Added encoder stats.

examples/common/example-glue.cpp

@@ -197,7 +197,45 @@ void showExampleDialog(entry::AppI* _app, const char* _errorText)
 
 				const float itemHeight = ImGui::GetTextLineHeightWithSpacing();
 
-				if (ImGui::ListBoxHeader("##empty", ImVec2(ImGui::GetWindowWidth(), stats->numViews*itemHeight) ) )
+				if (ImGui::ListBoxHeader("Encoders", ImVec2(ImGui::GetWindowWidth(), stats->numEncoders*itemHeight) ) )
+				{
+					ImGuiListClipper clipper(stats->numEncoders, itemHeight);
+					const double toCpuMs = 1000.0/double(stats->cpuTimerFreq);
+					const double toGpuMs = 1000.0/double(stats->gpuTimerFreq);
+					const float  scale   = 3.0f;
+
+					while (clipper.Step() )
+					{
+						for (int32_t pos = clipper.DisplayStart; pos < clipper.DisplayEnd; ++pos)
+						{
+							const bgfx::EncoderStats& encoderStats = stats->encoderStats[pos];
+
+							ImGui::Text("%3d", pos);
+							ImGui::SameLine(64.0f);
+
+							ImGui::PushStyleColor(ImGuiCol_Button,        cpuColor);
+							ImGui::PushStyleColor(ImGuiCol_ButtonHovered, cpuColor);
+							ImGui::PushStyleColor(ImGuiCol_ButtonActive,  cpuColor);
+
+							const float maxWidth = 30.0f*scale;
+							const float cpuMs    = float( (encoderStats.cpuTimeEnd-encoderStats.cpuTimeBegin)*toCpuMs);
+							const float cpuWidth = bx::fclamp(cpuMs*scale, 1.0f, maxWidth);
+
+							ImGui::Button("", ImVec2(cpuWidth, itemHeight) );
+							if (ImGui::IsItemHovered() )
+							{
+								ImGui::SetTooltip("CPU: %f [ms]", cpuMs);
+							}
+							ImGui::PopStyleColor(3);
+						}
+					}
+
+					ImGui::ListBoxFooter();
+				}
+
+				ImGui::Separator();
+
+				if (ImGui::ListBoxHeader("Views", ImVec2(ImGui::GetWindowWidth(), stats->numViews*itemHeight) ) )
 				{
 					ImGuiListClipper clipper(stats->numViews, itemHeight);
 

include/bgfx/bgfx.h

@@ -780,10 +780,20 @@ namespace bgfx
 	///
 	struct ViewStats
 	{
-		char     name[256];      //!< View name.
-		uint8_t  view;           //!< View id.
-		uint64_t cpuTimeElapsed; //!< CPU (submit) time elapsed.
-		uint64_t gpuTimeElapsed; //!< GPU time elapsed.
+		char    name[256];      //!< View name.
+		uint8_t view;           //!< View id.
+		int64_t cpuTimeElapsed; //!< CPU (submit) time elapsed.
+		int64_t gpuTimeElapsed; //!< GPU time elapsed.
+	};
+
+	/// Encoder stats.
+	///
+	/// @attention C99 equivalent is `bgfx_encoder_stats_t`.
+	///
+	struct EncoderStats
+	{
+		int64_t cpuTimeBegin; //!< Encoder thread CPU submit begin time.
+		int64_t cpuTimeEnd;   //!< Encoder thread CPU submit end time.
 	};
 
 	/// Renderer statistics data.
@@ -792,33 +802,36 @@ namespace bgfx
 	///
 	struct Stats
 	{
-		int64_t cpuTimeFrame;     //!< CPU time between two `bgfx::frame` calls.
-		int64_t cpuTimeBegin;     //!< Render thread CPU submit begin time.
-		int64_t cpuTimeEnd;       //!< Render thread CPU submit end time.
-		int64_t cpuTimerFreq;     //!< CPU timer frequency.
+		int64_t cpuTimeFrame;       //!< CPU time between two `bgfx::frame` calls.
+		int64_t cpuTimeBegin;       //!< Render thread CPU submit begin time.
+		int64_t cpuTimeEnd;         //!< Render thread CPU submit end time.
+		int64_t cpuTimerFreq;       //!< CPU timer frequency.
+
+		int64_t gpuTimeBegin;       //!< GPU frame begin time.
+		int64_t gpuTimeEnd;         //!< GPU frame end time.
+		int64_t gpuTimerFreq;       //!< GPU timer frequency.
 
-		int64_t gpuTimeBegin;     //!< GPU frame begin time.
-		int64_t gpuTimeEnd;       //!< GPU frame end time.
-		int64_t gpuTimerFreq;     //!< GPU timer frequency.
+		int64_t waitRender;         //!< Time spent waiting for render backend thread to finish issuing
+		                            //!  draw commands to underlying graphics API.
+		int64_t waitSubmit;         //!< Time spent waiting for submit thread to advance to next frame.
 
-		int64_t waitRender;       //!< Time spent waiting for render backend thread to finish issuing
-		                          //!  draw commands to underlying graphics API.
-		int64_t waitSubmit;       //!< Time spent waiting for submit thread to advance to next frame.
+		uint32_t numDraw;           //!< Number of draw calls submitted.
+		uint32_t numCompute;        //!< Number of compute calls submitted.
+		uint32_t maxGpuLatency;     //!< GPU driver latency.
 
-		uint32_t numDraw;         //!< Number of draw calls submitted.
-		uint32_t numCompute;      //!< Number of compute calls submitted.
-		uint32_t maxGpuLatency;   //!< GPU driver latency.
+		int64_t gpuMemoryMax;       //!< Maximum available GPU memory for application.
+		int64_t gpuMemoryUsed;      //!< Amount of GPU memory used.
 
-		int64_t gpuMemoryMax;     //!< Maximum available GPU memory for application.
-		int64_t gpuMemoryUsed;    //!< Amount of GPU memory used.
+		uint16_t width;             //!< Backbuffer width in pixels.
+		uint16_t height;            //!< Backbuffer height in pixels.
+		uint16_t textWidth;         //!< Debug text width in characters.
+		uint16_t textHeight;        //!< Debug text height in characters.
 
-		uint16_t width;           //!< Backbuffer width in pixels.
-		uint16_t height;          //!< Backbuffer height in pixels.
-		uint16_t textWidth;       //!< Debug text width in characters.
-		uint16_t textHeight;      //!< Debug text height in characters.
+		uint16_t   numViews;        //!< Number of view stats.
+		ViewStats* viewStats;       //!< View stats.
 
-		uint16_t  numViews;       //!< Number of view stats.
-		ViewStats viewStats[256]; //!< View stats.
+		uint8_t       numEncoders;  //!< Number of encoders used during frame.
+		EncoderStats* encoderStats; //!< Encoder stats.
 	};
 
 	/// Encoder for submitting draw calls from multiple threads. Use `bgfx::begin()`

include/bgfx/c99/bgfx.h

@@ -338,13 +338,20 @@ typedef struct bgfx_hmd
 /**/
 typedef struct bgfx_view_stats
 {
-    char     name[256];
-    uint8_t  view;
-    uint64_t cpuTimeElapsed;
-    uint64_t gpuTimeElapsed;
+    char    name[256];
+    uint8_t view;
+    int64_t cpuTimeElapsed;
+    int64_t gpuTimeElapsed;
 
 } bgfx_view_stats_t;
 
+typedef struct bgfx_encoder_stats
+{
+    int64_t cpuTimeBegin;
+    int64_t cpuTimeEnd;
+
+} bgfx_encoder_stats_t;
+
 /**/
 typedef struct bgfx_stats
 {
@@ -372,8 +379,11 @@ typedef struct bgfx_stats
     uint16_t textWidth;
     uint16_t textHeight;
 
-    uint16_t          numViews;
-    bgfx_view_stats_t viewStats[256];
+    uint16_t           numViews;
+    bgfx_view_stats_t* viewStats;
+
+    uint8_t               numEncoders;
+    bgfx_encoder_stats_t* encoderStats;
 
 } bgfx_stats_t;
 

src/bgfx.cpp

@@ -1721,12 +1721,13 @@ namespace bgfx
 
 	uint32_t Context::frame(bool _capture)
 	{
+		m_encoder[0].end();
+
 		bx::MutexScope resourceApiScope(m_resourceApiLock);
 
 		encoderApiWait();
 		bx::MutexScope encoderApiScope(m_encoderApiLock);
 
-		m_encoder[0].end();
 		m_submit->m_capture = _capture;
 
 		BGFX_PROFILER_SCOPE("bgfx/API thread frame", 0xff2040ff);

src/bgfx_p.h

@@ -1412,7 +1412,7 @@ namespace bgfx
 		VertexDeclHandle   m_decl;
 	};
 
-	struct RenderBind
+	BX_ALIGN_DECL_CACHE_LINE(struct) RenderBind
 	{
 		void clear()
 		{
@@ -1428,7 +1428,7 @@ namespace bgfx
 		Binding m_bind[BGFX_CONFIG_MAX_TEXTURE_SAMPLERS];
 	};
 
-	struct RenderDraw
+	BX_ALIGN_DECL_CACHE_LINE(struct) RenderDraw
 	{
 		void clear()
 		{
@@ -1494,7 +1494,7 @@ namespace bgfx
 		OcclusionQueryHandle m_occlusionQuery;
 	};
 
-	struct RenderCompute
+	BX_ALIGN_DECL_CACHE_LINE(struct) RenderCompute
 	{
 		void clear()
 		{
@@ -1534,7 +1534,7 @@ namespace bgfx
 		RenderCompute compute;
 	};
 
-	struct BlitItem
+	BX_ALIGN_DECL_CACHE_LINE(struct) BlitItem
 	{
 		uint16_t m_srcX;
 		uint16_t m_srcY;
@@ -1707,6 +1707,8 @@ namespace bgfx
 			m_sortKeys[BGFX_CONFIG_MAX_DRAW_CALLS]   = term.encodeDraw(SortKey::SortProgram);
 			m_sortValues[BGFX_CONFIG_MAX_DRAW_CALLS] = BGFX_CONFIG_MAX_DRAW_CALLS;
 			bx::memSet(m_occlusion, 0xff, sizeof(m_occlusion) );
+
+			m_perfStats.viewStats = m_viewStats;
 		}
 
 		~Frame()
@@ -1959,7 +1961,9 @@ namespace bgfx
 
 		TextVideoMem* m_textVideoMem;
 		HMD m_hmd;
-		Stats m_perfStats;
+
+		Stats     m_perfStats;
+		ViewStats m_viewStats[BGFX_CONFIG_MAX_VIEWS];
 
 		int64_t m_waitSubmit;
 		int64_t m_waitRender;
@@ -1979,6 +1983,8 @@ namespace bgfx
 		{
 			m_frame = _frame;
 
+			m_cpuTimeBegin = bx::getHPCounter();
+
 			m_uniformIdx = _idx;
 			m_uniformBegin = 0;
 			m_uniformEnd   = 0;
@@ -1994,6 +2000,8 @@ namespace bgfx
 		{
 			m_uniformBuffer->finish();
 
+			m_cpuTimeEnd = bx::getHPCounter();
+
 			if (BX_ENABLED(BGFX_CONFIG_DEBUG_OCCLUSION) )
 			{
 				m_occlusionQuerySet.clear();
@@ -2282,6 +2290,9 @@ namespace bgfx
 		HandleSet m_occlusionQuerySet;
 
 		UniformBuffer* m_uniformBuffer;
+
+		int64_t m_cpuTimeBegin;
+		int64_t m_cpuTimeEnd;
 	};
 
 	struct VertexDeclRef
@@ -2703,6 +2714,7 @@ namespace bgfx
 			const TextVideoMem* tvm = m_submit->m_textVideoMem;
 			stats.textWidth  = tvm->m_width;
 			stats.textHeight = tvm->m_height;
+			stats.encoderStats = m_encoderStats;
 			return &stats;
 		}
 
@@ -4384,11 +4396,18 @@ namespace bgfx
 
 		BX_NO_INLINE void encoderApiWait()
 		{
-			for (uint32_t ii = 1, num = m_numEncoders; ii < num; ++ii)
+			for (uint32_t ii = 0, num = m_numEncoders; ii < num; ++ii)
 			{
-				m_encoderApiSem.wait();
+				if (0 != ii)
+				{
+					m_encoderApiSem.wait();
+				}
+
+				m_encoderStats[ii].cpuTimeBegin = m_encoder[ii].m_cpuTimeBegin;
+				m_encoderStats[ii].cpuTimeEnd   = m_encoder[ii].m_cpuTimeEnd;
 			}
 
+			m_submit->m_perfStats.numEncoders = uint8_t(m_numEncoders);
 			m_numEncoders = 1;
 		}
 
@@ -4422,11 +4441,12 @@ namespace bgfx
 		}
 #endif // BGFX_CONFIG_MULTITHREADED
 
+		EncoderStats  m_encoderStats[BGFX_CONFIG_MAX_ENCODERS];
 		Encoder*      m_encoder0;
 		EncoderImpl   m_encoder[BGFX_CONFIG_MAX_ENCODERS];
 		uint32_t      m_numEncoders;
 
-		Frame m_frame[1+(BGFX_CONFIG_MULTITHREADED ? 1 : 0)];
+		Frame  m_frame[1+(BGFX_CONFIG_MULTITHREADED ? 1 : 0)];
 		Frame* m_render;
 		Frame* m_submit;