Files
FlaxEngine/Source/Engine/Profiler/ProfilingTools.cpp

232 lines
7.1 KiB
C++

// Copyright (c) 2012-2024 Wojciech Figat. All rights reserved.
#if COMPILE_WITH_PROFILER
#include "ProfilingTools.h"
#include "Engine/Core/Types/Pair.h"
#include "Engine/Engine/Engine.h"
#include "Engine/Engine/Time.h"
#include "Engine/Engine/EngineService.h"
#include "Engine/Graphics/GPUDevice.h"
#include "Engine/Networking/NetworkInternal.h"
ProfilingTools::MainStats ProfilingTools::Stats;
Array<ProfilingTools::ThreadStats, InlinedAllocation<64>> ProfilingTools::EventsCPU;
Array<ProfilerGPU::Event> ProfilingTools::EventsGPU;
Array<ProfilingTools::NetworkEventStat> ProfilingTools::EventsNetwork;
class ProfilingToolsService : public EngineService
{
public:
ProfilingToolsService()
: EngineService(TEXT("Profiling Tools"))
{
Platform::MemoryClear(&ProfilingTools::Stats, sizeof(ProfilingTools::MainStats));
}
void Update() override;
void Dispose() override;
};
ProfilingToolsService ProfilingToolsServiceInstance;
void ProfilingToolsService::Update()
{
ZoneScoped;
// Capture stats
{
auto& stats = ProfilingTools::Stats;
stats.ProcessMemory = Platform::GetProcessMemoryStats();
stats.MemoryCPU = Platform::GetMemoryStats();
stats.MemoryGPU.Total = GPUDevice::Instance->TotalGraphicsMemory;
stats.MemoryGPU.Used = GPUDevice::Instance->GetMemoryUsage();
stats.FPS = Engine::GetFramesPerSecond();
stats.UpdateTimeMs = static_cast<float>(Time::Update.LastLength * 1000.0);
stats.PhysicsTimeMs = static_cast<float>(Time::Physics.LastLength * 1000.0);
stats.DrawCPUTimeMs = static_cast<float>(Time::Draw.LastLength * 1000.0);
float presentTime;
ProfilerGPU::GetLastFrameData(stats.DrawGPUTimeMs, presentTime, stats.DrawStats);
stats.DrawCPUTimeMs = Math::Max(stats.DrawCPUTimeMs - presentTime, 0.0f); // Remove swapchain present wait time to exclude from drawing on CPU
}
// Extract CPU profiler events
Platform::MemoryBarrier();
const auto& threads = ProfilerCPU::Threads;
Platform::MemoryBarrier();
for (auto& pt : ProfilingTools::EventsCPU)
pt.Events.Clear();
ProfilingTools::EventsCPU.EnsureCapacity(threads.Count());
for (int32 i = 0; i < threads.Count(); i++)
{
ProfilerCPU::Thread* thread = threads[i];
if (thread == nullptr)
continue;
ProfilingTools::ThreadStats* pt = nullptr;
for (auto& e : ProfilingTools::EventsCPU)
{
if (e.Name == thread->GetName())
{
pt = &e;
break;
}
}
if (!pt)
{
pt = &ProfilingTools::EventsCPU.AddOne();
pt->Name = thread->GetName();
}
thread->Buffer.Extract(pt->Events, true);
}
#if 0
// Print CPU threads events to the log
for (auto& pt : ProfilingTools::EventsCPU)
{
auto& events = pt.Events;
if (events.HasItems())
{
LOG_FLOOR();
LOG(Info, "Thread: {0}", pt.Name);
for (int j = 0; j < events.Count(); j++)
{
auto e = events[j];
String prev;
for (int d = 0; d < e.Depth; d++)
prev += TEXT("\t");
LOG(Warning, "{2}{0}, Time: {1} ms", e.Name, ((int)((e.End - e.Start) * 1000.0f) / 1000.0f), prev);
}
LOG(Info, "");
LOG_FLOOR();
}
}
#endif
// Get the last resolved GPU frame events
ProfilingTools::EventsGPU.Clear();
uint64 maxFrame = 0;
int32 maxFrameIndex = -1;
auto& frames = ProfilerGPU::Buffers;
for (uint32 i = 0; i < ARRAY_COUNT(frames); i++)
{
if (frames[i].HasData() && frames[i].FrameIndex > maxFrame)
{
maxFrame = frames[i].FrameIndex;
maxFrameIndex = i;
}
}
if (maxFrameIndex != -1)
{
auto& frame = frames[maxFrameIndex];
frame.Extract(ProfilingTools::EventsGPU);
}
// Get the last events from networking runtime
{
auto& networkEvents = ProfilingTools::EventsNetwork;
networkEvents.Resize(NetworkInternal::ProfilerEvents.Count());
int32 i = 0;
for (const auto& e : NetworkInternal::ProfilerEvents)
{
const auto& src = e.Value;
auto& dst = networkEvents[i++];
dst.Count = src.Count;
dst.DataSize = src.DataSize;
dst.MessageSize = src.MessageSize;
dst.Receivers = src.Receivers;
const StringAnsiView& typeName = e.Key.First.GetType().Fullname;
uint64 len = Math::Min<uint64>(typeName.Length(), ARRAY_COUNT(dst.Name) - 10);
Platform::MemoryCopy(dst.Name, typeName.Get(), len);
const StringAnsiView& name = e.Key.Second;
if (name.HasChars())
{
uint64 pos = len;
dst.Name[pos++] = ':';
dst.Name[pos++] = ':';
len = Math::Min<uint64>(name.Length(), ARRAY_COUNT(dst.Name) - pos - 1);
Platform::MemoryCopy(dst.Name + pos, name.Get(), len);
dst.Name[pos + len] = 0;
}
else
{
dst.Name[len] = 0;
}
}
NetworkInternal::ProfilerEvents.Clear();
}
#if 0
// Print CPU events to the log
{
if (ProfilingTools::EventsCPU.HasItems())
{
LOG_FLOOR();
LOG(Info, "CPU");
for (auto& pt : ProfilingTools::EventsCPU)
{
LOG(Info, "");
LOG(Warning, "Thread {0}", pt.Name);
for (auto& e : pt.Events)
{
String prev;
for (int32 d = 0; d < e.Depth; d++)
prev += TEXT("\t");
const double time = e.End - e.Start;
LOG(Warning, "\t{2}{0}, Time: {1} ms", e.Name, ((int32)(time * 1000.0f) / 1000.0f), prev);
}
}
LOG(Info, "");
LOG_FLOOR();
}
}
#endif
#if 0
// Print GPU events to the log
{
auto& events = ProfilingTools::EventsGPU;
if (events.HasItems())
{
LOG_FLOOR();
LOG(Info, "GPU");
for (int j = 0; j < events.Count(); j++)
{
auto e = events[j];
String prev;
for (int d = 0; d < e.Depth; d++)
prev += TEXT("\t");
LOG(Warning, "{2}{0}, Time: {1} ms", e.Name, ((int)(e.Time * 1000.0f) / 1000.0f), prev);
}
LOG(Info, "");
LOG_FLOOR();
}
}
#endif
}
void ProfilingToolsService::Dispose()
{
ProfilingTools::EventsCPU.Clear();
ProfilingTools::EventsCPU.SetCapacity(0);
ProfilingTools::EventsGPU.SetCapacity(0);
ProfilingTools::EventsNetwork.SetCapacity(0);
}
bool ProfilingTools::GetEnabled()
{
return ProfilerCPU::Enabled && ProfilerGPU::Enabled;
}
void ProfilingTools::SetEnabled(bool enabled)
{
ProfilerCPU::Enabled = enabled;
ProfilerGPU::Enabled = enabled;
ProfilerGPU::EventsEnabled = enabled;
NetworkInternal::EnableProfiling = enabled;
}
#endif