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。
2026-09-24 19:45:17 +08:00
parent 99e6712cdf
commit 3bfe28b027
6 changed files with 182 additions and 0 deletions
@@ -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