Files
FlaxEngine/Source/Engine/Profiler/ProfilerCPU.h

418 lines
12 KiB
C++

// Copyright (c) Wojciech Figat. All rights reserved.
#pragma once
#include "Engine/Platform/Platform.h"
#include "Engine/Core/NonCopyable.h"
#include "Engine/Core/Types/String.h"
#include "Engine/Core/Math/Math.h"
#include "Engine/Core/Collections/Array.h"
#include "Engine/Scripting/ScriptingType.h"
#include <ThirdParty/tracy/tracy/Tracy.hpp>
#if COMPILE_WITH_PROFILER
/// <summary>
/// Provides CPU performance measuring methods.
/// </summary>
API_CLASS(Static) class FLAXENGINE_API ProfilerCPU
{
DECLARE_SCRIPTING_TYPE_NO_SPAWN(ProfilerCPU);
public:
/// <summary>
/// Represents single CPU profiling event data.
/// </summary>
API_STRUCT(NoDefault) struct Event
{
DECLARE_SCRIPTING_TYPE_MINIMAL(Event);
/// <summary>
/// The start time (in milliseconds).
/// </summary>
API_FIELD() double Start;
/// <summary>
/// The end time (in milliseconds).
/// </summary>
API_FIELD() double End;
/// <summary>
/// The event depth. Value 0 is used for the root event.
/// </summary>
API_FIELD() int32 Depth;
/// <summary>
/// The native dynamic memory allocation size during this event (excluding the child events). Given value is in bytes.
/// </summary>
API_FIELD() int32 NativeMemoryAllocation;
/// <summary>
/// The managed memory allocation size during this event (excluding the child events). Given value is in bytes.
/// </summary>
API_FIELD() int32 ManagedMemoryAllocation;
API_FIELD(Private, NoArray) Char Name[100];
};
/// <summary>
/// Implements simple profiling events ring-buffer.
/// </summary>
class EventBuffer : public NonCopyable
{
private:
Event* _data;
int32 _capacity;
int32 _capacityMask;
int32 _head;
int32 _count;
public:
EventBuffer();
~EventBuffer();
public:
/// <summary>
/// Gets the amount of the events in the buffer.
/// </summary>
FORCE_INLINE int32 GetCount() const
{
return _count;
}
/// <summary>
/// Gets the event at the specified index.
/// </summary>
/// <param name="index">The index.</param>
/// <returns>The event</returns>
Event& Get(int32 index) const
{
ASSERT_LOW_LAYER(index >= 0 && index < _capacity);
return _data[index];
}
/// <summary>
/// Adds new event to the buffer.
/// </summary>
/// <returns>The event index.</returns>
int32 Add()
{
const int32 index = _head;
_head = (_head + 1) & _capacityMask;
_count = Math::Min(_count + 1, _capacity);
return index;
}
/// <summary>
/// Extracts the buffer data (only ended events starting from the root level with depth=0).
/// </summary>
/// <param name="data">The output data.</param>
/// <param name="withRemoval">True if also remove extracted events to prevent double-gather, false if don't modify the buffer data.</param>
void Extract(Array<Event, HeapAllocation>& data, bool withRemoval);
public:
/// <summary>
/// Ring buffer iterator
/// </summary>
struct Iterator
{
friend EventBuffer;
private:
EventBuffer* _buffer;
int32 _index;
FORCE_INLINE Iterator(EventBuffer* buffer, const int32 index)
: _buffer(buffer)
, _index(index)
{
}
public:
FORCE_INLINE Iterator(const Iterator& other)
: _buffer(other._buffer)
, _index(other._index)
{
}
FORCE_INLINE Iterator(Iterator&& other) noexcept
: _buffer(other._buffer)
, _index(other._index)
{
}
FORCE_INLINE Iterator& operator=(Iterator&& other)
{
_buffer = other._buffer;
_index = other._index;
return *this;
}
FORCE_INLINE Iterator& operator=(const Iterator& other)
{
_buffer = other._buffer;
_index = other._index;
return *this;
}
FORCE_INLINE int32 Index() const
{
return _index;
}
FORCE_INLINE Event& Event() const
{
ASSERT_LOW_LAYER(_buffer && _index >= 0 && _index < _buffer->_capacity);
return _buffer->Get(_index);
}
FORCE_INLINE bool IsEnd() const
{
return _index == _buffer->_head;
}
FORCE_INLINE bool IsNotEnd() const
{
return _index != _buffer->_head;
}
FORCE_INLINE bool operator==(const Iterator& v) const
{
return _buffer == v._buffer && _index == v._index;
}
FORCE_INLINE bool operator!=(const Iterator& v) const
{
return _buffer != v._buffer || _index != v._index;
}
public:
FORCE_INLINE Iterator& operator++()
{
_index = (_index + 1) & _buffer->_capacityMask;
return *this;
}
FORCE_INLINE Iterator operator++(int)
{
Iterator temp = *this;
_index = (_index + 1) & _buffer->_capacityMask;
return temp;
}
FORCE_INLINE Iterator& operator--()
{
_index = (_index - 1) & _buffer->_capacityMask;
return *this;
}
FORCE_INLINE Iterator operator--(int)
{
Iterator temp = *this;
_index = (_index - 1) & _buffer->_capacityMask;
return temp;
}
};
public:
FORCE_INLINE Iterator Begin()
{
return Iterator(this, (_head - _count) & _capacityMask);
}
FORCE_INLINE Iterator Last()
{
ASSERT(_count > 0);
return Iterator(this, (_head - 1) & _capacityMask);
}
FORCE_INLINE Iterator End()
{
return Iterator(this, _head);
}
};
/// <summary>
/// Thread registered for profiling. Holds events data with read/write support.
/// </summary>
class Thread
{
private:
String _name;
int32 _depth = 0;
public:
Thread(const Char* name)
{
_name = name;
}
Thread(const String& name)
{
_name = name;
}
public:
/// <summary>
/// The current thread.
/// </summary>
static THREADLOCAL Thread* Current;
public:
/// <summary>
/// Gets the name.
/// </summary>
FORCE_INLINE const String& GetName() const
{
return _name;
}
/// <summary>
/// The events buffer.
/// </summary>
EventBuffer Buffer;
public:
/// <summary>
/// Begins the event running on a this thread. Call EndEvent with index parameter equal to the returned value by BeginEvent function.
/// </summary>
/// <returns>The event token.</returns>
int32 BeginEvent();
/// <summary>
/// Ends the event running on a this thread.
/// </summary>
/// <param name="index">The event index returned by the BeginEvent method.</param>
void EndEvent(int32 index);
/// <summary>
/// Ends the last event running on a this thread.
/// </summary>
void EndEvent();
};
public:
/// <summary>
/// The registered threads.
/// </summary>
static Array<Thread*, InlinedAllocation<64>> Threads;
/// <summary>
/// The profiling tools usage flag. Can be used to disable profiler. Engine turns it down before the exit and before platform startup.
/// </summary>
static bool Enabled;
public:
/// <summary>
/// Determines whether the current (calling) thread is being profiled by the service (it may has no active profile block but is registered).
/// </summary>
static bool IsProfilingCurrentThread();
/// <summary>
/// Gets the current thread (profiler service shadow object).
/// </summary>
static Thread* GetCurrentThread();
/// <summary>
/// Begins the event. Call EndEvent with index parameter equal to the returned value by BeginEvent function.
/// </summary>
/// <returns>The event token.</returns>
static int32 BeginEvent();
/// <summary>
/// Begins the event. Call EndEvent with index parameter equal to the returned value by BeginEvent function.
/// </summary>
/// <param name="name">The event name.</param>
/// <returns>The event token.</returns>
static int32 BeginEvent(const Char* name);
/// <summary>
/// Begins the event. Call EndEvent with index parameter equal to the returned value by BeginEvent function.
/// </summary>
/// <param name="name">The event name.</param>
/// <returns>The event token.</returns>
static int32 BeginEvent(const char* name);
/// <summary>
/// Ends the event.
/// </summary>
/// <param name="index">The event index returned by the BeginEvent method.</param>
static void EndEvent(int32 index);
/// <summary>
/// Ends the last event.
/// </summary>
static void EndEvent();
/// <summary>
/// Releases resources. Calls to the profiling API after Dispose are not valid.
/// </summary>
static void Dispose();
};
/// <summary>
/// Helper structure used to call BeginEvent/EndEvent within single code block.
/// </summary>
struct ScopeProfileBlockCPU
{
int32 Index;
FORCE_INLINE ScopeProfileBlockCPU(const Char* name)
{
Index = ProfilerCPU::BeginEvent(name);
}
FORCE_INLINE ScopeProfileBlockCPU(const char* name)
{
Index = ProfilerCPU::BeginEvent(name);
}
FORCE_INLINE ~ScopeProfileBlockCPU()
{
ProfilerCPU::EndEvent(Index);
}
};
template<>
struct TIsPODType<ProfilerCPU::Event>
{
enum { Value = true };
};
#include "ProfilerSrcLoc.h"
// Shortcut macros for profiling a single code block execution on CPU
// Use ZoneTransient for Tracy for code that can be hot-reloaded (eg. in Editor) or if name can be a variable
#define PROFILE_CPU_USE_TRANSIENT_DATA 0
#if PROFILE_CPU_USE_TRANSIENT_DATA
#define PROFILE_CPU() ZoneTransient(___tracy_scoped_zone, true); ScopeProfileBlockCPU ProfileBlockCPU(__FUNCTION__)
#define PROFILE_CPU_NAMED(name) ZoneTransientN(___tracy_scoped_zone, name, true); ScopeProfileBlockCPU ProfileBlockCPU(name)
#else
#define PROFILE_CPU() ZoneNamed(___tracy_scoped_zone, true); ScopeProfileBlockCPU ProfileBlockCPU(__FUNCTION__)
#define PROFILE_CPU_NAMED(name) ZoneNamedN(___tracy_scoped_zone, name, true); ScopeProfileBlockCPU ProfileBlockCPU(name)
#endif
#ifdef TRACY_ENABLE
#define PROFILE_CPU_SRC_LOC(srcLoc) tracy::ScopedZone ___tracy_scoped_zone( (tracy::SourceLocationData*)&(srcLoc) ); ScopeProfileBlockCPU ProfileBlockCPU((srcLoc).name)
#define PROFILE_CPU_ASSET(asset) ZoneScoped; const StringView __tracy_asset_name((asset)->GetPath()); ZoneName(*__tracy_asset_name, __tracy_asset_name.Length())
#define PROFILE_CPU_ACTOR(actor) ZoneScoped; const StringView __tracy_actor_name((actor)->GetName()); ZoneName(*__tracy_actor_name, __tracy_actor_name.Length())
#else
#define PROFILE_CPU_SRC_LOC(srcLoc) ScopeProfileBlockCPU ProfileBlockCPU((srcLoc).name)
#define PROFILE_CPU_ASSET(asset)
#define PROFILE_CPU_ACTOR(actor)
#endif
#else
// Empty macros for disabled profiler
#define PROFILE_CPU()
#define PROFILE_CPU_NAMED(name)
#define PROFILE_CPU_SRC_LOC(srcLoc)
#define PROFILE_CPU_ASSET(asset)
#define PROFILE_CPU_ACTOR(actor)
#endif
// CPU-wait zones can be marked with red color for better readability
#define TracyWaitZoneColor 0xba1904