Merge pull request #17475 from hrydgard/opengl-basic-profiler

Add a trivial profiling tool to the OpenGL backend

Common/GPU/OpenGL/GLFrameData.h

@@ -35,6 +35,13 @@ class GLDeleter {
 	std::vector<GLPushBuffer *> pushBuffers;
 };
 
+struct GLQueueProfileContext {
+	bool enabled;
+	double cpuStartTime;
+	double cpuEndTime;
+};
+
+
 // Per-frame data, round-robin so we can overlap submission with execution of the previous frame.
 struct GLFrameData {
 	bool skipSwap = false;
@@ -49,4 +56,6 @@ struct GLFrameData {
 	GLDeleter deleter;
 	GLDeleter deleter_prev;
 	std::set<GLPushBuffer *> activePushBuffers;
+
+	GLQueueProfileContext profile;
 };

Common/GPU/OpenGL/GLRenderManager.cpp

@@ -8,6 +8,7 @@
 #include "Common/Log.h"
 #include "Common/TimeUtil.h"
 #include "Common/MemoryUtil.h"
+#include "Common/StringUtils.h"
 #include "Common/Math/math_util.h"
 
 #if 0 // def _DEBUG
@@ -187,6 +188,14 @@ void GLRenderManager::StopThread() {
 	}
 }
 
+std::string GLRenderManager::GetGpuProfileString() const {
+	int curFrame = GetCurFrame();
+	const GLQueueProfileContext &profile = frameData_[curFrame].profile;
+
+	float cputime_ms = 1000.0f * (profile.cpuEndTime - profile.cpuStartTime);
+	return StringFromFormat("CPU time to run the list: %0.2f ms", cputime_ms);
+}
+
 void GLRenderManager::BindFramebufferAsRenderTarget(GLRFramebuffer *fb, GLRRenderPassAction color, GLRRenderPassAction depth, GLRRenderPassAction stencil, uint32_t clearColor, float clearDepth, uint8_t clearStencil, const char *tag) {
 	_assert_(insideFrame_);
 #ifdef _DEBUG
@@ -341,14 +350,15 @@ void GLRenderManager::CopyImageToMemorySync(GLRTexture *texture, int mipLevel, i
 	queueRunner_.CopyFromReadbackBuffer(nullptr, w, h, Draw::DataFormat::R8G8B8A8_UNORM, destFormat, pixelStride, pixels);
 }
 
-void GLRenderManager::BeginFrame() {
+void GLRenderManager::BeginFrame(bool enableProfiling) {
 #ifdef _DEBUG
 	curProgram_ = nullptr;
 #endif
 
 	int curFrame = GetCurFrame();
 
 	GLFrameData &frameData = frameData_[curFrame];
+	frameData.profile.enabled = enableProfiling;
 	{
 		VLOG("PUSH: BeginFrame (curFrame = %d, readyForFence = %d, time=%0.3f)", curFrame, (int)frameData.readyForFence, time_now_d());
 		std::unique_lock<std::mutex> lock(frameData.fenceMutex);
@@ -417,6 +427,10 @@ bool GLRenderManager::Run(GLRRenderThreadTask &task) {
 		}
 	}
 
+	if (frameData.profile.enabled) {
+		frameData.profile.cpuStartTime = time_now_d();
+	}
+
 	if (IsVREnabled()) {
 		int passes = GetVRPassesCount();
 		for (int i = 0; i < passes; i++) {
@@ -428,6 +442,10 @@ bool GLRenderManager::Run(GLRRenderThreadTask &task) {
 		queueRunner_.RunSteps(task.steps, skipGLCalls_, false, false);
 	}
 
+	if (frameData.profile.enabled) {
+		frameData.profile.cpuEndTime = time_now_d();
+	}
+
 	if (!skipGLCalls_) {
 		for (auto iter : frameData.activePushBuffers) {
 			iter->MapDevice(bufferStrategy_);

Common/GPU/OpenGL/GLRenderManager.h

@@ -409,8 +409,10 @@ class GLRenderManager {
 		caps_ = caps;
 	}
 
+	std::string GetGpuProfileString() const;
+
 	// Makes sure that the GPU has caught up enough that we can start writing buffers of this frame again.
-	void BeginFrame();
+	void BeginFrame(bool enableProfiling);
 	// Can run on a different thread!
 	void Finish(); 
 

Common/GPU/OpenGL/thin3d_gl.cpp

@@ -328,6 +328,9 @@ class OpenGLContext : public DrawContext {
 		DrawContext::SetTargetSize(w, h);
 		renderManager_.Resize(w, h);
 	}
+	void SetDebugFlags(DebugFlags flags) override {
+		debugFlags_ = flags;
+	}
 
 	const DeviceCaps &GetDeviceCaps() const override {
 		return caps_;
@@ -514,6 +517,8 @@ class OpenGLContext : public DrawContext {
 		GLPushBuffer *push;
 	};
 	FrameData frameData_[GLRenderManager::MAX_INFLIGHT_FRAMES]{};
+
+	DebugFlags debugFlags_ = DebugFlags::NONE;
 };
 
 static constexpr int MakeIntelSimpleVer(int v1, int v2, int v3) {
@@ -778,7 +783,7 @@ OpenGLContext::~OpenGLContext() {
 }
 
 void OpenGLContext::BeginFrame() {
-	renderManager_.BeginFrame();
+	renderManager_.BeginFrame(debugFlags_ & DebugFlags::PROFILE_TIMESTAMPS);
 	FrameData &frameData = frameData_[renderManager_.GetCurFrame()];
 	renderManager_.BeginPushBuffer(frameData.push);
 }

GPU/Common/GPUDebugInterface.h

@@ -257,6 +257,9 @@ class GPUDebugInterface {
 	// cached framebuffers / textures / vertices?
 	// get content of specific framebuffer / texture?
 	// vertex / texture decoding?
+
+	// Note: Wanted to name it GetProfileString but clashes with a Windows API.
+	virtual std::string GetGpuProfileString() { return ""; }
 };
 
 bool GPUDebugInitExpression(GPUDebugInterface *g, const char *str, PostfixExpression &exp);

GPU/GLES/GPU_GLES.cpp

@@ -305,3 +305,8 @@ void GPU_GLES::GetStats(char *buffer, size_t bufsize) {
 		shaderManagerGL_->GetNumPrograms()
 	);
 }
+
+std::string GPU_GLES::GetGpuProfileString() {
+	GLRenderManager *rm = (GLRenderManager *)draw_->GetNativeObject(Draw::NativeObject::RENDER_MANAGER);
+	return rm->GetGpuProfileString();
+}

GPU/GLES/GPU_GLES.h

@@ -51,6 +51,8 @@ class GPU_GLES : public GPUCommonHW {
 	void BeginHostFrame() override;
 	void EndHostFrame() override;
 
+	std::string GetGpuProfileString() override;
+
 protected:
 	void FinishDeferred() override;
 

GPU/Vulkan/DebugVisVulkan.cpp

@@ -37,6 +37,8 @@
 #include "GPU/Vulkan/VulkanUtil.h"
 #include "GPU/Vulkan/TextureCacheVulkan.h"
 
+#include "Core/Config.h"
+
 #undef DrawText
 
 bool comparePushBufferNames(const VulkanMemoryManager *a, const VulkanMemoryManager *b) {
@@ -107,9 +109,14 @@ void DrawGPUProfilerVis(UIContext *ui, GPUInterface *gpu) {
 
 	ui->Begin();
 
-	GPU_Vulkan *gpuVulkan = static_cast<GPU_Vulkan *>(gpu);
+	float scale = 0.4f;
+	if (g_Config.iGPUBackend == (int)GPUBackend::OPENGL) {
+		// Don't have as much info, let's go bigger.
+		scale = 0.7f;
+	}
 
-	std::string text = gpuVulkan->GetGpuProfileString();
+	GPUCommon *gpuCommon = static_cast<GPUCommon *>(gpu);
+	std::string text = gpuCommon->GetGpuProfileString();
 
 	ui->SetFontScale(0.4f, 0.4f);
 	ui->DrawTextShadow(text.c_str(), x, y, 0xFFFFFFFF, FLAG_DYNAMIC_ASCII);

GPU/Vulkan/GPU_Vulkan.h

@@ -59,7 +59,7 @@ class GPU_Vulkan : public GPUCommonHW {
 		return textureCacheVulkan_;
 	}
 
-	std::string GetGpuProfileString();
+	std::string GetGpuProfileString() override;
 
 protected:
 	void FinishDeferred() override;

UI/DevScreens.cpp

@@ -102,6 +102,8 @@ void DevMenuScreen::CreatePopupContents(UI::ViewGroup *parent) {
 	items->Add(new Choice(dev->T("Shader Viewer")))->OnClick.Handle(this, &DevMenuScreen::OnShaderView);
 	if (g_Config.iGPUBackend == (int)GPUBackend::VULKAN) {
 		items->Add(new CheckBox(&g_Config.bShowAllocatorDebug, dev->T("Allocator Viewer")));
+	}
+	if (g_Config.iGPUBackend == (int)GPUBackend::VULKAN || g_Config.iGPUBackend == (int)GPUBackend::OPENGL) {
 		items->Add(new CheckBox(&g_Config.bShowGpuProfile, dev->T("GPU Profile")));
 	}
 	items->Add(new Choice(dev->T("Toggle Freeze")))->OnClick.Handle(this, &DevMenuScreen::OnFreezeFrame);

UI/EmuScreen.cpp

@@ -1616,7 +1616,7 @@ void EmuScreen::renderUI() {
 		DrawAllocatorVis(ctx, gpu);
 	}
 
-	if (g_Config.iGPUBackend == (int)GPUBackend::VULKAN && g_Config.bShowGpuProfile) {
+	if ((g_Config.iGPUBackend == (int)GPUBackend::VULKAN || g_Config.iGPUBackend == (int)GPUBackend::OPENGL) && g_Config.bShowGpuProfile) {
 		DrawGPUProfilerVis(ctx, gpu);
 	}