UE4 stats埋点

2024-01-11 10:29:10

某个Stats文件所统计到大类（Group Name）如下：

某个Stats文件Group Name为Memory的所统计到细项如下：

统计项类型：

为int或float数字类型 // 用于Stat HUD展示，如下图所示

Stat GPU // 显示帧的GPU统计数据注：android平台上没有输出

为Memory类型 // 用于Stat HUD展示，如下图所示

Stat Memory // 显示有关虚幻引擎中各个子系统使用多少内存的统计数据

为hierarchy类别，可以嵌套子节点，包含CallCount、InclusiveTime、ExclusiveTime等字段 // 用于Stat HUD展示和Profiler工具CallStack展示如下图所示

Stat Component // 显示组件列表及组件性能信息

Profiler工具CallStack展示

本文重点讲述如何自定义hierarchy类别来埋点，并在Profiler工具的CallStack树中查看数据。

定义分组

DECLARE_STATS_GROUP(TEXT("AI"),STATGROUP_AI, STATCAT_Advanced); // 3个参数分别对应Description、GroupName、GroupCategory

// 展开后为如下结构体：
struct FStatGroup_STATGROUP_AI
{
    enum { DefaultEnable = true, CompileTimeEnable = true, SortByName = false };

    static __forceinline const char* GetGroupName() { return "STATGROUP_AI"; }
    static __forceinline const char* GetGroupCategory() { return "STATCAT_Advanced"; }
    static __forceinline const TCHAR* GetDescription() { return L"AI"; }
    static __forceinline bool IsDefaultEnabled() { return (bool)DefaultEnable; }
    static __forceinline bool IsCompileTimeEnable() { return (bool)CompileTimeEnable; }
    static __forceinline bool GetSortByName() { return (bool)SortByName; }
};;;

DECLARE_STATS_GROUP_VERBOSE(TEXT("LoadTimeVerbose"), STATGROUP_LoadTimeVerbose, STATCAT_Advanced); // 缺省不开启统计

// 展开后为如下结构体：
struct FStatGroup_STATGROUP_LoadTimeVerbose
{
    enum { DefaultEnable = false, CompileTimeEnable = true, SortByName = false };

    static __forceinline const char* GetGroupName() { return "STATGROUP_LoadTimeVerbose"; }
    static __forceinline const char* GetGroupCategory() { return "STATCAT_Advanced"; }
    static __forceinline const TCHAR* GetDescription() { return L"LoadTimeVerbose"; }
    static __forceinline bool IsDefaultEnabled() { return (bool)DefaultEnable; }
    static __forceinline bool IsCompileTimeEnable() { return (bool)CompileTimeEnable; }
    static __forceinline bool GetSortByName() { return (bool)SortByName; }
};;;

DECLARE_STATS_GROUP_SORTBYNAME(TEXT("Streaming Overview"),STATGROUP_StreamingOverview, STATCAT_Advanced); // 会进行排序，消耗会大一些

// 展开后为如下结构体：
struct FStatGroup_STATGROUP_StreamingOverview
{
    enum { DefaultEnable = true, CompileTimeEnable = true, SortByName = true };

    static __forceinline const char* GetGroupName() { return "STATGROUP_StreamingOverview"; }
    static __forceinline const char* GetGroupCategory() { return "STATCAT_Advanced"; }
    static __forceinline const TCHAR* GetDescription() { return L"Streaming Overview"; }
    static __forceinline bool IsDefaultEnabled() { return (bool)DefaultEnable; }
    static __forceinline bool IsCompileTimeEnable() { return (bool)CompileTimeEnable; }
    static __forceinline bool GetSortByName() { return (bool)SortByName; }
};;;

DECLARE_STATS_GROUP_MAYBE_COMPILED_OUT(TEXT("SlateVeryVerbose"), STATGROUP_SlateVeryVerbose, STATCAT_Advanced, WITH_VERY_VERBOSE_SLATE_STATS); // 缺省不开启统计宏WITH_VERY_VERBOSE_SLATE_STATS为0：表示不编译该统计项的逻辑

// 展开后为如下结构体：
struct FStatGroup_STATGROUP_SlateVeryVerbose
{
    enum { DefaultEnable = false, CompileTimeEnable = 0, SortByName = false };

    static __forceinline const char* GetGroupName() { return "STATGROUP_SlateVeryVerbose"; }
    static __forceinline const char* GetGroupCategory() { return "STATCAT_Advanced"; }
    static __forceinline const TCHAR* GetDescription() { return L"SlateVeryVerbose"; }
    static __forceinline bool IsDefaultEnabled() { return (bool)DefaultEnable; }
    static __forceinline bool IsCompileTimeEnable() { return (bool)CompileTimeEnable; }
    static __forceinline bool GetSortByName() { return (bool)SortByName; }
};;;

定义埋点方式1

在cpp代码的全局区域，定义埋点结构体类型和static全局变量

DECLARE_CYCLE_STAT(TEXT("Test1"), STAT_Test1, STATGROUP_TestGroup); // 3个参数分别对应Description、埋点结构体类型、GroupName

// 宏展开后为：
struct FStat_STAT_Test1
{
    typedef FStatGroup_STATGROUP_TestGroup TGroup;
    static __forceinline const char* GetStatName() { return "STAT_Test1"; }
    static __forceinline const TCHAR* GetDescription() { return L"Test1"; }
    static __forceinline EStatDataType::Type GetStatType() { return EStatDataType::ST_int64; }
    static __forceinline bool IsClearEveryFrame() { return true; }
    static __forceinline bool IsCycleStat() { return true; }

    static __forceinline FPlatformMemory::EMemoryCounterRegion GetMemoryRegion()
    {
        return FPlatformMemory::MCR_Invalid;
    }
};;

static struct FThreadSafeStaticStat<FStat_STAT_Test1> StatPtr_STAT_Test1;;

在函数中插入埋点

DECLARE_STATS_GROUP(TEXT("TestGroupDesc"), STATGROUP_TestGroup, STATCAT_Advanced); // 定义名为TestGroup的分组


DECLARE_CYCLE_STAT(TEXT("Test1"), STAT_Test1, STATGROUP_TestGroup); // 定义Test1的埋点类型与static埋点变量，并放在TestGroup分组中
DECLARE_CYCLE_STAT(TEXT("Test2"), STAT_Test2, STATGROUP_TestGroup); // 定义Test2的埋点类型与static埋点变量，并放在TestGroup分组中
DECLARE_CYCLE_STAT(TEXT("Test3"), STAT_Test3, STATGROUP_TestGroup); // 定义Test3的埋点类型与static埋点变量，并放在TestGroup分组中
DECLARE_CYCLE_STAT(TEXT("Test4"), STAT_Test4, STATGROUP_TestGroup); // 定义Test4的埋点类型与static埋点变量，并放在TestGroup分组中
DECLARE_CYCLE_STAT(TEXT("Test5"), STAT_Test5, STATGROUP_TestGroup); // 定义Test5的埋点类型与static埋点变量，并放在TestGroup分组中
DECLARE_CYCLE_STAT(TEXT("Test6"), STAT_Test6, STATGROUP_TestGroup); // 定义Test6的埋点类型与static埋点变量，并放在TestGroup分组中

// LoopCall(1)在我的电脑耗时约为16ms
#define LoopCall(n)                            \
{                                                uint64 sum = 1;                                for (int32 i = 1; i < 10000000*n; i++)        {                                                sum *= i;                                }                                        }

void AMyTest1Character::StatTest()
{
    SCOPE_CYCLE_COUNTER(STAT_Test1); // 宏展开后的代码为：FScopeCycleCounter CycleCount_STAT_Test1((StatPtr_STAT_Test1.GetStatId()));;

    FPlatformProcess::Sleep(0.002); // 统计到CPU Stall - Sleep

// 条件埋点: 条件成立时，才会埋点成功
    int a = 100;
    CONDITIONAL_SCOPE_CYCLE_COUNTER(STAT_Test2, a > 50); // 宏展开后的代码为：FScopeCycleCounter CycleCount_STAT_Test2(a > 50 ? (StatPtr_STAT_Test2.GetStatId()) : TStatId());;

    LoopCall(1);

    SCOPE_CYCLE_COUNTER(STAT_Test3); // 宏展开后的代码为：FScopeCycleCounter CycleCount_STAT_Test3((StatPtr_STAT_Test3.GetStatId()));;

    FPlatformProcess::Sleep(0.005); // 统计到CPU Stall - Sleep

    {
        SCOPE_CYCLE_COUNTER(STAT_Test6); // 宏展开后的代码为：FScopeCycleCounter CycleCount_STAT_Test6((StatPtr_STAT_Test6.GetStatId()));;
    }
    
    FPlatformProcess::Sleep(0.003); // 统计到CPU Stall - Sleep

    SCOPE_CYCLE_COUNTER(STAT_Test4); // 宏展开后的代码为：FScopeCycleCounter CycleCount_STAT_Test4((StatPtr_STAT_Test4.GetStatId()));;

    {
        SCOPE_CYCLE_COUNTER(STAT_Test5); // 宏展开后的代码为：FScopeCycleCounter CycleCount_STAT_Test5((StatPtr_STAT_Test5.GetStatId()));;
        LoopCall(2);
    }

    {
        SCOPE_CYCLE_COUNTER(STAT_Test5); // 宏展开后的代码为：FScopeCycleCounter CycleCount_STAT_Test5((StatPtr_STAT_Test5.GetStatId()));;
        FPlatformProcess::SleepNoStats(0.5); // 会被统计到当前作用域埋点的IncTime中
    }

    SCOPE_CYCLE_COUNTER(STAT_Test6); // 宏展开后的代码为：FScopeCycleCounter CycleCount_STAT_Test6((StatPtr_STAT_Test6.GetStatId()));;
}

定义埋点方式2

相比方式1，该方式不需要提前定义埋点类型，比较方便

DECLARE_SCOPE_CYCLE_COUNTER(TEXT("UnhashUnreachableObjects"), STAT_UnhashUnreachableObjects, STATGROUP_GC)展开为如下代码：

struct FStat_STAT_UnhashUnreachableObjects
{
    typedef FStatGroup_STATGROUP_GC TGroup;
    static __forceinline const char* GetStatName() { return "STAT_UnhashUnreachableObjects"; }
    static __forceinline const TCHAR* GetDescription() { return L"UnhashUnreachableObjects"; }
    static __forceinline EStatDataType::Type GetStatType() { return EStatDataType::ST_int64; }
    static __forceinline bool IsClearEveryFrame() { return true; }
    static __forceinline bool IsCycleStat() { return true; }

    static __forceinline FPlatformMemory::EMemoryCounterRegion GetMemoryRegion()
    {
        return FPlatformMemory::MCR_Invalid;
    }
};;
static struct FThreadSafeStaticStat<FStat_STAT_UnhashUnreachableObjects> StatPtr_STAT_UnhashUnreachableObjects;
FScopeCycleCounter CycleCount_STAT_UnhashUnreachableObjects((StatPtr_STAT_UnhashUnreachableObjects.GetStatId()));;

在函数中，定义埋点结构体类型和static局部变量，并插入埋点

bool UnhashUnreachableObjects(bool bUseTimeLimit, float TimeLimit)
{
    DECLARE_SCOPE_CYCLE_COUNTER(TEXT("UnhashUnreachableObjects"), STAT_UnhashUnreachableObjects, STATGROUP_GC);
    
    // ... ...
}

定义埋点方式3

在DECLARE_SCOPE_CYCLE_COUNTER基础上封装，放到Quick分组下，更易于使用

QUICK_SCOPE_CYCLE_COUNTER(STAT_QuickTest1); // 即：DECLARE_SCOPE_CYCLE_COUNTER(TEXT("STAT_QuickTest1"),STAT_QuickTest1,STATGROUP_Quick)

// 用宏QUICK_SCOPE_CYCLE_COUNTER定义的埋点会放到GroupName为FStatGroup_STATGROUP_Quick下

QUICK_SCOPE_CYCLE_COUNTER(STAT_QuickTest1)展开为：

struct FStat_STAT_QuickTest1
{
    typedef FStatGroup_STATGROUP_Quick TGroup;
    static __forceinline const char* GetStatName() { return "STAT_QuickTest1"; }
    static __forceinline const TCHAR* GetDescription() { return L"STAT_QuickTest1"; }
    static __forceinline EStatDataType::Type GetStatType() { return EStatDataType::ST_int64; }
    static __forceinline bool IsClearEveryFrame() { return true; }
    static __forceinline bool IsCycleStat() { return true; }

    static __forceinline FPlatformMemory::EMemoryCounterRegion GetMemoryRegion()
    {
        return FPlatformMemory::MCR_Invalid;
    }
};;
static struct FThreadSafeStaticStat<FStat_STAT_QuickTest1> StatPtr_STAT_QuickTest1;
FScopeCycleCounter CycleCount_STAT_QuickTest1((StatPtr_STAT_QuickTest1.GetStatId()));;

在函数中，定义埋点结构体类型和static局部变量，并插入埋点

void AMyTest1Character::StatTest()
{
    QUICK_SCOPE_CYCLE_COUNTER(STAT_QuickTest1);

    // ... ...
}

对UObject对象埋点

DECLARE_STATS_GROUP(TEXT("TestGroupDesc"), STATGROUP_TestGroup, STATCAT_Advanced); // 定义名为TestGroup的分组

DECLARE_CYCLE_STAT(TEXT("ObjTest1"), STAT_ObjTest1, STATGROUP_TestGroup); // 定义ObjTest1的埋点类型与static埋点变量，并放在TestGroup分组中
DECLARE_CYCLE_STAT(TEXT("ObjTest2"), STAT_ObjTest2, STATGROUP_TestGroup); // 定义ObjTest2的埋点类型与static埋点变量，并放在TestGroup分组中

void AMyTest1Character::StatTest()
{
    FString MyBPObjectPath = TEXT("/Game/ThirdPersonCPP/Blueprints/MyBlueprintObject.MyBlueprintObject_C");
    UClass* MyBPObjectClass = LoadClass<UObject>(nullptr, *MyBPObjectPath);
    UMyBPObject* BPObj1 = NewObject<UMyBPObject>(this, MyBPObjectClass);

    FString TexturePath1 = TEXT("/Engine/EngineMaterials/DefaultDiffuse_TC_Masks");
    UTexture2D* TextureObj1 = LoadObject<UTexture2D>(nullptr, *TexturePath1);

    {
        FScopeCycleCounterUObject ObjScope(MyBPObjectClass);

        LoopCall(1);

        FPlatformProcess::Sleep(0.002);

        {
            FScopeCycleCounterUObject ObjScope2(MyBPObjectClass);

            LoopCall(1);

            {
// 通过GET_STATID宏来获取，会被展开为：(StatPtr_STAT_ObjTest1.GetStatId())
                TStatId StatBPObj1 = GET_STATID(STAT_ObjTest1);
                FScopeCycleCounterUObject ObjScope3(BPObj1, StatBPObj1);

                FPlatformProcess::SleepNoStats(0.5);
            }

            {
// 通过GET_STATID宏来获取，会被展开为：(StatPtr_STAT_ObjTest2.GetStatId())
                FScopeCycleCounterUObject ObjScope4(this, GET_STATID(STAT_ObjTest2));

                FPlatformProcess::Sleep(0.003);
            }
        }

        LoopCall(2);
        
        // 动态创建TStatId对象
        TStatId StatObjTest3 = FDynamicStats::CreateStatId<FStatGroup_STATGROUP_TestGroup>(FString(TEXT("ObjTest3")));
        FScopeCycleCounterUObject ObjScope3(MyBPObjectClass, StatObjTest3);
    }
}

即使对UObject指定了埋点类型，但对UObject的统计会放到STATGROUP_UObjects分组中，如下所示：

Tickable对象的Tick耗时

UCLASS()
class UMyBPObject : public UObject, public FTickableGameObject
{
    GENERATED_BODY()
public:
    UMyBPObject();
    ~UMyBPObject();

    virtual TStatId GetStatId() const override
    {
        RETURN_QUICK_DECLARE_CYCLE_STAT(MyBPObject, STATGROUP_Tickables); // 如果不希望被统计，直接返回return TStatId();即可
    }
    virtual bool IsTickable() const override { return !this->IsDefaultSubobject(); }
    virtual void Tick(float DeltaTime) override 
    {
        if (GFrameCounter % 300 == 0)
        {
            FPlatformProcess::SleepNoStats(0.03);
        }
    }
};


void AMyTest1Character::StatTest()
{
    FString MyBPObjectPath = TEXT("/Game/ThirdPersonCPP/Blueprints/MyBlueprintObject.MyBlueprintObject_C");
    UClass* MyBPObjectClass = LoadClass<UObject>(nullptr, *MyBPObjectPath); // MyBPObjectClass为UBlueprintGeneratedClass*类型

    // 创建UMyBPObject对象，并赋值给成员变量UMyBPObject* m_BPObj1
    m_BPObj1 = NewObject<UMyBPObject>(this, MyBPObjectClass);
}

Tickable对象一般会放在STATGROUP_Tickables分组中。以下为某个Stats文件收集到的Tickable对象名称：

Task执行任务的耗时

DECLARE_STATS_GROUP(TEXT("TestGroupDesc"), STATGROUP_TestGroup, STATCAT_Advanced); // 定义名为TestGroup的分组

class FMyTestTask
{
public:
    FMyTestTask()
    {
    }
    static const TCHAR* GetTaskName()
    {
        return TEXT("FMyTestTask");
    }
    FORCEINLINE static TStatId GetStatId()
    {
        RETURN_QUICK_DECLARE_CYCLE_STAT(FMyTestTask, STATGROUP_TestGroup);
    }
    /** return the thread for this task **/
    static ENamedThreads::Type GetDesiredThread()
    {
        return ENamedThreads::AnyThread;
    }

    static ESubsequentsMode::Type GetSubsequentsMode()
    {
        return ESubsequentsMode::TrackSubsequents;
    }

    void DoTask(ENamedThreads::Type CurrentThread, const FGraphEventRef& MyCompletionGraphEvent)
    {
        LoopCall(5);
    }
};

void AMyTest1Character::StatTest()
{
    FGraphEventRef MyTestTaskEvent = TGraphTask<FMyTestTask>::CreateTask().ConstructAndDispatchWhenReady();
// 当前线程挂起，等待Task任务执行完成
    FTaskGraphInterface::Get().WaitUntilTaskCompletes(MyTestTaskEvent);  // 统计到CPU Stall - Wait For Event
}

TaskGraph的任务一般会放在STATGROUP_TaskGraphTasks和STATGROUP_ThreadPoolAsyncTasks分组中。以下为某个Stats文件收集到的Task名称：

其他统计说明

CPU停转的统计：

Stat系统自己开销的统计：

参考

https://docs.unrealengine.com/4.26/zh-CN/TestingAndOptimization/PerformanceAndProfiling/StatCommands/StatsSystemOverview/

https://docs.unrealengine.com/4.26/en-US/TestingAndOptimization/PerformanceAndProfiling/StatCommands/StatsSystemOverview/

UE4 stats埋点

码农公寓

相关文章