atdunbg
a975845214
定位数据（100 背包 + SunTemple，RTX 5060）：GPU 全帧仅 ~3ms （Forward 1.91 + Shadow 0.82），CPU BuildFrame 收集段 5.94ms 才是 帧率瓶颈。三段细分测量：pre 0.2 / each+cull 5.0 / flatten 0.4。 优化（collect 5.94 → 4.6ms，-23%）： 1. Mesh 全局连续槽位（GetSlot）；ViewScratch batch 查找 hash → 数组寻址（slot × kMeshLodCount 槽） 2. Model 级联合包围球（各 mesh bounds 并集）：实体在四视图全部 不可见 → 跳过整个 mesh 循环（多 mesh 模型的粗剔除） 3. worlds/prevWorlds 合并为 stride-2 InstanceMatrices 单次写 （每实例每视图 2 次随机堆访问 → 1 次，cache miss 减半） 遥测：Renderer::GetCollectMs 暴露收集段耗时。剩余 4.6ms 的进一步 压缩（并行收集 / GPU-driven culling）单独立项。
441 lines
17 KiB
C++
441 lines
17 KiB
C++
//
|
||
// Created by atdunbg on 2026/9/21.
|
||
//
|
||
|
||
#include <Engine/Renderer/Model.h>
|
||
#include <Engine/Renderer/Material.h>
|
||
#include <Engine/Vulkan/Device.h>
|
||
#include <Engine/Vulkan/MemoryAllocator.h>
|
||
#include <Engine/Vulkan/DescriptorManager.h>
|
||
#include <Engine/Vulkan/Texture.h>
|
||
#include <Engine/Core/Log.h>
|
||
|
||
#include <assimp/Importer.hpp>
|
||
#include <assimp/scene.h>
|
||
#include <assimp/postprocess.h>
|
||
|
||
#include <meshoptimizer.h>
|
||
|
||
#include <filesystem>
|
||
#include <algorithm>
|
||
|
||
namespace Engine {
|
||
|
||
Model::Model(Device& device, MemoryAllocator& allocator,
|
||
DescriptorManager& descriptorManager,
|
||
vk::DescriptorSetLayout textureLayout,
|
||
const std::string& path)
|
||
: Model(device, allocator, descriptorManager, textureLayout)
|
||
{
|
||
// 同步便捷路径 = 两段式串行（等价异步收割后的主线程消费）
|
||
LoadModel(path);
|
||
}
|
||
|
||
// 空模型骨架：只建默认纹理；mesh/材质由 FromParsed 填充
|
||
Model::Model(Device& device, MemoryAllocator& allocator,
|
||
DescriptorManager& descriptorManager,
|
||
vk::DescriptorSetLayout textureLayout)
|
||
: m_Device(device)
|
||
, m_Allocator(allocator)
|
||
, m_DescriptorManager(descriptorManager)
|
||
, m_TextureLayout(textureLayout)
|
||
{
|
||
InitDefaultTextures();
|
||
}
|
||
|
||
Model::~Model() = default;
|
||
|
||
void Model::RebuildUnionBounds()
|
||
{
|
||
if (m_Meshes.empty()) {
|
||
m_Bounds = {};
|
||
return;
|
||
}
|
||
// 各 mesh 包围球 → AABB 并集 → 外接球（保守，宁多画不漏画）
|
||
glm::vec3 mn(1e9f), mx(-1e9f);
|
||
for (const auto& mesh : m_Meshes) {
|
||
const auto& b = mesh->GetBounds();
|
||
mn = glm::min(mn, b.center - glm::vec3(b.radius));
|
||
mx = glm::max(mx, b.center + glm::vec3(b.radius));
|
||
}
|
||
m_Bounds.center = (mn + mx) * 0.5f;
|
||
m_Bounds.radius = 0.5f * glm::length(mx - mn);
|
||
}
|
||
|
||
void Model::InitDefaultTextures()
|
||
{
|
||
// 各材质槽位的兜底纹理（1x1，开销可忽略）：
|
||
// albedo 白 / MR 平参数（roughness 0.5, metallic 0）/ emissive 黑
|
||
m_DefaultTextures[Material::Albedo] =
|
||
Texture::CreateSolidColor(m_Device, m_Allocator, 255, 255, 255);
|
||
m_DefaultTextures[Material::MetallicRoughness] =
|
||
Texture::CreateSolidColor(m_Device, m_Allocator, 0, 128, 0, false);
|
||
m_DefaultTextures[Material::Normal] =
|
||
Texture::CreateSolidColor(m_Device, m_Allocator, 128, 128, 255, false);
|
||
m_DefaultTextures[Material::Emissive] =
|
||
Texture::CreateSolidColor(m_Device, m_Allocator, 0, 0, 0);
|
||
m_DefaultTextures[Material::Ao] =
|
||
Texture::CreateSolidColor(m_Device, m_Allocator, 255, 255, 255, false);
|
||
}
|
||
|
||
Model::Model(Device& device, MemoryAllocator& allocator,
|
||
DescriptorManager& descriptorManager,
|
||
vk::DescriptorSetLayout textureLayout,
|
||
ProceduralTag, f32 halfSize, const std::array<u8, 3>& color)
|
||
: m_Device(device)
|
||
, m_Allocator(allocator)
|
||
, m_DescriptorManager(descriptorManager)
|
||
, m_TextureLayout(textureLayout)
|
||
{
|
||
// 纯色 1x1 贴图（albedo 兜底）
|
||
InitDefaultTextures();
|
||
auto tex = Texture::CreateSolidColor(device, allocator, color[0], color[1], color[2]);
|
||
|
||
// XZ 平面，法线 +Y，从上往下看 CCW 绕序
|
||
const f32 h = halfSize;
|
||
std::vector<Vertex> vertices = {
|
||
{{-h, 0.0f, -h}, {0, 1, 0}, {0.0f, 0.0f}},
|
||
{{-h, 0.0f, h}, {0, 1, 0}, {0.0f, 0.0f}},
|
||
{{ h, 0.0f, h}, {0, 1, 0}, {0.0f, 0.0f}},
|
||
{{ h, 0.0f, -h}, {0, 1, 0}, {0.0f, 0.0f}},
|
||
};
|
||
const std::vector<u32> indices = {0, 1, 2, 0, 2, 3};
|
||
const std::vector<std::vector<u32>> lodIndices = {indices};
|
||
|
||
// 材质/纹理生命周期随模型（缓存持有）
|
||
std::string paths[Material::kSlotCount] = {"", "", "", ""};
|
||
Material* mat = GetOrCreateMaterial(paths, MaterialParams{}); // 全默认材质
|
||
|
||
m_Meshes.push_back(CreateScope<Mesh>(
|
||
m_Device, m_Allocator, vertices, lodIndices, mat));
|
||
RebuildUnionBounds();
|
||
}
|
||
|
||
Scope<Model> Model::CreatePlane(Device& device, MemoryAllocator& allocator,
|
||
DescriptorManager& descriptorManager,
|
||
vk::DescriptorSetLayout textureLayout,
|
||
f32 halfSize, u8 r, u8 g, u8 b)
|
||
{
|
||
return Scope<Model>(new Model(device, allocator, descriptorManager,
|
||
textureLayout, ProceduralTag{},
|
||
halfSize, std::array<u8, 3>{r, g, b}));
|
||
}
|
||
|
||
ParsedModel Model::ParseFromDisk(const std::string& path)
|
||
{
|
||
ParsedModel out;
|
||
out.path = path;
|
||
|
||
// UV 翻转按格式区分：
|
||
// OBJ/FBX —— v 原点在底部，需翻到左上惯例（配合"v=0=图像顶"采样）
|
||
// glTF/GLB —— UV 本就是左上原点（v=0=图像顶），翻转会把所有
|
||
// 贴图上下颠倒（法线/MR 全反 = 塑料感元凶）
|
||
const std::string lower = [&] {
|
||
std::string s = path;
|
||
std::transform(s.begin(), s.end(), s.begin(),
|
||
[](unsigned char c) { return std::tolower(c); });
|
||
return s;
|
||
}();
|
||
out.isGltf = lower.size() >= 5 &&
|
||
(lower.rfind(".gltf") == lower.size() - 5 ||
|
||
lower.rfind(".glb") == lower.size() - 4);
|
||
|
||
Assimp::Importer importer;
|
||
const aiScene* scene = importer.ReadFile(path,
|
||
aiProcess_Triangulate |
|
||
(out.isGltf ? 0 : aiProcess_FlipUVs) |
|
||
aiProcess_GenSmoothNormals |
|
||
aiProcess_CalcTangentSpace | // 法线贴图 TBN
|
||
aiProcess_JoinIdenticalVertices);
|
||
|
||
if (!scene || (scene->mFlags & AI_SCENE_FLAGS_INCOMPLETE) || !scene->mRootNode) {
|
||
ENGINE_ERROR("Assimp error: {}", importer.GetErrorString());
|
||
return out; // meshes 空 = 无效
|
||
}
|
||
|
||
const std::filesystem::path modelDir =
|
||
std::filesystem::path(path).parent_path();
|
||
ProcessNode(scene->mRootNode, scene, out.meshes, modelDir, out.isGltf);
|
||
ENGINE_INFO("Model parsed: {} ({} meshes)", path, out.meshes.size());
|
||
return out;
|
||
}
|
||
|
||
Scope<Model> Model::FromParsed(Device& device, MemoryAllocator& allocator,
|
||
DescriptorManager& descriptorManager,
|
||
vk::DescriptorSetLayout textureLayout,
|
||
ParsedModel&& parsed)
|
||
{
|
||
if (!parsed.Valid()) {
|
||
ENGINE_ERROR("Model::FromParsed: invalid parsed model ({})", parsed.path);
|
||
return nullptr;
|
||
}
|
||
|
||
auto model = Scope<Model>(new Model(device, allocator,
|
||
descriptorManager, textureLayout));
|
||
model->m_ModelPath = parsed.path;
|
||
model->m_IsGltf = parsed.isGltf;
|
||
|
||
// 材质缓存去重 + 纹理上传 + mesh 缓冲（全部 GPU，主线程）
|
||
for (auto& pm : parsed.meshes) {
|
||
Material* mat = model->GetOrCreateMaterial(pm.texPaths.data(), pm.params);
|
||
model->m_Meshes.push_back(CreateScope<Mesh>(
|
||
device, allocator, pm.vertices, pm.lods, mat));
|
||
}
|
||
ENGINE_INFO("Model loaded: {} ({} meshes, {} textures)",
|
||
parsed.path, model->m_Meshes.size(), model->m_TextureCache.size());
|
||
model->RebuildUnionBounds();
|
||
return model;
|
||
}
|
||
|
||
void Model::LoadModel(const std::string& path)
|
||
{
|
||
// 同步路径：解析 + GPU 构建串行（与 FromParsed 相同的消费逻辑）
|
||
ParsedModel parsed = ParseFromDisk(path);
|
||
if (!parsed.Valid()) {
|
||
return;
|
||
}
|
||
m_ModelPath = parsed.path;
|
||
m_IsGltf = parsed.isGltf;
|
||
for (auto& pm : parsed.meshes) {
|
||
Material* mat = GetOrCreateMaterial(pm.texPaths.data(), pm.params);
|
||
m_Meshes.push_back(CreateScope<Mesh>(
|
||
m_Device, m_Allocator, pm.vertices, pm.lods, mat));
|
||
}
|
||
ENGINE_INFO("Model loaded: {} ({} meshes, {} textures)",
|
||
path, m_Meshes.size(), m_TextureCache.size());
|
||
RebuildUnionBounds();
|
||
}
|
||
|
||
void Model::ProcessNode(const aiNode* node, const aiScene* scene,
|
||
std::vector<ParsedMesh>& out,
|
||
const std::filesystem::path& modelDir, bool isGltf)
|
||
{
|
||
for (u32 i = 0; i < node->mNumMeshes; ++i) {
|
||
aiMesh* mesh = scene->mMeshes[node->mMeshes[i]];
|
||
out.push_back(ProcessMesh(mesh, scene, modelDir, isGltf));
|
||
}
|
||
for (u32 i = 0; i < node->mNumChildren; ++i) {
|
||
ProcessNode(node->mChildren[i], scene, out, modelDir, isGltf);
|
||
}
|
||
}
|
||
|
||
Texture* Model::GetOrLoadTexture(const std::string& rawPath, bool srgb)
|
||
{
|
||
// FBX/DCC 导出的路径常带反斜杠（Windows 语义），统一成正斜杠
|
||
std::string path = rawPath;
|
||
std::replace(path.begin(), path.end(), '\\', '/');
|
||
|
||
// 缓存键带 sRGB 语义：同一路径可能同时被颜色槽与数据槽引用
|
||
std::string key = (srgb ? "s:" : "l:") + path;
|
||
auto it = m_TextureCache.find(key);
|
||
if (it != m_TextureCache.end()) {
|
||
return it->second.get();
|
||
}
|
||
|
||
// 文件不存在则用兜底纹理
|
||
if (!std::filesystem::exists(path)) {
|
||
ENGINE_WARN("Texture not found, using default: {}", path);
|
||
return nullptr;
|
||
}
|
||
|
||
auto tex = CreateScope<Texture>(m_Device, m_Allocator, path, srgb);
|
||
Texture* ptr = tex.get();
|
||
m_TextureCache.emplace(key, std::move(tex));
|
||
return ptr;
|
||
}
|
||
|
||
Material* Model::GetOrCreateMaterial(const std::string paths[Material::kSlotCount],
|
||
const MaterialParams& params)
|
||
{
|
||
const std::string key = Material::MakeCacheKey(paths, params);
|
||
if (auto it = m_MaterialCache.find(key); it != m_MaterialCache.end()) {
|
||
return it->second.get();
|
||
}
|
||
|
||
// 每槽位解析纹理：空/缺失折到该槽位默认纹理
|
||
std::array<Texture*, Material::kSlotCount> textures{};
|
||
// 槽位 sRGB 语义：颜色图 true，数据图 false
|
||
const bool slotSrgb[Material::kSlotCount] = {true, false, false, true};
|
||
|
||
for (u32 slot = 0; slot < Material::kSlotCount; ++slot) {
|
||
Texture* tex = nullptr;
|
||
if (!paths[slot].empty()) {
|
||
tex = GetOrLoadTexture(paths[slot], slotSrgb[slot]);
|
||
}
|
||
textures[slot] = tex ? tex : m_DefaultTextures[slot].get();
|
||
}
|
||
|
||
auto mat = CreateScope<Material>(
|
||
m_Device, m_Allocator, m_DescriptorManager, m_TextureLayout,
|
||
textures, params);
|
||
Material* ptr = mat.get();
|
||
m_MaterialCache.emplace(key, std::move(mat));
|
||
return ptr;
|
||
}
|
||
|
||
ParsedMesh Model::ProcessMesh(const aiMesh* mesh, const aiScene* scene,
|
||
const std::filesystem::path& modelDir,
|
||
bool isGltf)
|
||
{
|
||
ParsedMesh out;
|
||
std::vector<Vertex>& vertices = out.vertices;
|
||
std::vector<u32> indices;
|
||
|
||
vertices.reserve(mesh->mNumVertices);
|
||
|
||
for (u32 i = 0; i < mesh->mNumVertices; ++i) {
|
||
Vertex vertex{};
|
||
|
||
vertex.position = {
|
||
mesh->mVertices[i].x,
|
||
mesh->mVertices[i].y,
|
||
mesh->mVertices[i].z
|
||
};
|
||
|
||
if (mesh->HasNormals()) {
|
||
vertex.normal = {
|
||
mesh->mNormals[i].x,
|
||
mesh->mNormals[i].y,
|
||
mesh->mNormals[i].z
|
||
};
|
||
}
|
||
|
||
if (mesh->mTextureCoords[0]) {
|
||
vertex.uv = {
|
||
mesh->mTextureCoords[0][i].x,
|
||
mesh->mTextureCoords[0][i].y
|
||
};
|
||
}
|
||
|
||
// 切线（法线贴图 TBN；无切线数据置零，shader 侧退回几何法线）
|
||
if (mesh->mTangents) {
|
||
vertex.tangent = {
|
||
mesh->mTangents[i].x,
|
||
mesh->mTangents[i].y,
|
||
mesh->mTangents[i].z
|
||
};
|
||
}
|
||
|
||
vertices.push_back(vertex);
|
||
}
|
||
|
||
for (u32 i = 0; i < mesh->mNumFaces; ++i) {
|
||
aiFace face = mesh->mFaces[i];
|
||
for (u32 j = 0; j < face.mNumIndices; ++j) {
|
||
indices.push_back(face.mIndices[j]);
|
||
}
|
||
}
|
||
|
||
// 读取材质纹理路径（PBR metallic-roughness 槽位）+ 金属度/粗糙度因子
|
||
std::string texPaths[Material::kSlotCount];
|
||
MaterialParams params;
|
||
|
||
if (mesh->mMaterialIndex >= 0) {
|
||
const aiMaterial* material = scene->mMaterials[mesh->mMaterialIndex];
|
||
|
||
// glTF2 贴图在 assimp 5.x 的落点：MR 合图 → METALNESS(15)/
|
||
// DIFFUSE_ROUGHNESS(16)（同图双槽）；baseColor → BASE_COLOR(12)。
|
||
// OBJ：map_Ns → SHININESS（独立 roughness 图）、map_Ka 常绑 AO。
|
||
// 先到先得，数组顺序即优先级
|
||
const std::pair<aiTextureType, u32> slots[] = {
|
||
{aiTextureType_DIFFUSE, Material::Albedo},
|
||
{aiTextureType_BASE_COLOR, Material::Albedo},
|
||
{aiTextureType_NORMALS, Material::Normal},
|
||
{aiTextureType_HEIGHT, Material::Normal}, // OBJ map_Bump 落点
|
||
{aiTextureType_METALNESS, Material::MetallicRoughness},
|
||
{aiTextureType_DIFFUSE_ROUGHNESS, Material::MetallicRoughness},
|
||
{aiTextureType_UNKNOWN, Material::MetallicRoughness},
|
||
{aiTextureType_SHININESS, Material::MetallicRoughness},
|
||
{aiTextureType_AMBIENT_OCCLUSION, Material::Ao},
|
||
{aiTextureType_LIGHTMAP, Material::Ao},
|
||
{aiTextureType_AMBIENT, Material::Ao},
|
||
{aiTextureType_EMISSIVE, Material::Emissive},
|
||
{aiTextureType_EMISSION_COLOR, Material::Emissive},
|
||
};
|
||
|
||
for (const auto& [type, slot] : slots) {
|
||
if (!texPaths[slot].empty()) continue; // 高优先级已命中
|
||
aiString path;
|
||
if (material->GetTexture(type, 0, &path) == AI_SUCCESS &&
|
||
path.length > 0) {
|
||
texPaths[slot] = (modelDir / path.C_Str()).string();
|
||
}
|
||
}
|
||
// 注：不做文件名探测/自动补槽——资产缺什么就是什么；
|
||
// 补槽是将来手动材质编辑的职责，加载期保持确定性
|
||
|
||
// 材质加载可见性：贴图槽缺失/路径解析错误会静默退化成默认
|
||
// 纹理（金属全没 → 塑料感），这里留下诊断锚点
|
||
ENGINE_INFO("Material '{}': albedo={} normal={} MR={} emissive={} ao={}",
|
||
material->GetName().C_Str(),
|
||
texPaths[Material::Albedo].empty() ? "<none>"
|
||
: texPaths[Material::Albedo],
|
||
texPaths[Material::Normal].empty() ? "<none>"
|
||
: texPaths[Material::Normal],
|
||
texPaths[Material::MetallicRoughness].empty() ? "<none>"
|
||
: texPaths[Material::MetallicRoughness],
|
||
texPaths[Material::Emissive].empty() ? "<none>"
|
||
: texPaths[Material::Emissive],
|
||
texPaths[Material::Ao].empty() ? "<none>"
|
||
: texPaths[Material::Ao]);
|
||
|
||
// 因子按工作流分支：
|
||
// glTF2：MR 合图（B=金属度语义），因子按声明（缺省 1/1）
|
||
// OBJ：specular 工作流无金属概念 → metallic=0；
|
||
// roughness 有独立贴图 → 因子 1（G 通道直达）；
|
||
// 无贴图 → Blinn-Phong Ns 换算 R=sqrt(2/(Ns+2))，
|
||
// 补偿默认 MR 贴图 G=0.5 → 因子 = 2R
|
||
if (isGltf) {
|
||
ai_real metallic = 1.0f, roughness = 1.0f;
|
||
material->Get(AI_MATKEY_METALLIC_FACTOR, metallic);
|
||
material->Get(AI_MATKEY_ROUGHNESS_FACTOR, roughness);
|
||
params.factors = glm::vec4(static_cast<f32>(metallic),
|
||
static_cast<f32>(roughness), 0.0f, 0.0f);
|
||
} else {
|
||
f32 roughnessFactor = 1.0f;
|
||
if (texPaths[Material::MetallicRoughness].empty()) {
|
||
ai_real shininess = 10.0f;
|
||
material->Get(AI_MATKEY_SHININESS, shininess);
|
||
const f32 rough =
|
||
std::sqrt(2.0f / (static_cast<f32>(shininess) + 2.0f));
|
||
roughnessFactor = 2.0f * glm::clamp(rough, 0.05f, 1.0f);
|
||
}
|
||
params.factors = glm::vec4(0.0f, roughnessFactor, 0.0f, 0.0f);
|
||
}
|
||
}
|
||
|
||
// 材质创建延迟到 GPU 构建段（FromParsed）：解析期只留引用
|
||
for (u32 slot = 0; slot < Material::kSlotCount; ++slot) {
|
||
out.texPaths[slot] = std::move(texPaths[slot]);
|
||
}
|
||
out.params = params;
|
||
|
||
// ---- LOD 链（解析期生成，worker 线程免费时间）：----
|
||
// LOD0 = 原始；每级目标索引数减半，相对误差 1e-2。顶点共享，
|
||
// 简化只重排索引（meshopt_simplify 从 LOD0 单次起点逐级向下）
|
||
out.lods.push_back(std::move(indices));
|
||
for (u32 lod = 1; lod < kMeshLodCount; ++lod) {
|
||
const size_t target = out.lods[0].size() >> lod;
|
||
// 太小的网格不简化：简化收益低于误差风险，直接复用上一级
|
||
if (target < 96 || out.lods[0].size() < 192) {
|
||
out.lods.push_back(out.lods.back());
|
||
continue;
|
||
}
|
||
std::vector<u32> dest(out.lods[0].size());
|
||
float error = 0.0f;
|
||
const size_t count = meshopt_simplify(
|
||
dest.data(),
|
||
out.lods[0].data(), out.lods[0].size(),
|
||
&out.vertices[0].position.x, out.vertices.size(), sizeof(Vertex),
|
||
target, 1e-2f, 0, &error);
|
||
if (count == 0) {
|
||
out.lods.push_back(out.lods.back()); // 简化失败兜底
|
||
continue;
|
||
}
|
||
dest.resize(count);
|
||
out.lods.push_back(std::move(dest));
|
||
}
|
||
return out;
|
||
}
|
||
|
||
} // namespace Engine
|