feat: Tracy GPU 时间线——RG pass 级 GPU zone

RenderGraph 新增 IGpuProfiler 可选注入（pass 边界 Begin/End 钩子 +
帧尾 Collect，zone 覆盖 barrier+lambda+postBarriers），RG 本体不感知
任何 profiler 实现。Renderer 侧 TracyGpuProfiler 实现：TracyVkContext
（校准用临时 cb，池必须带 RESET_COMMAND_BUFFER_BIT——tracy 校准流程
对同一 cb 连续 begin 三次）+ TracyVkZoneTransient（pass 名是运行时
字符串）。TRACY_ENABLE 关闭全链路退化为空。

用法：Tracy GUI（0.14.1 配对）连接后按 pass 名看 GPU 段时间线，
与 ImGui 数表互补；LOD 调优的测量前置。

Engine/src/RenderGraph/RenderGraph.cpp

@@ -728,6 +728,14 @@ namespace Engine::RG {
             auto& pass = m_Passes[passIdx];
             if (pass.culled) continue;
 
+            // profiler zone：覆盖本 pass 的 barrier + lambda + postBarriers
+            // （GPU 时间线里 pass 是完整段，含同步开销）
+            Scope<IGpuProfiler::IScope> zoneScope;
+            if (m_GpuProfiler) {
+                zoneScope = m_GpuProfiler->BeginPass(pass.name.c_str(),
+                                                     frame.cmd);
+            }
+
             if (m_QueryPool) {
                 frame.cmd.writeTimestamp2(
                     vk::PipelineStageFlagBits2::eBottomOfPipe,
@@ -798,8 +806,14 @@ namespace Engine::RG {
                     vk::PipelineStageFlagBits2::eBottomOfPipe,
                     m_QueryPool, frame.frameSlot * m_QueriesPerFrame + tsPos * 2 + 1);
             }
+            zoneScope.reset();   // END 时间戳
             ++tsPos;
         }
+
+        // 帧尾收集：profiler 在当帧 cmd 内解析查询并上报（提交前录制）
+        if (m_GpuProfiler) {
+            m_GpuProfiler->EndFrame(frame.cmd);
+        }
     }
 
     // ---------- RenderGraphContext ----------

Engine/src/Renderer/Renderer.cpp

@@ -17,6 +17,7 @@
 #include <Engine/Renderer/Passes/ForwardPass.h>
 #include <Engine/Renderer/Cubemap.h>
 #include <Engine/Renderer/IblGenerator.h>
+#include <Engine/Renderer/TracyGpuProfiler.h>
 #include <Engine/Renderer/Passes/TAAPass.h>
 #include <Engine/Renderer/Passes/BloomPass.h>
 #include <Engine/Renderer/Passes/TonemapPass.h>
@@ -126,6 +127,10 @@ namespace Engine
         m_TimestampPeriod = static_cast<f32>(
             m_Device->GetPhysicalDevice().getProperties().limits.timestampPeriod);
 
+        // 10. Tracy GPU 时间线（可选注入；TRACY_ENABLE 关闭时空实现）
+        m_TracyProfiler = CreateScope<TracyGpuProfiler>(*m_Device);
+        m_RenderGraph->SetGpuProfiler(m_TracyProfiler.get());
+
         ENGINE_INFO("Renderer initialized");
     }
 

Engine/src/Renderer/TracyGpuProfiler.cpp

@@ -0,0 +1,99 @@
+//
+// Created by atdunbg on 2026/9/24.
+//
+
+#include <Engine/Renderer/TracyGpuProfiler.h>
+#include <Engine/Vulkan/Device.h>
+#include <Engine/Core/Log.h>
+
+#include <tracy/TracyVulkan.hpp>
+#include <cstring>
+
+namespace Engine {
+
+    // pass scope：构造 = BEGIN 时间戳，析构 = END（覆盖 pass lambda 录制期）
+    class TracyGpuProfiler::ScopeImpl final : public RG::RenderGraph::IGpuProfiler::IScope {
+    public:
+#ifdef TRACY_ENABLE
+        ScopeImpl(TracyVkCtx ctx, const char* name, vk::CommandBuffer cmd)
+            : m_Zone(ctx, static_cast<uint32_t>(__LINE__), __FILE__,
+                     sizeof(__FILE__) - 1, __FUNCTION__, sizeof(__FUNCTION__) - 1,
+                     name, std::strlen(name),
+                     static_cast<VkCommandBuffer>(cmd), true) {}
+        tracy::VkCtxScope m_Zone;
+#else
+        ScopeImpl(TracyVkCtx, const char*, vk::CommandBuffer) {}
+#endif
+    };
+
+    TracyGpuProfiler::TracyGpuProfiler(Device& device)
+    {
+#ifdef TRACY_ENABLE
+        // 校准用临时 command buffer：tracy 校准流程对同一 cb 连续
+        // begin 三次（无中间 reset）——池必须带 RESET_COMMAND_BUFFER_BIT，
+        // 否则 VUID-vkBeginCommandBuffer-commandBuffer-00050
+        vk::CommandPoolCreateInfo poolInfo;
+        poolInfo.setQueueFamilyIndex(device.GetGraphicsFamily())
+                .setFlags(vk::CommandPoolCreateFlagBits::eTransient |
+                          vk::CommandPoolCreateFlagBits::eResetCommandBuffer);
+        vk::CommandPool pool =
+            device.GetDevice().createCommandPool(poolInfo);
+        vk::CommandBuffer cb = device.GetDevice().allocateCommandBuffers(
+            vk::CommandBufferAllocateInfo{}
+                .setCommandPool(pool)
+                .setLevel(vk::CommandBufferLevel::ePrimary)
+                .setCommandBufferCount(1))[0];
+
+        m_Ctx = TracyVkContext(
+            static_cast<VkPhysicalDevice>(device.GetPhysicalDevice()),
+            static_cast<VkDevice>(device.GetDevice()),
+            static_cast<VkQueue>(device.GetGraphicsQueue()),
+            static_cast<VkCommandBuffer>(cb));
+
+        device.GetDevice().freeCommandBuffers(pool, cb);
+        device.GetDevice().destroyCommandPool(pool);
+
+        if (m_Ctx) {
+            ENGINE_INFO("Tracy GPU profiler attached (pass zones active)");
+        } else {
+            ENGINE_WARN("Tracy GPU profiler: context creation failed, disabled");
+        }
+#endif
+    }
+
+    TracyGpuProfiler::~TracyGpuProfiler()
+    {
+#ifdef TRACY_ENABLE
+        if (m_Ctx) {
+            TracyVkDestroy(static_cast<TracyVkCtx>(m_Ctx));
+        }
+#endif
+    }
+
+    Scope<TracyGpuProfiler::IScope> TracyGpuProfiler::BeginPass(
+        const char* name, vk::CommandBuffer cmd)
+    {
+#ifdef TRACY_ENABLE
+        if (!m_Ctx) return nullptr;
+        return Scope<IScope>(new ScopeImpl(
+            static_cast<TracyVkCtx>(m_Ctx), name, cmd));
+#else
+        (void)name;
+        (void)cmd;
+        return nullptr;
+#endif
+    }
+
+    void TracyGpuProfiler::EndFrame(vk::CommandBuffer cmd)
+    {
+#ifdef TRACY_ENABLE
+        if (m_Ctx) {
+            TracyVkCollect(static_cast<TracyVkCtx>(m_Ctx),
+                           static_cast<VkCommandBuffer>(cmd));
+        }
+#else
+        (void)cmd;
+#endif
+    }
+
+} // namespace Engine