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

#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::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 (mesh_pipeline_)
    wgpuRenderPipelineRelease(mesh_pipeline_);
  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_);

  // Clear mesh cache
  for (auto& pair : mesh_cache_) {
    if (pair.second.vertex_buffer)
      wgpuBufferRelease(pair.second.vertex_buffer);
    if (pair.second.index_buffer)
      wgpuBufferRelease(pair.second.index_buffer);
  }
  mesh_cache_.clear();
}

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::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, obj.local_extent.x, obj.local_extent.y, obj.local_extent.z);
    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 = temp_mesh_override_ ? temp_mesh_override_ : 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 = obj.local_extent;
      if (obj.type == ObjectType::TORUS) {
        extent = vec3(1.5f, 0.5f, 1.5f);
      } else if (obj.type != ObjectType::MESH) {
        extent = vec3(1.0f, 1.0f, 1.0f);
      }
      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);
}