Read pending vertex data before exiting gpu loop.

Fixes save pictures in Crimson Gem Saga.

GPU/Directx9/GPU_DX9.cpp

@@ -621,6 +621,11 @@ void DIRECTX9_GPU::FastRunLoop(DisplayList &list) {
 	}
 }
 
+void DIRECTX9_GPU::FinishDeferred() {
+	// This finishes reading any vertex data that is pending.
+	transformDraw_.FinishDeferred();
+}
+
 void DIRECTX9_GPU::ProcessEvent(GPUEvent ev) {
 	switch (ev.type) {
 	case GPU_EVENT_INIT_CLEAR:

GPU/Directx9/GPU_DX9.h

@@ -144,9 +144,10 @@ public:
 	void Execute_BoneMtxData(u32 op, u32 diff);
 
 protected:
-	virtual void FastRunLoop(DisplayList &list);
-	virtual void ProcessEvent(GPUEvent ev);
-	virtual void FastLoadBoneMatrix(u32 target);
+	void FastRunLoop(DisplayList &list) override;
+	void ProcessEvent(GPUEvent ev) override;
+	void FastLoadBoneMatrix(u32 target) override;
+	void FinishDeferred() override;
 
 private:
 	void UpdateCmdInfo();

GPU/Directx9/TransformPipelineDX9.h

@@ -171,6 +171,12 @@ public:
 		DoFlush();
 	}
 
+	void FinishDeferred() {
+		if (!numDrawCalls)
+			return;
+		DecodeVerts();
+	}
+
 	bool IsCodePtrVertexDecoder(const u8 *ptr) const;
 
 protected:

GPU/GLES/GLES_GPU.cpp

@@ -705,6 +705,10 @@ void GLES_GPU::FastRunLoop(DisplayList &list) {
 	downcount = 0;
 }
 
+void GLES_GPU::FinishDeferred() {
+	// This finishes reading any vertex data that is pending.
+	transformDraw_.FinishDeferred();
+}
 
 void GLES_GPU::ProcessEvent(GPUEvent ev) {
 	switch (ev.type) {

GPU/GLES/GLES_GPU.h

@@ -146,9 +146,10 @@ public:
 	void Execute_BlockTransferStart(u32 op, u32 diff);
 
 protected:
-	virtual void FastRunLoop(DisplayList &list);
-	virtual void ProcessEvent(GPUEvent ev);
-	virtual void FastLoadBoneMatrix(u32 target);
+	void FastRunLoop(DisplayList &list) override;
+	void ProcessEvent(GPUEvent ev) override;
+	void FastLoadBoneMatrix(u32 target) override;
+	void FinishDeferred() override;
 
 private:
 	void Flush() {

GPU/GLES/TransformPipeline.h

@@ -174,6 +174,12 @@ public:
 		DoFlush();
 	}
 
+	void FinishDeferred() {
+		if (!numDrawCalls)
+			return;
+		DecodeVerts();
+	}
+
 	bool IsCodePtrVertexDecoder(const u8 *ptr) const;
 
 protected:

GPU/GPUCommon.cpp

@@ -533,6 +533,8 @@ bool GPUCommon::InterpretList(DisplayList &list) {
 		}
 	}
 
+	FinishDeferred();
+
 	// We haven't run the op at list.pc, so it shouldn't count.
 	if (cycleLastPC != list.pc) {
 		UpdatePC(list.pc - 4, list.pc);

GPU/GPUCommon.h

@@ -139,6 +139,8 @@ protected:
 	virtual bool ShouldExitEventLoop() {
 		return coreState != CORE_RUNNING;
 	}
+	virtual void FinishDeferred() {
+	}
 
 	// Allows early unlocking with a guard.  Do not double unlock.
 	class easy_guard {