From 6b26e1a591aaa00c6765700cb3aa822618acaf09 Mon Sep 17 00:00:00 2001
From: "Unknown W. Brackets" <checkins@unknownbrackets.org>
Date: Sat, 14 Mar 2015 18:11:00 -0700
Subject: [PATCH] Read pending vertex data before exiting gpu loop.

Fixes save pictures in Crimson Gem Saga.
---
 GPU/Directx9/GPU_DX9.cpp            | 5 +++++
 GPU/Directx9/GPU_DX9.h              | 7 ++++---
 GPU/Directx9/TransformPipelineDX9.h | 6 ++++++
 GPU/GLES/GLES_GPU.cpp               | 4 ++++
 GPU/GLES/GLES_GPU.h                 | 7 ++++---
 GPU/GLES/TransformPipeline.h        | 6 ++++++
 GPU/GPUCommon.cpp                   | 2 ++
 GPU/GPUCommon.h                     | 2 ++
 8 files changed, 33 insertions(+), 6 deletions(-)

diff --git a/GPU/Directx9/GPU_DX9.cpp b/GPU/Directx9/GPU_DX9.cpp
index 04e1391d4a..f4acb85add 100644
--- a/GPU/Directx9/GPU_DX9.cpp
+++ b/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:
diff --git a/GPU/Directx9/GPU_DX9.h b/GPU/Directx9/GPU_DX9.h
index 8b093d310a..290ad14300 100644
--- a/GPU/Directx9/GPU_DX9.h
+++ b/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();
diff --git a/GPU/Directx9/TransformPipelineDX9.h b/GPU/Directx9/TransformPipelineDX9.h
index 47f1133de5..37dbdfcd3d 100644
--- a/GPU/Directx9/TransformPipelineDX9.h
+++ b/GPU/Directx9/TransformPipelineDX9.h
@@ -171,6 +171,12 @@ public:
 		DoFlush();
 	}
 
+	void FinishDeferred() {
+		if (!numDrawCalls)
+			return;
+		DecodeVerts();
+	}
+
 	bool IsCodePtrVertexDecoder(const u8 *ptr) const;
 
 protected:
diff --git a/GPU/GLES/GLES_GPU.cpp b/GPU/GLES/GLES_GPU.cpp
index d8caced3e9..6b93fe6537 100644
--- a/GPU/GLES/GLES_GPU.cpp
+++ b/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) {
diff --git a/GPU/GLES/GLES_GPU.h b/GPU/GLES/GLES_GPU.h
index c9937951b1..814f08065c 100644
--- a/GPU/GLES/GLES_GPU.h
+++ b/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() {
diff --git a/GPU/GLES/TransformPipeline.h b/GPU/GLES/TransformPipeline.h
index a91d2f264e..34fcb6bcc6 100644
--- a/GPU/GLES/TransformPipeline.h
+++ b/GPU/GLES/TransformPipeline.h
@@ -174,6 +174,12 @@ public:
 		DoFlush();
 	}
 
+	void FinishDeferred() {
+		if (!numDrawCalls)
+			return;
+		DecodeVerts();
+	}
+
 	bool IsCodePtrVertexDecoder(const u8 *ptr) const;
 
 protected:
diff --git a/GPU/GPUCommon.cpp b/GPU/GPUCommon.cpp
index 51a0453032..a8e3c5bbfa 100644
--- a/GPU/GPUCommon.cpp
+++ b/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);
diff --git a/GPU/GPUCommon.h b/GPU/GPUCommon.h
index ab0967b5be..fe09c08edd 100644
--- a/GPU/GPUCommon.h
+++ b/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 {