feat: ACES tonemap + Bloom 后处理链（Extract/BlurH/BlurV 半分辨率）

- ACES(Narkowicz) 经 push 开关启用，exposure 接 ImGui 滑条
- RG: TextureDesc 支持 sizeScale 相对尺寸；per-pass 渲染尺寸解析
  （半分辨率 pass 的 viewport 由此决定）
- RG Toposort 重写：RAW+WAR+WAW 按声明序逐资源建边，支持多写者
  ping-pong 资源（旧逻辑对 reader 连接所有 writer，ping-pong 成环）
- GetPhysicalView 缺物理镜像时 throw（ENGINE_FATAL 只记日志不中止）
- 全屏 pass 通用 16B push + 公共录制 lambda，Extract/Blur/Tonemap 复用

App/main.cpp

@@ -15,6 +15,7 @@
 #include <Engine/ECS/EntityHandle.h>
 #include <Engine/ECS/Components.h>
 #include <Engine/Assets/AssetManager.h>
+#include <Engine/Renderer/Renderer.h>
 
 #include <imgui.h>
 
@@ -130,6 +131,29 @@ protected:
             }
         }
 
+        // Tonemap + Bloom 参数（HDR 后处理输出端）
+        auto& renderer = GetRenderer();
+        ImGui::Separator();
+        ImGui::Text("Tonemap");
+        f32 exposure = renderer.GetExposure();
+        if (ImGui::SliderFloat("exposure", &exposure, 0.1f, 4.0f)) {
+            renderer.SetExposure(exposure);
+        }
+        bool aces = renderer.GetUseACES();
+        if (ImGui::Checkbox("ACES", &aces)) {
+            renderer.SetUseACES(aces);
+        }
+
+        ImGui::Text("Bloom");
+        f32 threshold = renderer.GetBloomThreshold();
+        if (ImGui::SliderFloat("threshold", &threshold, 0.0f, 2.0f)) {
+            renderer.SetBloomThreshold(threshold);
+        }
+        f32 strength = renderer.GetBloomStrength();
+        if (ImGui::SliderFloat("strength", &strength, 0.0f, 2.0f)) {
+            renderer.SetBloomStrength(strength);
+        }
+
         ImGui::Separator();
         ImGui::Text("fps: %.1f  frame: %llu",
                     1.0f / GetDeltaTime(),

Engine/include/Engine/RenderGraph/RenderGraph.h

@@ -70,8 +70,9 @@ namespace Engine::RG {
 
     struct TextureDesc {
         vk::Format format = vk::Format::eUndefined;
-        u32 width = 0;   // 0 = 跟随目标（swapchain）尺寸
+        u32 width = 0;   // 0 = 跟随目标（swapchain）尺寸 × sizeScale
         u32 height = 0;
+        f32 sizeScale = 1.0f;   // 相对尺寸（后处理降采样用，如 bloom 0.5）
     };
 
     // ---------- 物理镜像 ----------
@@ -154,6 +155,7 @@ namespace Engine::RG {
         ExecuteFn exec;
         bool sideEffect = false;              // 无读写但必须执行（compute 写 buffer 等）
         bool culled = false;
+        vk::Extent2D extent{};                // 本 pass 的渲染尺寸（写入附件的最大分辨率，Compile 期解析）
 
         // Compile 期预算好的 barrier：每个资源两个变体
         //   steady = 该槽位已被使用过（从 resting 状态出发）

Engine/include/Engine/Renderer/Renderer.h

@@ -64,7 +64,7 @@ namespace Engine {
         void BuildGraph();          // 声明 pass 结构 + 首次 Compile
         void HandleResize();
         void UpdateShadowDescriptors();
-        void UpdateTonemapDescriptors();
+        void UpdateFullscreenDescriptors();
 
         Window& m_Window;
 
@@ -97,7 +97,8 @@ namespace Engine {
         vk::ShaderEXT m_ShadowVertShader{};
         vk::ShaderEXT m_ShadowFragShader{};
 
-        // Tonemap pass 装配：Set0 = HDR 颜色图，push = {exposure}（fragment）
+        // Tonemap pass 装配：Set0 = {HDR 颜色图, Bloom 图}，push = 16B
+        //（全屏 pass 通用：exposure/aces/bloomStrength/pad）
         vk::PipelineLayout m_TonemapPipelineLayout;
         vk::DescriptorSetLayout m_TonemapSetLayout;
         std::vector<vk::DescriptorSet> m_TonemapDescriptorSets; // 每帧槽位一个
@@ -105,11 +106,21 @@ namespace Engine {
         vk::ShaderEXT m_TonemapVertShader{};
         vk::ShaderEXT m_TonemapFragShader{};
 
+        // Bloom：Extract(HDR→A) → BlurH(A→B) → BlurV(B→A)，半分辨率
+        // 与 Tonemap 共用 set layout/pipeline layout/vertex shader
+        std::vector<vk::DescriptorSet> m_BloomExtractSets;  // 每帧槽位
+        std::vector<vk::DescriptorSet> m_BloomBlurHSets;
+        std::vector<vk::DescriptorSet> m_BloomBlurVSets;
+        vk::ShaderEXT m_BloomExtractFrag{};
+        vk::ShaderEXT m_BloomBlurFrag{};
+
         // ---- RenderGraph 资源句柄 ----
         RG::TextureHandle m_Backbuffer{};
         RG::TextureHandle m_Depth{};
         RG::TextureHandle m_ShadowMap{};
         RG::TextureHandle m_HDRColor{};   // Forward 输出（线性 HDR），Tonemap 归一
+        RG::TextureHandle m_BloomA{};     // 半分辨率：提取 + 垂直模糊结果
+        RG::TextureHandle m_BloomB{};     // 半分辨率：水平模糊中间结果
 
         // ---- 当帧绑定（Render() 期间有效，供 pass 回调读取）----
         ECS::Scene* m_BoundScene = nullptr;
@@ -119,6 +130,24 @@ namespace Engine {
         // 可选叠加层（不拥有；Application 持有）
         ImGuiLayer* m_ImGuiLayer = nullptr;
 
+        // ---- Tonemap/Bloom 参数（ImGui 可调）----
+        f32 m_Exposure = 1.0f;
+        bool m_UseACES = true;
+        f32 m_BloomThreshold = 1.0f;   // HDR 域亮度阈值
+        f32 m_BloomStrength = 0.8f;    // 叠加强度
+
+    public:
+        f32 GetExposure() const { return m_Exposure; }
+        void SetExposure(f32 e) { m_Exposure = e; }
+        bool GetUseACES() const { return m_UseACES; }
+        void SetUseACES(bool b) { m_UseACES = b; }
+        f32 GetBloomThreshold() const { return m_BloomThreshold; }
+        void SetBloomThreshold(f32 v) { m_BloomThreshold = v; }
+        f32 GetBloomStrength() const { return m_BloomStrength; }
+        void SetBloomStrength(f32 v) { m_BloomStrength = v; }
+
+    private:
+
         vk::Extent2D m_Extent{};
         u32 m_CurrentFrame = 0;
     };

Engine/src/RenderGraph/RenderGraph.cpp

@@ -8,6 +8,7 @@
 
 #include <algorithm>
 #include <set>
+#include <stdexcept>
 
 namespace Engine::RG {
 
@@ -164,6 +165,22 @@ namespace Engine::RG {
 
         for (auto& entry : m_Textures) ResolveUsage(entry);
 
+        // 每 pass 渲染尺寸 = 其写入附件的最大分辨率（半分辨率后处理的
+        // viewport/渲染区域由此决定）；无写入（如 Present）回退目标尺寸
+        for (auto& pass : m_Passes) {
+            u32 w = 0, h = 0;
+            for (const auto& use : pass.uses) {
+                if (!IsWriteUse(use.use)) continue;
+                const auto& d = m_Textures[use.resIndex].desc;
+                w = std::max(w, d.width ? d.width
+                                        : std::max(1u, u32(m_TargetExtent.width * d.sizeScale)));
+                h = std::max(h, d.height ? d.height
+                                         : std::max(1u, u32(m_TargetExtent.height * d.sizeScale)));
+            }
+            pass.extent = vk::Extent2D(w ? w : m_TargetExtent.width,
+                                       h ? h : m_TargetExtent.height);
+        }
+
         CullPasses();
         Toposort();
         ComputeLifetimes();
@@ -272,34 +289,43 @@ namespace Engine::RG {
 
     void RenderGraph::Toposort()
     {
-        // writer → reader 建边（去重），Kahn 拓扑排序。
+        // 按资源在声明序上的使用序列建边，Kahn 拓扑排序。
-        // 声明顺序通常已是拓扑序，但引擎级实现必须显式校验，
+        //   RAW：reader → 依赖声明序上它的上一个 writer
-        // 避免上层重构 pass 顺序时引入环。
+        //   WAR：writer → 依赖上一次 write 之后出现过的所有 reader
+        //   WAW：writer → 上一个 writer
+        // 多写者（ping-pong 资源）合法：若对 reader 连接"所有"
+        // writer（含声明序在其后的），BlurH 会同时收到 Extract→BlurH 与
+        // BlurV→BlurH，而 BlurH→BlurV 又存在，直接成环。
+        // 声明顺序通常已是拓扑序，但引擎级实现必须显式校验。
 
-        std::vector<std::vector<u32>> writers(m_Textures.size());
+        const u32 N = static_cast<u32>(m_Passes.size());
-        for (u32 i = 0; i < m_Passes.size(); ++i) {
+        std::vector<std::set<u32>> adj(N);
+        std::vector<u32> indegree(N, 0);
+
+        auto addEdge = [&](u32 from, u32 to) {
+            if (from == to) return;
+            if (adj[from].insert(to).second) ++indegree[to];
+        };
+
+        constexpr u32 kNoWriter = UINT32_MAX;
+        std::vector<u32> lastWriter(m_Textures.size(), kNoWriter);
+        std::vector<std::set<u32>> pendingReaders(m_Textures.size());
+
+        for (u32 i = 0; i < N; ++i) {
             if (m_Passes[i].culled) continue;
+
             for (const auto& use : m_Passes[i].uses) {
+                auto& lw = lastWriter[use.resIndex];
+                auto& readers = pendingReaders[use.resIndex];
+
                 if (IsWriteUse(use.use)) {
-                    writers[use.resIndex].push_back(i);
+                    for (u32 r : readers) addEdge(r, i);  // WAR
-                }
+                    readers.clear();
-            }
+                    if (lw != kNoWriter) addEdge(lw, i);  // WAW
-        }
+                    lw = i;
-
+                } else {
-        std::vector<std::set<u32>> adj(m_Passes.size());
+                    if (lw != kNoWriter) addEdge(lw, i);  // RAW
-        std::vector<u32> indegree(m_Passes.size(), 0);
+                    readers.insert(i);
-
-        for (u32 q = 0; q < m_Passes.size(); ++q) {
-            if (m_Passes[q].culled) continue;
-            std::set<u32> seenWriters;
-            for (const auto& use : m_Passes[q].uses) {
-                if (IsWriteUse(use.use)) continue;
-                for (u32 p : writers[use.resIndex]) {
-                    if (p == q || m_Passes[p].culled) continue;
-                    if (seenWriters.insert(p).second) {
-                        adj[p].insert(q);
-                        ++indegree[q];
-                    }
                 }
             }
         }
@@ -309,7 +335,7 @@ namespace Engine::RG {
 
         // 同层按声明序出队，保证确定性
         std::set<u32> ready;
-        for (u32 i = 0; i < m_Passes.size(); ++i) {
+        for (u32 i = 0; i < N; ++i) {
             if (!m_Passes[i].culled && indegree[i] == 0) ready.insert(i);
         }
 
@@ -458,9 +484,14 @@ namespace Engine::RG {
         for (auto& entry : m_Textures) {
             if (entry.imported || entry.firstPass < 0) continue;
 
+            // 显式尺寸优先；否则 = 目标尺寸 × sizeScale（后处理降采样）
             vk::Extent2D extent{
-                entry.desc.width ? entry.desc.width : m_TargetExtent.width,
+                entry.desc.width
-                entry.desc.height ? entry.desc.height : m_TargetExtent.height
+                    ? entry.desc.width
+                    : std::max(1u, u32(m_TargetExtent.width * entry.desc.sizeScale)),
+                entry.desc.height
+                    ? entry.desc.height
+                    : std::max(1u, u32(m_TargetExtent.height * entry.desc.sizeScale))
             };
 
             for (u32 slot = 0; slot < m_FramesInFlight; ++slot) {
@@ -567,8 +598,11 @@ namespace Engine::RG {
         const auto& entry = GetEntry(h);
         if (entry.imported) return entry.importedView;
         if (frameSlot >= entry.physical.size() || !entry.physical[frameSlot]) {
-            ENGINE_FATAL("RenderGraph: no physical image for '{}' slot {} "
+            // ENGINE_FATAL 只记日志不中止，这里必须 throw 才能拦住空指针
-                         "(compile first?)", entry.name, frameSlot);
+            //（Application::Run 的 catch 会接住并干净退出）
+            throw std::runtime_error(
+                "RenderGraph: no physical image for '" + entry.name +
+                "' slot " + std::to_string(frameSlot));
         }
         return entry.physical[frameSlot]->view;
     }
@@ -614,7 +648,7 @@ namespace Engine::RG {
                 RenderGraphContext ctx;
                 ctx.cmd = frame.cmd;
                 ctx.frame = frame;
-                ctx.extent = m_TargetExtent;
+                ctx.extent = pass.extent;
                 ctx.graph = this;
                 pass.exec(ctx);
             }

Engine/src/Renderer/Renderer.cpp

@@ -56,6 +56,16 @@ namespace Engine
         constexpr u32 kShadowMapSize = 2048;
         constexpr u32 kShadowPushSize = static_cast<u32>(2 * sizeof(glm::mat4));
 
+        // 全屏 pass 通用 push（16B）：所有全屏 pass（Tonemap/Bloom 系）
+        // 统一声明与推送，规避 VUID-08878 的"range 定义必须一致"
+        struct FullscreenPush {
+            f32 exposure;
+            f32 param0;   // 各 pass 自定语义（tonemap: aces 开关）
+            f32 param1;
+            f32 param2;
+        };
+        constexpr u32 kFullscreenPushSize = static_cast<u32>(sizeof(FullscreenPush));
+
         // 光源正交包围盒（覆盖当前场景；后续按场景 AABB 自适应）
         constexpr f32 kLightOrthoExtent = 15.0f;
         constexpr f32 kLightDistance = 30.0f;
@@ -147,16 +157,19 @@ namespace Engine
                          .setCompareEnable(VK_FALSE);
         m_ShadowSampler = m_Device->GetDevice().createSampler(shadowSamplerInfo);
 
-        // 7.5 Tonemap pass 装配：Set0 = HDR 颜色图，push = {exposure}
+        // 7.5 全屏后处理装配（Tonemap + Bloom 共用一套 layout/sampler）：
-        std::vector<DescriptorBinding> tonemapBindings = {
+        // Set0 = {HDR 颜色图, Bloom 图}，push = 16B 通用结构
+        std::vector<DescriptorBinding> fullscreenBindings = {
             {0, vk::DescriptorType::eCombinedImageSampler,
              vk::ShaderStageFlagBits::eFragment},
+            {1, vk::DescriptorType::eCombinedImageSampler,
+             vk::ShaderStageFlagBits::eFragment},
         };
-        m_TonemapSetLayout = m_DescriptorManager->CreateLayout(tonemapBindings);
+        m_TonemapSetLayout = m_DescriptorManager->CreateLayout(fullscreenBindings);
 
         m_TonemapPipelineLayout = m_DescriptorManager->CreatePipelineLayout(
             std::vector<vk::DescriptorSetLayout>{m_TonemapSetLayout},
-            sizeof(f32));
+            kFullscreenPushSize);
 
         m_TonemapSampler = m_Device->GetDevice().createSampler(
             vk::SamplerCreateInfo{}
@@ -170,17 +183,32 @@ namespace Engine
         auto tonemapVert = m_ShaderManager->CreateFromFile(
             shaderDir + "tonemap.vert.spv",
             vk::ShaderStageFlagBits::eVertex,
-            std::vector<vk::DescriptorSetLayout>{m_TonemapSetLayout}, sizeof(f32));
+            std::vector<vk::DescriptorSetLayout>{m_TonemapSetLayout},
+            kFullscreenPushSize);
         auto tonemapFrag = m_ShaderManager->CreateFromFile(
             shaderDir + "tonemap.frag.spv",
             vk::ShaderStageFlagBits::eFragment,
-            std::vector<vk::DescriptorSetLayout>{m_TonemapSetLayout}, sizeof(f32));
+            std::vector<vk::DescriptorSetLayout>{m_TonemapSetLayout},
+            kFullscreenPushSize);
+        auto bloomExtractFrag = m_ShaderManager->CreateFromFile(
+            shaderDir + "bloom_extract.frag.spv",
+            vk::ShaderStageFlagBits::eFragment,
+            std::vector<vk::DescriptorSetLayout>{m_TonemapSetLayout},
+            kFullscreenPushSize);
+        auto bloomBlurFrag = m_ShaderManager->CreateFromFile(
+            shaderDir + "bloom_blur.frag.spv",
+            vk::ShaderStageFlagBits::eFragment,
+            std::vector<vk::DescriptorSetLayout>{m_TonemapSetLayout},
+            kFullscreenPushSize);
 
         m_TonemapVertShader = tonemapVert.Handle;
         m_TonemapFragShader = tonemapFrag.Handle;
+        m_BloomExtractFrag = bloomExtractFrag.Handle;
+        m_BloomBlurFrag = bloomBlurFrag.Handle;
 
-        if (!m_TonemapVertShader || !m_TonemapFragShader) {
+        if (!m_TonemapVertShader || !m_TonemapFragShader ||
-            ENGINE_FATAL("Failed to create tonemap shader objects");
+            !m_BloomExtractFrag || !m_BloomBlurFrag) {
+            ENGINE_FATAL("Failed to create fullscreen shader objects");
         }
 
         // 8. Swapchain（只管呈现；深度图由 RenderGraph 创建）
@@ -199,6 +227,9 @@ namespace Engine
         m_UBODescriptorSets.resize(kMaxFramesInFlight);
         m_ShadowDescriptorSets.resize(kMaxFramesInFlight);
         m_TonemapDescriptorSets.resize(kMaxFramesInFlight);
+        m_BloomExtractSets.resize(kMaxFramesInFlight);
+        m_BloomBlurHSets.resize(kMaxFramesInFlight);
+        m_BloomBlurVSets.resize(kMaxFramesInFlight);
         for (u32 slot = 0; slot < kMaxFramesInFlight; ++slot) {
             m_UBODescriptorSets[slot] =
                 m_DescriptorManager->AllocateDescriptorSet(m_UBOSetLayout);
@@ -211,6 +242,13 @@ namespace Engine
 
             m_TonemapDescriptorSets[slot] =
                 m_DescriptorManager->AllocateDescriptorSet(m_TonemapSetLayout);
+
+            m_BloomExtractSets[slot] =
+                m_DescriptorManager->AllocateDescriptorSet(m_TonemapSetLayout);
+            m_BloomBlurHSets[slot] =
+                m_DescriptorManager->AllocateDescriptorSet(m_TonemapSetLayout);
+            m_BloomBlurVSets[slot] =
+                m_DescriptorManager->AllocateDescriptorSet(m_TonemapSetLayout);
         }
 
         // 11. RenderGraph：声明 + 编译（场景在 Render(Scene&) 时按帧绑定）
@@ -218,7 +256,7 @@ namespace Engine
 
         // Compile 已分配物理镜像：把各槽位的阴影图/HDR view 绑进描述符
         UpdateShadowDescriptors();
-        UpdateTonemapDescriptors();
+        UpdateFullscreenDescriptors();
 
         ENGINE_INFO("Renderer initialized");
     }
@@ -247,6 +285,12 @@ namespace Engine
         if (m_TonemapFragShader) {
             vkDevice.destroyShaderEXT(m_TonemapFragShader);
         }
+        if (m_BloomExtractFrag) {
+            vkDevice.destroyShaderEXT(m_BloomExtractFrag);
+        }
+        if (m_BloomBlurFrag) {
+            vkDevice.destroyShaderEXT(m_BloomBlurFrag);
+        }
 
         vkDevice.destroySampler(m_ShadowSampler);
         vkDevice.destroySampler(m_TonemapSampler);
@@ -300,6 +344,14 @@ namespace Engine
             "HDRColor",
             RG::TextureDesc{vk::Format::eR16G16B16A16Sfloat});
 
+        // Bloom ping-pong：半分辨率（sizeScale 相对目标尺寸，resize 自动跟随）
+        m_BloomA = m_RenderGraph->CreateTexture(
+            "BloomA",
+            RG::TextureDesc{vk::Format::eR16G16B16A16Sfloat, 0, 0, 0.5f});
+        m_BloomB = m_RenderGraph->CreateTexture(
+            "BloomB",
+            RG::TextureDesc{vk::Format::eR16G16B16A16Sfloat, 0, 0, 0.5f});
+
         // ---- Shadow pass：光源视角写深度（所有渲染体都写入，不做实体级开关——
         //      castShadow 应由材质/实体数据驱动剔除，而非 pass 逻辑特判）----
         m_RenderGraph->AddPass("Shadow",
@@ -527,11 +579,12 @@ namespace Engine
             .Read(m_ShadowMap, RG::ResourceUse::ShaderRead,
                   vk::ShaderStageFlagBits::eFragment);
 
-        // ---- Tonemap pass：HDR → 显示空间，全屏三角无顶点缓冲 ----
+        // ---- 全屏 pass 公共录制：动态状态/draw 全一致，仅附件 view、
-        // 显示变换在此显式建模（此前隐式依赖 Srgb 附件，是设计欠账）。
+        //      输入 set、push 值不同（Extract/BlurH/BlurV/Tonemap 复用）----
-        // Backbuffer 从此只接收"显示空间"颜色：tonemap 直接写、ImGui 叠加。
+        auto recordFullscreen =
-        m_RenderGraph->AddPass("Tonemap",
+            [this](RG::RenderGraphContext& ctx, vk::ImageView view,
-            [this](RG::RenderGraphContext& ctx) {
+                   vk::ShaderEXT frag, const vk::DescriptorSet& set,
+                   const FullscreenPush& push) {
                 const vk::CommandBuffer cmd = ctx.cmd;
 
                 vk::ClearValue clearValue;
@@ -539,7 +592,7 @@ namespace Engine
                     std::array<float, 4>{0.0f, 0.0f, 0.0f, 1.0f});
 
                 vk::RenderingAttachmentInfo colorAttachment;
-                colorAttachment.setImageView(ctx.View(m_Backbuffer))
+                colorAttachment.setImageView(view)
                                .setImageLayout(vk::ImageLayout::eColorAttachmentOptimal)
                                .setLoadOp(vk::AttachmentLoadOp::eClear)
                                .setStoreOp(vk::AttachmentStoreOp::eStore)
@@ -557,12 +610,11 @@ namespace Engine
                     vk::ShaderStageFlagBits::eFragment
                 };
                 std::array<vk::ShaderEXT, 2> shaders = {
-                    m_TonemapVertShader, m_TonemapFragShader};
+                    m_TonemapVertShader, frag};
                 cmd.bindShadersEXT(stages, shaders);
 
                 cmd.bindDescriptorSets(vk::PipelineBindPoint::eGraphics,
-                                       m_TonemapPipelineLayout, 0,
+                                       m_TonemapPipelineLayout, 0, set, {});
-                                       m_TonemapDescriptorSets[ctx.frame.frameSlot], {});
 
                 // ---- 动态状态（无深度附件、无顶点输入）----
                 cmd.setRasterizerDiscardEnable(VK_FALSE);
@@ -592,17 +644,65 @@ namespace Engine
                 cmd.setViewportWithCount(viewport);
                 cmd.setScissorWithCount(vk::Rect2D({0, 0}, ctx.extent));
 
-                const f32 exposure = 1.0f;  // TODO: 接 ImGui 曝光滑条
                 cmd.pushConstants(m_TonemapPipelineLayout,
                                   vk::ShaderStageFlagBits::eVertex |
                                   vk::ShaderStageFlagBits::eFragment,
-                                  0, sizeof(f32), &exposure);
+                                  0, kFullscreenPushSize, &push);
 
                 cmd.draw(3, 1, 0, 0);
                 cmd.endRendering();
+            };
+
+        // ---- Bloom：Extract（HDR 软阈值）→ BlurH → BlurV，全部半分辨率 ----
+        m_RenderGraph->AddPass("BloomExtract",
+            [this, recordFullscreen](RG::RenderGraphContext& ctx) {
+                recordFullscreen(ctx, ctx.View(m_BloomA), m_BloomExtractFrag,
+                                 m_BloomExtractSets[ctx.frame.frameSlot],
+                                 FullscreenPush{0.0f, m_BloomThreshold, 0.0f, 0.0f});
+            })
+            .Read(m_HDRColor, RG::ResourceUse::ShaderRead,
+                  vk::ShaderStageFlagBits::eFragment)
+            .Write(m_BloomA, RG::ResourceUse::ColorAttachment);
+
+        m_RenderGraph->AddPass("BloomBlurH",
+            [this, recordFullscreen](RG::RenderGraphContext& ctx) {
+                recordFullscreen(ctx, ctx.View(m_BloomB), m_BloomBlurFrag,
+                                 m_BloomBlurHSets[ctx.frame.frameSlot],
+                                 FullscreenPush{0.0f,
+                                                1.0f / ctx.extent.width, 0.0f,
+                                                0.0f});
+            })
+            .Read(m_BloomA, RG::ResourceUse::ShaderRead,
+                  vk::ShaderStageFlagBits::eFragment)
+            .Write(m_BloomB, RG::ResourceUse::ColorAttachment);
+
+        m_RenderGraph->AddPass("BloomBlurV",
+            [this, recordFullscreen](RG::RenderGraphContext& ctx) {
+                recordFullscreen(ctx, ctx.View(m_BloomA), m_BloomBlurFrag,
+                                 m_BloomBlurVSets[ctx.frame.frameSlot],
+                                 FullscreenPush{0.0f, 0.0f,
+                                                1.0f / ctx.extent.height,
+                                                0.0f});
+            })
+            .Read(m_BloomB, RG::ResourceUse::ShaderRead,
+                  vk::ShaderStageFlagBits::eFragment)
+            .Write(m_BloomA, RG::ResourceUse::ColorAttachment);
+
+        // ---- Tonemap pass：HDR（+bloom）→ 显示空间，全屏三角无顶点缓冲 ----
+        // 显示变换在此显式建模（此前隐式依赖 Srgb 附件，是设计欠账）。
+        // Backbuffer 从此只接收"显示空间"颜色：tonemap 直接写、ImGui 叠加。
+        m_RenderGraph->AddPass("Tonemap",
+            [this, recordFullscreen](RG::RenderGraphContext& ctx) {
+                recordFullscreen(ctx, ctx.View(m_Backbuffer), m_TonemapFragShader,
+                                 m_TonemapDescriptorSets[ctx.frame.frameSlot],
+                                 FullscreenPush{m_Exposure,
+                                                m_UseACES ? 1.0f : 0.0f,
+                                                m_BloomStrength, 0.0f});
             })
             .Write(m_Backbuffer, RG::ResourceUse::ColorAttachment)
             .Read(m_HDRColor, RG::ResourceUse::ShaderRead,
+                  vk::ShaderStageFlagBits::eFragment)
+            .Read(m_BloomA, RG::ResourceUse::ShaderRead,
                   vk::ShaderStageFlagBits::eFragment);
 
         // ---- ImGui pass：叠加层（loadOp=Load，不清屏）----
@@ -661,7 +761,7 @@ namespace Engine
 
         // 重编译后物理镜像可能变化，描述符重新绑定
         UpdateShadowDescriptors();
-        UpdateTonemapDescriptors();
+        UpdateFullscreenDescriptors();
     }
 
     void Renderer::UpdateShadowDescriptors()
@@ -675,13 +775,32 @@ namespace Engine
         }
     }
 
-    void Renderer::UpdateTonemapDescriptors()
+    void Renderer::UpdateFullscreenDescriptors()
     {
+        // 全屏 pass 的输入绑定（per-slot 物理镜像）：
+        //   Tonemap   = {0: HDRColor, 1: BloomA}
+        //   Extract   = {0: HDRColor}        （binding1 不用，可不写）
+        //   BlurH     = {0: BloomA}
+        //   BlurV     = {0: BloomB}
         for (u32 slot = 0; slot < kMaxFramesInFlight; ++slot) {
+            const vk::ImageView hdr =
+                m_RenderGraph->GetPhysicalView(m_HDRColor, slot);
+            const vk::ImageView bloomA =
+                m_RenderGraph->GetPhysicalView(m_BloomA, slot);
+            const vk::ImageView bloomB =
+                m_RenderGraph->GetPhysicalView(m_BloomB, slot);
+
             m_DescriptorManager->UpdateTexture(
-                m_TonemapDescriptorSets[slot], 0,
+                m_TonemapDescriptorSets[slot], 0, m_TonemapSampler, hdr);
-                m_TonemapSampler,
+            m_DescriptorManager->UpdateTexture(
-                m_RenderGraph->GetPhysicalView(m_HDRColor, slot));
+                m_TonemapDescriptorSets[slot], 1, m_TonemapSampler, bloomA);
+
+            m_DescriptorManager->UpdateTexture(
+                m_BloomExtractSets[slot], 0, m_TonemapSampler, hdr);
+            m_DescriptorManager->UpdateTexture(
+                m_BloomBlurHSets[slot], 0, m_TonemapSampler, bloomA);
+            m_DescriptorManager->UpdateTexture(
+                m_BloomBlurVSets[slot], 0, m_TonemapSampler, bloomB);
         }
     }
 

shaders/bloom_blur.frag

@@ -0,0 +1,29 @@
+#version 450
+
+// Bloom 高斯模糊：可分离 9-tap，方向由 push 给出（H/V 各跑一次）。
+// 源图为半分辨率，线性采样 + 权重对应的偏移天然抗锯齿。
+
+layout(location = 0) in vec2 fragUV;
+layout(location = 0) out vec4 outColor;
+
+layout(set = 0, binding = 0) uniform sampler2D srcImage;
+
+layout(push_constant) uniform PushConstants {
+    float exposure;   // 未用（程序级统一布局）
+    float stepX;      // param0: UV 步长 x（水平 pass = 1/width，垂直 = 0）
+    float stepY;      // param1: UV 步长 y
+    float _pad0;
+} pc;
+
+const float weights[5] = float[](0.227027, 0.1945946, 0.1216216, 0.054054, 0.016216);
+
+void main() {
+    vec2 off = vec2(pc.stepX, pc.stepY);
+
+    vec3 c = texture(srcImage, fragUV).rgb * weights[0];
+    for (int i = 1; i < 5; ++i) {
+        c += texture(srcImage, fragUV + off * float(i)).rgb * weights[i];
+        c += texture(srcImage, fragUV - off * float(i)).rgb * weights[i];
+    }
+    outColor = vec4(c, 1.0);
+}

shaders/bloom_extract.frag

@@ -0,0 +1,25 @@
+#version 450
+
+// Bloom 提取：从 HDR 颜色图中滤出超过阈值的部分（软阈值，保留超出比例）。
+// 全屏 pass 通用 push（16B）：param0 = 亮度阈值（HDR 域，1.0 = 仅高光溢出）。
+
+layout(location = 0) in vec2 fragUV;
+layout(location = 0) out vec4 outColor;
+
+layout(set = 0, binding = 0) uniform sampler2D hdrImage;
+
+layout(push_constant) uniform PushConstants {
+    float exposure;   // 未用（程序级统一布局）
+    float threshold;  // param0
+    float _pad0;
+    float _pad1;
+} pc;
+
+void main() {
+    vec3 c = texture(hdrImage, fragUV).rgb;
+    float l = max(c.r, max(c.g, c.b));
+
+    // 软阈值：只保留超出 threshold 的能量比例，避免硬切产生的块状边缘
+    float k = max(l - pc.threshold, 0.0) / max(l, 1e-4);
+    outColor = vec4(c * k, 1.0);
+}

shaders/tonemap.frag

@@ -2,22 +2,37 @@
 
 // Tonemap pass：HDR 线性颜色 → 显示空间。
 // 这是显示变换的显式建模（对齐 UE/Unity：HDR SceneColor → tonemapper → UI）。
-// 当前 = 曝光 + gamma 编码（与旧 Srgb 附件行为等价，观感不变）；
+// ACES = Narkowicz 拟合（线性域映射到 [0,1]），可经 push constant 关闭
-// ACES/Reinhard 色调曲线、bloom 等后处理在此 pass 基础上扩展。
+// 退回纯 gamma（旧观感）；bloom 等后处理在 HDR 域（Tonemap 之前）插 pass。
 
 layout(location = 0) in vec2 fragUV;
 layout(location = 0) out vec4 outColor;
 
 layout(set = 0, binding = 0) uniform sampler2D hdrImage;
+layout(set = 0, binding = 1) uniform sampler2D bloomImage;
 
+// 全屏 pass 通用 push（16B，程序级统一声明，VUID-08878）
 layout(push_constant) uniform PushConstants {
-    float exposure;   // 曝光系数（后续接 ImGui）
+    float exposure;      // 曝光系数
+    float aces;          // >0.5 启用 ACES 曲线，否则纯 gamma
+    float bloomStrength; // bloom 叠加强度（0 = 关闭）
+    float _pad0;
 } pc;
 
-void main() {
+vec3 acesFilm(vec3 x) {
-    vec3 hdr = texture(hdrImage, fragUV).rgb * pc.exposure;
+    return clamp((x * (2.51 * x + 0.03)) / (x * (2.43 * x + 0.59) + 0.14), 0.0, 1.0);
-
+}
-    // 显示编码：linear → sRGB（gamma 2.2 近似）。
+
-    // TODO: ACES 拟合曲线接入点（HDR 泛光/色调映射时替换此行）
+void main() {
-    outColor = vec4(pow(clamp(hdr, 0.0, 1.0), vec3(1.0 / 2.2)), 1.0);
+    // bloom 在 HDR 域叠加（半分辨率模糊图线性采样上采样）
+    vec3 hdr = texture(hdrImage, fragUV).rgb
+             + texture(bloomImage, fragUV).rgb * pc.bloomStrength;
+
+    hdr *= pc.exposure;
+
+    vec3 mapped = (pc.aces > 0.5)
+        ? acesFilm(hdr)
+        : clamp(hdr, 0.0, 1.0);
+
+    outColor = vec4(pow(mapped, vec3(1.0 / 2.2)), 1.0);
 }

shaders/tonemap.vert

@@ -6,6 +6,14 @@
 //  光栅化插值后屏幕内 UV 恰好落在 [0,1]。
 layout(location = 0) out vec2 fragUV;
 
+// 全屏 pass 通用 push（16B，与 tonemap.frag 及所有全屏 pass 一致，VUID-08878）
+layout(push_constant) uniform PushConstants {
+    float exposure;
+    float aces;
+    float _pad0;
+    float _pad1;
+} pc;
+
 void main() {
     vec2 p = vec2(
         float((gl_VertexIndex << 1) & 2),