUShaderSourceWGSL wgsl_desc = {};
  wgsl_desc.chain.sType = WGPUSType_ShaderSourceWGSL;
  wgsl_desc.code = str_view(shader_source.c_str());
  WGPUShaderModuleDescriptor shader_desc = {};
  shader_desc.nextInChain = (const WGPUChainedStruct*)&wgsl_desc.chain;
#endif
  WGPUShaderModule shader_module =
      wgpuDeviceCreateShaderModule(device_, &shader_desc);

  WGPURenderPipelineDescriptor desc = {};
  desc.layout = pipeline_layout;
  desc.vertex.module = shader_module;
#if defined(DEMO_CROSS_COMPILE_WIN32)
  desc.vertex.entryPoint = "vs_main";
#else
  desc.vertex.entryPoint = {"vs_main", 7};
#endif
  WGPUColorTargetState color_target = {};
  color_target.format = format_;
  color_target.writeMask = WGPUColorWriteMask_All;
  WGPUFragmentState fragment = {};
  fragment.module = shader_module;
#if defined(DEMO_CROSS_COMPILE_WIN32)
  fragment.entryPoint = "fs_main";
#else
  fragment.entryPoint = {"fs_main", 7};
#endif
  fragment.targetCount = 1;
  fragment.targets = &color_target;
  desc.fragment = &fragment;
  desc.primitive.topology = WGPUPrimitiveTopology_TriangleList;
  desc.primitive.cullMode = WGPUCullMode_Back;
  desc.primitive.frontFace = WGPUFrontFace_CCW;
  desc.multisample.count = 1;
  desc.multisample.mask = 0xFFFFFFFF;

  WGPUDepthStencilState depth_stencil = {};
  depth_stencil.format = WGPUTextureFormat_Depth24Plus;
  depth_stencil.depthWriteEnabled = WGPUOptionalBool_False;
  depth_stencil.depthCompare = WGPUCompareFunction_Always;
  desc.depthStencil = &depth_stencil;

  skybox_pipeline_ = wgpuDeviceCreateRenderPipeline(device_, &desc);
  wgpuBindGroupLayoutRelease(bgl);
  wgpuPipelineLayoutRelease(pipeline_layout);
  wgpuShaderModuleRelease(shader_module);
}

void Renderer3D::shutdown() {
#if !defined(STRIP_ALL)
  visual_debug_.shutdown();
#endif

  if (default_sampler_)
    wgpuSamplerRelease(default_sampler_);
  if (pipeline_)
    wgpuRenderPipelineRelease(pipeline_);
  if (pipeline_no_bvh_)
    wgpuRenderPipelineRelease(pipeline_no_bvh_);
  if (bind_group_)
    wgpuBindGroupRelease(bind_group_);
  if (skybox_pipeline_)
    wgpuRenderPipelineRelease(skybox_pipeline_);
  if (skybox_bind_group_)
    wgpuBindGroupRelease(skybox_bind_group_);
  if (global_uniform_buffer_)
    wgpuBufferRelease(global_uniform_buffer_);
  if (object_storage_buffer_)
    wgpuBufferRelease(object_storage_buffer_);
  if (bvh_storage_buffer_)
    wgpuBufferRelease(bvh_storage_buffer_);
  if (depth_view_)
    wgpuTextureViewRelease(depth_view_);
  if (depth_texture_)
    wgpuTextureRelease(depth_texture_);
}

void Renderer3D::resize(int width, int height) {
  if (width == width_ && height == height_)
    return;
  width_ = width;
  height_ = height;

  if (depth_view_)
    wgpuTextureViewRelease(depth_view_);
  if (depth_texture_)
    wgpuTextureRelease(depth_texture_);

  WGPUTextureDescriptor desc = {};
  desc.usage = WGPUTextureUsage_RenderAttachment;
  desc.dimension = WGPUTextureDimension_2D;
  desc.size = {(uint32_t)width, (uint32_t)height, 1};
  desc.format = WGPUTextureFormat_Depth24Plus;
  desc.mipLevelCount = 1;
  desc.sampleCount = 1;
  depth_texture_ = wgpuDeviceCreateTexture(device_, &desc);

  WGPUTextureViewDescriptor view_desc = {};
  view_desc.format = WGPUTextureFormat_Depth24Plus;
  view_desc.dimension = WGPUTextureViewDimension_2D;
  view_desc.aspect = WGPUTextureAspect_DepthOnly;
  view_desc.arrayLayerCount = 1;
  view_desc.mipLevelCount = 1;
  depth_view_ = wgpuTextureCreateView(depth_texture_, &view_desc);
}

void Renderer3D::create_default_resources() {
  global_uniform_buffer_ =
      gpu_create_buffer(device_, sizeof(GlobalUniforms),
                        WGPUBufferUsage_Uniform | WGPUBufferUsage_CopyDst,
                        nullptr)
          .buffer;
  object_storage_buffer_ =
      gpu_create_buffer(device_, sizeof(ObjectData) * kMaxObjects,
                        WGPUBufferUsage_Storage | WGPUBufferUsage_CopyDst,
                        nullptr)
          .buffer;
  bvh_storage_buffer_ = gpu_create_buffer(
      device_, sizeof(BVHNode) * kMaxObjects * 2, // Capacity for a full tree
      WGPUBufferUsage_Storage | WGPUBufferUsage_CopyDst, nullptr)
                            .buffer;
}

void Renderer3D::set_noise_texture(WGPUTextureView noise_view) {
  noise_texture_view_ = noise_view;
}

void Renderer3D::set_sky_texture(WGPUTextureView sky_view) {
  sky_texture_view_ = sky_view;
}

void Renderer3D::add_debug_aabb(const vec3& min, const vec3& max,
                                const vec3& color) {
#if !defined(STRIP_ALL)
  visual_debug_.add_aabb(min, max, color);
#endif
}

void Renderer3D::create_pipeline() {
  pipeline_ = create_pipeline_impl(true);       // BVH enabled
  pipeline_no_bvh_ = create_pipeline_impl(false); // BVH disabled
  create_mesh_pipeline();
}

void Renderer3D::create_mesh_pipeline() {
  std::vector<WGPUBindGroupLayoutEntry> entries;

  // Binding 0: Global Uniforms
  {
    WGPUBindGroupLayoutEntry e = {};
    e.binding = 0;
    e.visibility = WGPUShaderStage_Vertex | WGPUShaderStage_Fragment;
    e.buffer.type = WGPUBufferBindingType_Uniform;
    e.buffer.minBindingSize = sizeof(GlobalUniforms);
    entries.push_back(e);
  }

  // Binding 1: Object Data
  {
    WGPUBindGroupLayoutEntry e = {};
    e.binding = 1;
    e.visibility = WGPUShaderStage_Vertex | WGPUShaderStage_Fragment;
    e.buffer.type = WGPUBufferBindingType_ReadOnlyStorage;
    e.buffer.minBindingSize = sizeof(ObjectData) * kMaxObjects;
    entries.push_back(e);
  }

  // Binding 2: BVH Nodes (Optional)
  if (bvh_enabled_) {
    WGPUBindGroupLayoutEntry e = {};
    e.binding = 2;
    e.visibility = WGPUShaderStage_Fragment;
    e.buffer.type = WGPUBufferBindingType_ReadOnlyStorage;
    e.buffer.minBindingSize = sizeof(BVHNode) * kMaxObjects * 2;
    entries.push_back(e);
  }

  // Binding 3: Noise Texture
  {
    WGPUBindGroupLayoutEntry e = {};
    e.binding = 3;
    e.visibility = WGPUShaderStage_Fragment;
    e.texture.sampleType = WGPUTextureSampleType_Float;
    e.texture.viewDimension = WGPUTextureViewDimension_2D;
    entries.push_back(e);
  }

  // Binding 4: Default Sampler
  {
    WGPUBindGroupLayoutEntry e = {};
    e.binding = 4;
    e.visibility = WGPUShaderStage_Fragment;
    e.sampler.type = WGPUSamplerBindingType_Filtering;
    entries.push_back(e);
  }

  // Binding 5: Sky Texture
  {
    WGPUBindGroupLayoutEntry e = {};
    e.binding = 5;
    e.visibility = WGPUShaderStage_Fragment;
    e.texture.sampleType = WGPUTextureSampleType_Float;
    e.texture.viewDimension = WGPUTextureViewDimension_2D;
    entries.push_back(e);
  }

  WGPUBindGroupLayoutDescriptor bgl_desc = {};
  bgl_desc.entryCount = (uint32_t)entries.size();
  bgl_desc.entries = entries.data();
  WGPUBindGroupLayout bgl = wgpuDeviceCreateBindGroupLayout(device_, &bgl_desc);

  WGPUPipelineLayoutDescriptor pl_desc = {};
  pl_desc.bindGroupLayoutCount = 1;
  pl_desc.bindGroupLayouts = &bgl;
  WGPUPipelineLayout pipeline_layout =
      wgpuDeviceCreatePipelineLayout(device_, &pl_desc);

  const char* shader_code_asset = (const char*)GetAsset(AssetId::ASSET_SHADER_MESH);
  
  ShaderComposer::CompositionMap composition_map;
  if (bvh_enabled_) {
    composition_map["render/scene_query_mode"] = "render/scene_query_bvh";
  } else {
    composition_map["render/scene_query_mode"] = "render/scene_query_linear";
  }
  std::string shader_source = ShaderComposer::Get().Compose({}, shader_code_asset, composition_map);

#if defined(DEMO_CROSS_COMPILE_WIN32)
  WGPUShaderModuleWGSLDescriptor wgsl_desc = {};
  wgsl_desc.chain.sType = WGPUSType_ShaderModuleWGSLDescriptor;
  wgsl_desc.code = shader_source.c_str();
  WGPUShaderModuleDescriptor shader_desc = {};
  shader_desc.nextInChain = (const WGPUChainedStruct*)&wgsl_desc.chain;
#else
  WGPUShaderSourceWGSL wgsl_desc = {};
  wgsl_desc.chain.sType = WGPUSType_ShaderSourceWGSL;
  wgsl_desc.code = str_view(shader_source.c_str());
  WGPUShaderModuleDescriptor shader_desc = {};
  shader_desc.nextInChain = (const WGPUChainedStruct*)&wgsl_desc.chain;
#endif
  WGPUShaderModule shader_module =
      wgpuDeviceCreateShaderModule(device_, &shader_desc);

  WGPUVertexAttribute vert_attrs[3] = {};
  // pos
  vert_attrs[0].format = WGPUVertexFormat_Float32x3;
  vert_attrs[0].offset = offsetof(MeshVertex, p);
  vert_attrs[0].shaderLocation = 0;
  // norm
  vert_attrs[1].format = WGPUVertexFormat_Float32x3;
  vert_attrs[1].offset = offsetof(MeshVertex, n);
  vert_attrs[1].shaderLocation = 1;
  // uv
  vert_attrs[2].format = WGPUVertexFormat_Float32x2;
  vert_attrs[2].offset = offsetof(MeshVertex, u);
  vert_attrs[2].shaderLocation = 2;

  WGPUVertexBufferLayout vert_layout = {};
  vert_layout.arrayStride = sizeof(MeshVertex);
  vert_layout.stepMode = WGPUVertexStepMode_Vertex;
  vert_layout.attributeCount = 3;
  vert_layout.attributes = vert_attrs;

  WGPURenderPipelineDescriptor desc = {};
  desc.layout = pipeline_layout;
  desc.vertex.module = shader_module;
#if defined(DEMO_CROSS_COMPILE_WIN32)
  desc.vertex.entryPoint = "vs_main";
#else
  desc.vertex.entryPoint = {"vs_main", 7};
#endif
  desc.vertex.bufferCount = 1;
  desc.vertex.buffers = &vert_layout;

  WGPUColorTargetState color_target = {};
  color_target.format = format_;
  color_target.writeMask = WGPUColorWriteMask_All;
  WGPUFragmentState fragment = {};
  fragment.module = shader_module;
#if defined(DEMO_CROSS_COMPILE_WIN32)
  fragment.entryPoint = "fs_main";
#else
  fragment.entryPoint = {"fs_main", 7};
#endif
  fragment.targetCount = 1;
  fragment.targets = &color_target;
  desc.fragment = &fragment;

  desc.primitive.topology = WGPUPrimitiveTopology_TriangleList;
  desc.primitive.cullMode = WGPUCullMode_Back;
  desc.primitive.frontFace = WGPUFrontFace_CCW;

  WGPUDepthStencilState depth_stencil = {};
  depth_stencil.format = WGPUTextureFormat_Depth24Plus;
  depth_stencil.depthWriteEnabled = WGPUOptionalBool_True;
  depth_stencil.depthCompare = WGPUCompareFunction_Less;
  desc.depthStencil = &depth_stencil;

  desc.multisample.count = 1;
  desc.multisample.mask = 0xFFFFFFFF;

  mesh_pipeline_ = wgpuDeviceCreateRenderPipeline(device_, &desc);
  wgpuBindGroupLayoutRelease(bgl);
  wgpuPipelineLayoutRelease(pipeline_layout);
  wgpuShaderModuleRelease(shader_module);
}

WGPURenderPipeline Renderer3D::create_pipeline_impl(bool use_bvh) {
  std::vector<WGPUBindGroupLayoutEntry> entries;

  // Binding 0: Global Uniforms
  {
    WGPUBindGroupLayoutEntry e = {};
    e.binding = 0;
    e.visibility = WGPUShaderStage_Vertex | WGPUShaderStage_Fragment;
    e.buffer.type = WGPUBufferBindingType_Uniform;
    e.buffer.minBindingSize = sizeof(GlobalUniforms);
    entries.push_back(e);
  }

  // Binding 1: Object Data
  {
    WGPUBindGroupLayoutEntry e = {};
    e.binding = 1;
    e.visibility = WGPUShaderStage_Vertex | WGPUShaderStage_Fragment;
    e.buffer.type = WGPUBufferBindingType_ReadOnlyStorage;
    e.buffer.minBindingSize = sizeof(ObjectData) * kMaxObjects;
    entries.push_back(e);
  }

  // Binding 2: BVH Nodes (only if BVH is used)
  if (use_bvh) {
    WGPUBindGroupLayoutEntry e = {};
    e.binding = 2;
    e.visibility = WGPUShaderStage_Fragment;
    e.buffer.type = WGPUBufferBindingType_ReadOnlyStorage;
    e.buffer.minBindingSize = sizeof(BVHNode) * kMaxObjects * 2;
    entries.push_back(e);
  }

  // Binding 3: Noise Texture
  {
    WGPUBindGroupLayoutEntry e = {};
    e.binding = 3;
    e.visibility = WGPUShaderStage_Fragment;
    e.texture.sampleType = WGPUTextureSampleType_Float;
    e.texture.viewDimension = WGPUTextureViewDimension_2D;
    entries.push_back(e);
  }

  // Binding 4: Default Sampler
  {
    WGPUBindGroupLayoutEntry e = {};
    e.binding = 4;
    e.visibility = WGPUShaderStage_Fragment;
    e.sampler.type = WGPUSamplerBindingType_Filtering;
    entries.push_back(e);
  }

  // Binding 5: Sky Texture
  {
    WGPUBindGroupLayoutEntry e = {};
    e.binding = 5;
    e.visibility = WGPUShaderStage_Fragment;
    e.texture.sampleType = WGPUTextureSampleType_Float;
    e.texture.viewDimension = WGPUTextureViewDimension_2D;
    entries.push_back(e);
  }

  WGPUBindGroupLayoutDescriptor bgl_desc = {};
  bgl_desc.entryCount = (uint32_t)entries.size();
  bgl_desc.entries = entries.data();
  WGPUBindGroupLayout bgl = wgpuDeviceCreateBindGroupLayout(device_, &bgl_desc);

  WGPUPipelineLayoutDescriptor pl_desc = {};
  pl_desc.bindGroupLayoutCount = 1;
  pl_desc.bindGroupLayouts = &bgl;
  WGPUPipelineLayout pipeline_layout =
      wgpuDeviceCreatePipelineLayout(device_, &pl_desc);

  const char* asset_data =
      (const char*)GetAsset(AssetId::ASSET_SHADER_RENDERER_3D);
  std::string main_code = asset_data;

  // Use ShaderComposer to dynamically include the correct scene_query snippet
  ShaderComposer::CompositionMap composition_map;
  if (use_bvh) {
    composition_map["render/scene_query_mode"] = "render/scene_query_bvh";
  } else {
    composition_map["render/scene_query_mode"] = "render/scene_query_linear";
  }
  std::string shader_source = ShaderComposer::Get().Compose({}, main_code, composition_map);

#if defined(DEMO_CROSS_COMPILE_WIN32)
  WGPUShaderModuleWGSLDescriptor wgsl_desc = {};
  wgsl_desc.chain.sType = WGPUSType_ShaderModuleWGSLDescriptor;
  wgsl_desc.code = shader_source.c_str();
  WGPUShaderModuleDescriptor shader_desc = {};
  shader_desc.nextInChain = (const WGPUChainedStruct*)&wgsl_desc.chain;
#else
  WGPUShaderSourceWGSL wgsl_desc = {};
  wgsl_desc.chain.sType = WGPUSType_ShaderSourceWGSL;
  wgsl_desc.code = str_view(shader_source.c_str());
  WGPUShaderModuleDescriptor shader_desc = {};
  shader_desc.nextInChain = (const WGPUChainedStruct*)&wgsl_desc.chain;
#endif
  WGPUShaderModule shader_module =
      wgpuDeviceCreateShaderModule(device_, &shader_desc);

  WGPUDepthStencilState depth_stencil = {};
  depth_stencil.format = WGPUTextureFormat_Depth24Plus;
  depth_stencil.depthWriteEnabled = WGPUOptionalBool_True;
  depth_stencil.depthCompare = WGPUCompareFunction_Less;

  WGPURenderPipelineDescriptor desc = {};
  desc.layout = pipeline_layout;
  desc.vertex.module = shader_module;
#if defined(DEMO_CROSS_COMPILE_WIN32)
  desc.vertex.entryPoint = "vs_main";
#else
  desc.vertex.entryPoint = {"vs_main", 7};
#endif
  WGPUColorTargetState color_target = {};
  color_target.format = format_;
  color_target.writeMask = WGPUColorWriteMask_All;
  WGPUFragmentState fragment = {};
  fragment.module = shader_module;
#if defined(DEMO_CROSS_COMPILE_WIN32)
  fragment.entryPoint = "fs_main";
#else
  fragment.entryPoint = {"fs_main", 7};
#endif
  fragment.targetCount = 1;
  fragment.targets = &color_target;
  desc.fragment = &fragment;
  desc.primitive.topology = WGPUPrimitiveTopology_TriangleList;
  desc.primitive.cullMode = WGPUCullMode_Back;
  desc.primitive.frontFace = WGPUFrontFace_CCW;
  desc.depthStencil = &depth_stencil;
  desc.multisample.count = 1;
  desc.multisample.mask = 0xFFFFFFFF;

  WGPURenderPipeline pipeline = wgpuDeviceCreateRenderPipeline(device_, &desc);
  wgpuBindGroupLayoutRelease(bgl);
  wgpuPipelineLayoutRelease(pipeline_layout);
  wgpuShaderModuleRelease(shader_module);

  return pipeline;
}

const Renderer3D::MeshGpuData*
Renderer3D::get_or_create_mesh(AssetId asset_id) {
  auto it = mesh_cache_.find(asset_id);
  if (it != mesh_cache_.end()) {
    return &it->second;
  }

  MeshAsset asset = GetMeshAsset(asset_id);
  if (!asset.vertices || asset.num_vertices == 0) {
    return nullptr;
  }

  MeshGpuData data;
  data.num_indices = asset.num_indices;

  data.vertex_buffer =
      gpu_create_buffer(device_, asset.num_vertices * sizeof(MeshVertex),
                        WGPUBufferUsage_Vertex | WGPUBufferUsage_CopyDst,
                        asset.vertices)
          .buffer;
  data.index_buffer =
      gpu_create_buffer(device_, asset.num_indices * sizeof(uint32_t),
                        WGPUBufferUsage_Index | WGPUBufferUsage_CopyDst,
                        asset.indices)
          .buffer;

  mesh_cache_[asset_id] = data;
  return &mesh_cache_[asset_id];
}

void Renderer3D::update_uniforms(const Scene& scene, const Camera& camera,
                                 float time) {
  GlobalUniforms globals;
  globals.view_proj = camera.get_projection_matrix() * camera.get_view_matrix();
  globals.inv_view_proj = globals.view_proj.inverse();
  globals.camera_pos_time =
      vec4(camera.position.x, camera.position.y, camera.position.z, time);
  globals.params =
      vec4((float)std::min((size_t)kMaxObjects, scene.objects.size()), 0.0f,
           0.0f, 0.0f);
  globals.resolution = vec2((float)width_, (float)height_);
  globals.padding = vec2(0.0f, 0.0f);

  wgpuQueueWriteBuffer(queue_, global_uniform_buffer_, 0, &globals,
                       sizeof(GlobalUniforms));

  std::vector<ObjectData> obj_data;
  for (const auto& obj : scene.objects) {
    ObjectData data;
    data.model = obj.get_model_matrix();

    // Calculate Inverse for point transformation
    data.inv_model = data.model.inverse();

    data.color = obj.color;
    float type_id = 0.0f;
    if (obj.type == ObjectType::SPHERE)
      type_id = 1.0f;
    else if (obj.type == ObjectType::BOX)
      type_id = 2.0f;
    else if (obj.type == ObjectType::TORUS)
      type_id = 3.0f;
    else if (obj.type == ObjectType::PLANE)
      type_id = 4.0f;
    else if (obj.type == ObjectType::MESH)
      type_id = 5.0f;
    data.params = vec4(type_id, 0, 0, 0);
    obj_data.push_back(data);
    if (obj_data.size() >= kMaxObjects)
      break;
  }
  if (!obj_data.empty()) {
    wgpuQueueWriteBuffer(queue_, object_storage_buffer_, 0, obj_data.data(),
                         obj_data.size() * sizeof(ObjectData));
  }

  // Build and upload BVH (always uploaded, used by BVH pipeline)
  BVHBuilder::build(cpu_bvh_, scene.objects);
  if (!cpu_bvh_.nodes.empty()) {
    wgpuQueueWriteBuffer(queue_, bvh_storage_buffer_, 0, cpu_bvh_.nodes.data(),
                         cpu_bvh_.nodes.size() * sizeof(BVHNode));
  }
}

void Renderer3D::draw(WGPURenderPassEncoder pass, const Scene& scene,
                      const Camera& camera, float time) {
  update_uniforms(scene, camera, time);

  // Lazy Bind Group creation

  if (bind_group_)
    wgpuBindGroupRelease(bind_group_);

  std::vector<WGPUBindGroupEntry> bg_entries;

  {
    WGPUBindGroupEntry e = {};
    e.binding = 0;
    e.buffer = global_uniform_buffer_;
    e.size = sizeof(GlobalUniforms);
    bg_entries.push_back(e);
  }

  {
    WGPUBindGroupEntry e = {};
    e.binding = 1;
    e.buffer = object_storage_buffer_;
    e.size = sizeof(ObjectData) * kMaxObjects;
    bg_entries.push_back(e);
  }

  if (bvh_enabled_) {
    WGPUBindGroupEntry e = {};
    e.binding = 2;
    e.buffer = bvh_storage_buffer_;
    e.size = sizeof(BVHNode) * kMaxObjects * 2;
    bg_entries.push_back(e);
  }

  {
    WGPUBindGroupEntry e = {};
    e.binding = 3;
    e.textureView = noise_texture_view_;
    bg_entries.push_back(e);
  }

  {
    WGPUBindGroupEntry e = {};
    e.binding = 4;
    e.sampler = default_sampler_;
    bg_entries.push_back(e);
  }

  {
    WGPUBindGroupEntry e = {};
    e.binding = 5;
    e.textureView =
        sky_texture_view_ ? sky_texture_view_ : noise_texture_view_;
    bg_entries.push_back(e);
  }

  // Select the correct pipeline and bind group layout
  WGPURenderPipeline current_pipeline =
      bvh_enabled_ ? pipeline_ : pipeline_no_bvh_;
  WGPUBindGroupLayout current_layout =
      wgpuRenderPipelineGetBindGroupLayout(current_pipeline, 0);

  WGPUBindGroupDescriptor bg_desc = {};
  bg_desc.layout = current_layout;
  bg_desc.entryCount = (uint32_t)bg_entries.size();
  bg_desc.entries = bg_entries.data();

  bind_group_ = wgpuDeviceCreateBindGroup(device_, &bg_desc);

  wgpuBindGroupLayoutRelease(current_layout);

  wgpuRenderPassEncoderSetPipeline(pass, current_pipeline);

  wgpuRenderPassEncoderSetBindGroup(pass, 0, bind_group_, 0, nullptr);

  uint32_t instance_count =
      (uint32_t)std::min((size_t)kMaxObjects, scene.objects.size());

  if (instance_count > 0) {
    wgpuRenderPassEncoderSetPipeline(pass, current_pipeline);
    wgpuRenderPassEncoderSetBindGroup(pass, 0, bind_group_, 0, nullptr);
    wgpuRenderPassEncoderDraw(pass, 36, instance_count, 0, 0);

    // Mesh pass
    if (mesh_pipeline_) {
      wgpuRenderPassEncoderSetPipeline(pass, mesh_pipeline_);
      // Bind group is the same layout
      wgpuRenderPassEncoderSetBindGroup(pass, 0, bind_group_, 0, nullptr);

      for (uint32_t i = 0; i < instance_count; ++i) {
        const auto& obj = scene.objects[i];
        if (obj.type == ObjectType::MESH) {
          const MeshGpuData* mesh = get_or_create_mesh(obj.mesh_asset_id);
          if (mesh) {
            wgpuRenderPassEncoderSetVertexBuffer(pass, 0, mesh->vertex_buffer, 0,
                                                  WGPU_WHOLE_SIZE);
            wgpuRenderPassEncoderSetIndexBuffer(
                pass, mesh->index_buffer, WGPUIndexFormat_Uint32, 0,
                WGPU_WHOLE_SIZE);
            wgpuRenderPassEncoderDrawIndexed(pass, mesh->num_indices, 1, 0, 0,
                                             i);
          }
        }
      }
    }
  }

#if !defined(STRIP_ALL)
  if (s_debug_enabled_) {
    for (const auto& obj : scene.objects) {
      vec3 extent(1.0f, 1.0f, 1.0f);
      if (obj.type == ObjectType::TORUS) {
        extent = vec3(1.5f, 0.5f, 1.5f);
      }
      visual_debug_.add_box(obj.get_model_matrix(), extent,
                            vec3(1.0f, 1.0f, 0.0f)); // Yellow boxes
    }

    // Calculate ViewProj matrix for the debug renderer
    mat4 view_proj = camera.get_projection_matrix() * camera.get_view_matrix();
    visual_debug_.render(pass, view_proj);
  }
#endif
}

void Renderer3D::render(const Scene& scene, const Camera& camera, float time,

                        WGPUTextureView target_view,

                        WGPUTextureView depth_view_opt) {
  WGPUTextureView depth_view = depth_view_opt ? depth_view_opt : depth_view_;

  if (!depth_view)

    return;

  WGPUCommandEncoder encoder = wgpuDeviceCreateCommandEncoder(device_, nullptr);

  // --- Pass 1: Render Skybox (Clears color, clears and stores depth) ---
  bool skybox_rendered = false;
  if (sky_texture_view_ && skybox_pipeline_) {
    WGPURenderPassColorAttachment sky_color_attachment = {};
    gpu_init_color_attachment(sky_color_attachment, target_view);
    sky_color_attachment.loadOp = WGPULoadOp_Clear; // Clear to black
    sky_color_attachment.storeOp = WGPUStoreOp_Store;
    sky_color_attachment.clearValue = {0.0f, 0.0f, 0.0f, 1.0f};

    WGPURenderPassDepthStencilAttachment sky_depth_attachment = {};
    sky_depth_attachment.view = depth_view;
    sky_depth_attachment.depthLoadOp = WGPULoadOp_Clear; // Clear depth
    sky_depth_attachment.depthStoreOp =
        WGPUStoreOp_Store;                       // Store cleared depth
    sky_depth_attachment.depthClearValue = 1.0f; // Farthest possible depth

    WGPURenderPassDescriptor sky_pass_desc = {};
    sky_pass_desc.colorAttachmentCount = 1;
    sky_pass_desc.colorAttachments = &sky_color_attachment;
    sky_pass_desc.depthStencilAttachment = &sky_depth_attachment;

    WGPURenderPassEncoder sky_pass =
        wgpuCommandEncoderBeginRenderPass(encoder, &sky_pass_desc);
    wgpuRenderPassEncoderSetViewport(sky_pass, 0.0f, 0.0f, (float)width_,
                                     (float)height_, 0.0f, 1.0f);

    if (skybox_bind_group_)
      wgpuBindGroupRelease(skybox_bind_group_);
    WGPUBindGroupEntry bg_entries[3] = {};
    bg_entries[0].binding = 0;
    bg_entries[0].textureView = sky_texture_view_;
    bg_entries[1].binding = 1;
    bg_entries[1].sampler = default_sampler_;
    bg_entries[2].binding = 2;
    bg_entries[2].buffer = global_uniform_buffer_;
    bg_entries[2].size = sizeof(GlobalUniforms);

    WGPUBindGroupDescriptor bg_desc = {};
    bg_desc.layout = wgpuRenderPipelineGetBindGroupLayout(skybox_pipeline_, 0);
    bg_desc.entryCount = 3;
    bg_desc.entries = bg_entries;
    skybox_bind_group_ = wgpuDeviceCreateBindGroup(device_, &bg_desc);
    wgpuBindGroupLayoutRelease(bg_desc.layout);

    wgpuRenderPassEncoderSetPipeline(sky_pass, skybox_pipeline_);
    wgpuRenderPassEncoderSetBindGroup(sky_pass, 0, skybox_bind_group_, 0,
                                      nullptr);
    wgpuRenderPassEncoderDraw(sky_pass, 3, 1, 0, 0); // Draw a full-screen quad
    wgpuRenderPassEncoderEnd(sky_pass);
    wgpuRenderPassEncoderRelease(sky_pass);
    skybox_rendered = true;
  }

  // --- Pass 2: Render Scene Objects (Loads depth, writes depth) ---
  WGPURenderPassColorAttachment obj_color_attachment = {};
  gpu_init_color_attachment(obj_color_attachment, target_view);
  obj_color_attachment.loadOp =
      skybox_rendered ? WGPULoadOp_Load : WGPULoadOp_Clear; // Load or Clear
  obj_color_attachment.storeOp = WGPUStoreOp_Store;
  obj_color_attachment.clearValue = {0.05f, 0.05f, 0.05f,
                                     1.0f}; // Dark gray if no sky

  WGPURenderPassDepthStencilAttachment obj_depth_attachment = {};
  obj_depth_attachment.view = depth_view;
  obj_depth_attachment.depthLoadOp =
      skybox_rendered ? WGPULoadOp_Load : WGPULoadOp_Clear; // Load or Clear
  obj_depth_attachment.depthStoreOp = WGPUStoreOp_Store; // Store object depth
  obj_depth_attachment.depthClearValue = 1.0f;

  WGPURenderPassDescriptor obj_pass_desc = {};
  obj_pass_desc.colorAttachmentCount = 1;
  obj_pass_desc.colorAttachments = &obj_color_attachment;
  obj_pass_desc.depthStencilAttachment = &obj_depth_attachment;

  WGPURenderPassEncoder obj_pass =
      wgpuCommandEncoderBeginRenderPass(encoder, &obj_pass_desc);
  wgpuRenderPassEncoderSetViewport(obj_pass, 0.0f, 0.0f, (float)width_,
                                   (float)height_, 0.0f, 1.0f);
  draw(obj_pass, scene, camera, time);
  wgpuRenderPassEncoderEnd(obj_pass);
  wgpuRenderPassEncoderRelease(obj_pass);

  WGPUCommandBuffer commands = wgpuCommandEncoderFinish(encoder, nullptr);

  wgpuQueueSubmit(queue_, 1, &commands);

  wgpuCommandBufferRelease(commands);

  wgpuCommandEncoderRelease(encoder);
}