Add a thread for background depth drawing (doesn't fit very well in our task model)

GPU/Common/DrawEngineCommon.cpp

@@ -69,6 +69,7 @@ DrawEngineCommon::DrawEngineCommon() : decoderMap_(32) {
 	}
 	if (useDepthRaster_) {
 		depthDraws_.reserve(256);
+		depthThread_ = std::thread([this]() { DepthThreadFunc(); });
 	}
 }
 
@@ -87,6 +88,11 @@ DrawEngineCommon::~DrawEngineCommon() {
 		delete decoder;
 	});
 	ClearSplineBezierWeights();
+	if (depthThread_.joinable()) {
+		exitDepthThread_ = true;
+		depthEnqueueCond_.notify_one();
+		depthThread_.join();
+	}
 }
 
 void DrawEngineCommon::Init() {
@@ -1071,6 +1077,18 @@ void DrawEngineCommon::FlushQueuedDepth() {
 	inDepthDrawPass_ = false;
 }
 
+void DrawEngineCommon::DepthThreadFunc() {
+	while (true) {
+		{
+			std::unique_lock<std::mutex> lock(depthEnqueueMutex_);
+			depthEnqueueCond_.wait(lock);
+			if (exitDepthThread_) {
+				break;
+			}
+		}
+	}
+}
+
 void DrawEngineCommon::DepthRasterTransform(GEPrimitiveType prim, VertexDecoder *dec, uint32_t vertTypeID, int vertexCount) {
 	if (!gstate.isModeClear() && (!gstate.isDepthTestEnabled() || !gstate.isDepthWriteEnabled())) {
 		return;

GPU/Common/DrawEngineCommon.h

@@ -18,6 +18,8 @@
 #pragma once
 
 #include <vector>
+#include <mutex>
+#include <condition_variable>
 
 #include "Common/CommonTypes.h"
 #include "Common/Data/Collections/Hashmaps.h"
@@ -365,6 +367,7 @@ protected:
 
 private:
 	// Internal implementation details.
+	void DepthThreadFunc();
 
 	inline void EnqueueDepthDraw(const DepthDraw &draw);
 	inline void ProcessDepthDraw(const DepthDraw &draw);
@@ -381,5 +384,14 @@ private:
 
 	double depthRasterPassStart_ = 0.0;
 
+	std::thread depthThread_;
+	bool exitDepthThread_ = false;  // notify depthEnqueueCond_ to check this.
+
 	bool inDepthDrawPass_ = false;
+	std::mutex depthEnqueueMutex_;
+	std::condition_variable depthEnqueueCond_;
+
+	bool processingDepthDraw_ = false;
+	std::mutex depthFinishMutex_;
+	std::condition_variable depthFinishCond_;
 };