diff --git a/CMakeLists.txt b/CMakeLists.txt
index 4694274ce5..07ee2c1d5d 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1508,6 +1508,8 @@ add_library(GPU OBJECT
 	GPU/GPUInterface.h
 	GPU/GeDisasm.cpp
 	GPU/GeDisasm.h
+	GPU/GPU.cpp
+	GPU/GPU.h
 	GPU/GPUCommon.cpp
 	GPU/GPUCommon.h
 	GPU/GPUState.cpp
diff --git a/Core/HLE/ReplaceTables.cpp b/Core/HLE/ReplaceTables.cpp
index 6fae1c03a8..f4164dc2f2 100644
--- a/Core/HLE/ReplaceTables.cpp
+++ b/Core/HLE/ReplaceTables.cpp
@@ -32,6 +32,7 @@
 #include "Core/HLE/FunctionWrappers.h"
 
 #include "GPU/Math3D.h"
+#include "GPU/GPU.h"
 #include "GPU/GPUInterface.h"
 #include "GPU/GPUState.h"
 
@@ -687,14 +688,14 @@ static int Hook_brandish_download_frame() {
 }
 
 static int Hook_growlanser_create_saveicon() {
-    const u32 fb_address = Memory::Read_U32(currentMIPS->r[MIPS_REG_SP] + 4);
-    const u32 fmt = Memory::Read_U32(currentMIPS->r[MIPS_REG_SP]);
-    const u32 sz = fmt == GE_FORMAT_8888 ? 0x00088000 : 0x00044000;
-    if (Memory::IsVRAMAddress(fb_address) && fmt <= 3) {
-        gpu->PerformMemoryDownload(fb_address, sz);
-        CBreakPoints::ExecMemCheck(fb_address, true, sz, currentMIPS->pc);
-    }
-    return 0;
+	const u32 fb_address = Memory::Read_U32(currentMIPS->r[MIPS_REG_SP] + 4);
+	const u32 fmt = Memory::Read_U32(currentMIPS->r[MIPS_REG_SP]);
+	const u32 sz = fmt == GE_FORMAT_8888 ? 0x00088000 : 0x00044000;
+	if (Memory::IsVRAMAddress(fb_address) && fmt <= 3) {
+		gpu->PerformMemoryDownload(fb_address, sz);
+		CBreakPoints::ExecMemCheck(fb_address, true, sz, currentMIPS->pc);
+	}
+	return 0;
 }
 
 static int Hook_sd_gundam_g_generation_download_frame() {
@@ -984,7 +985,7 @@ static int Hook_utawarerumono_download_frame() {
 	if (Memory::IsVRAMAddress(fb_address)) {
 		gpu->PerformMemoryDownload(fb_address, 0x00088000);
 		CBreakPoints::ExecMemCheck(fb_address, true, 0x00088000, currentMIPS->pc);
-}
+	}
 	return 0;
 }
 
@@ -1011,7 +1012,7 @@ static int Hook_gakuenheaven_download_frame() {
 	if (Memory::IsVRAMAddress(fb_address)) {
 		gpu->PerformMemoryDownload(fb_address, 0x00088000);
 		CBreakPoints::ExecMemCheck(fb_address, true, 0x00088000, currentMIPS->pc);
-}
+	}
 	return 0;
 }
 
@@ -1020,7 +1021,7 @@ static int Hook_youkosohitsujimura_download_frame() {
 	if (Memory::IsVRAMAddress(fb_address)) {
 		gpu->PerformMemoryDownload(fb_address, 0x00088000);
 		CBreakPoints::ExecMemCheck(fb_address, true, 0x00088000, currentMIPS->pc);
-}
+	}
 	return 0;
 }
 
diff --git a/Core/HLE/sceDisplay.cpp b/Core/HLE/sceDisplay.cpp
index 43d5b5b472..7d3b595c8b 100644
--- a/Core/HLE/sceDisplay.cpp
+++ b/Core/HLE/sceDisplay.cpp
@@ -50,6 +50,7 @@
 #include "Core/HLE/sceKernelThread.h"
 #include "Core/HLE/sceKernelInterrupt.h"
 
+#include "GPU/GPU.h"
 #include "GPU/GPUState.h"
 #include "GPU/GPUInterface.h"
 #include "GPU/Common/FramebufferCommon.h"
@@ -215,8 +216,6 @@ void __DisplayInit() {
 	fpsHistoryValid = 0;
 	fpsHistoryPos = 0;
 
-	InitGfxState();
-
 	__KernelRegisterWaitTypeFuncs(WAITTYPE_VBLANK, __DisplayVblankBeginCallback, __DisplayVblankEndCallback);
 }
 
@@ -272,6 +271,11 @@ void __DisplayDoState(PointerWrap &p) {
 	}
 
 	p.Do(gstate);
+
+	// TODO: GPU stuff is really not the responsibility of sceDisplay.
+	// Display just displays the buffers the GPU has drawn, they are really completely distinct.
+	// Maybe a bit tricky to move at this point, though...
+
 	gstate_c.DoState(p);
 #ifndef _XBOX
 	if (s < 2) {
@@ -298,7 +302,6 @@ void __DisplayDoState(PointerWrap &p) {
 void __DisplayShutdown() {
 	vblankListeners.clear();
 	vblankWaitingThreads.clear();
-	ShutdownGfxState();
 }
 
 void __DisplayListenVblank(VblankCallback callback) {
diff --git a/GPU/CMakeLists.txt b/GPU/CMakeLists.txt
deleted file mode 100644
index 1e0a3d7688..0000000000
--- a/GPU/CMakeLists.txt
+++ /dev/null
@@ -1,34 +0,0 @@
-set(SRCS
-	GPUCommon.cpp
-	GPUState.cpp
-	Math3D.cpp
-	GLES/GLES_GPU.cpp
-	GLES/FragmentShaderGenerator.cpp
-	GLES/Framebuffer.cpp
-	GLES/IndexGenerator.cpp
-	GLES/ShaderManager.cpp
-	GLES/Spline.cpp
-	GLES/StateMapping.cpp
-	GLES/TextureCache.cpp
-	GLES/TextureScaler.cpp
-	GLES/TransformPipeline.cpp
-	GLES/VertexDecoder.cpp
-	GLES/VertexShaderGenerator.cpp
-	Null/NullGpu.cpp
-	Software/Clipper.cpp
-	Software/Lighting.cpp
-	Software/Rasterizer.cpp
-	Software/SoftGpu.cpp
-	Software/TransformUnit.cpp
-)
-
-set(SRCS ${SRCS})
-
-add_library(gpu STATIC ${SRCS})
-target_link_libraries(gpu general gfx_es2)
-target_link_libraries(gpu general lin)
-
-if(UNIX)
-  add_definitions(-fPIC)
-endif(UNIX)
-
diff --git a/GPU/Common/FramebufferCommon.cpp b/GPU/Common/FramebufferCommon.cpp
index 086317c5be..798fe680da 100644
--- a/GPU/Common/FramebufferCommon.cpp
+++ b/GPU/Common/FramebufferCommon.cpp
@@ -340,7 +340,7 @@ void FramebufferManagerCommon::DoSetRenderFrameBuffer() {
 		vfb->drawnHeight = 0;
 		vfb->drawnFormat = fmt;
 		vfb->usageFlags = FB_USAGE_RENDERTARGET;
-		SetColorUpdated(vfb);
+		SetColorUpdated(vfb, gstate_c.skipDrawReason);
 		vfb->depthUpdated = false;
 
 		u32 byteSize = FramebufferByteSize(vfb);
@@ -454,7 +454,7 @@ void FramebufferManagerCommon::UpdateFromMemory(u32 addr, int size, bool safe) {
 						fmt = displayFormat_;
 					}
 					DrawPixels(vfb, 0, 0, Memory::GetPointer(addr | 0x04000000), fmt, vfb->fb_stride, vfb->width, vfb->height);
-					SetColorUpdated(vfb);
+					SetColorUpdated(vfb, gstate_c.skipDrawReason);
 				} else {
 					INFO_LOG(SCEGE, "Invalidating FBO for %08x (%i x %i x %i)", vfb->fb_address, vfb->width, vfb->height, vfb->format);
 					DestroyFramebuf(vfb);
@@ -543,7 +543,7 @@ bool FramebufferManagerCommon::NotifyFramebufferCopy(u32 src, u32 dst, int size,
 			// Just do the blit!
 			if (g_Config.bBlockTransferGPU) {
 				BlitFramebuffer(dstBuffer, 0, dstY, srcBuffer, 0, srcY, srcBuffer->width, srcH, 0);
-				SetColorUpdated(dstBuffer);
+				SetColorUpdated(dstBuffer, gstate_c.skipDrawReason);
 				RebindFramebuffer();
 			}
 		}
@@ -554,7 +554,7 @@ bool FramebufferManagerCommon::NotifyFramebufferCopy(u32 src, u32 dst, int size,
 			FlushBeforeCopy();
 			const u8 *srcBase = Memory::GetPointerUnchecked(src);
 			DrawPixels(dstBuffer, 0, dstY, srcBase, dstBuffer->format, dstBuffer->fb_stride, dstBuffer->width, dstH);
-			SetColorUpdated(dstBuffer);
+			SetColorUpdated(dstBuffer, gstate_c.skipDrawReason);
 			RebindFramebuffer();
 			// This is a memcpy, let's still copy just in case.
 			return false;
@@ -683,7 +683,7 @@ bool FramebufferManagerCommon::NotifyBlockTransferBefore(u32 dstBasePtr, int dst
 					FlushBeforeCopy();
 					BlitFramebuffer(dstBuffer, dstX, dstY, srcBuffer, srcX, srcY, dstWidth, dstHeight, bpp);
 					RebindFramebuffer();
-					SetColorUpdated(dstBuffer);
+					SetColorUpdated(dstBuffer, gstate_c.skipDrawReason);
 					return true;
 				}
 			} else {
@@ -699,7 +699,7 @@ bool FramebufferManagerCommon::NotifyBlockTransferBefore(u32 dstBasePtr, int dst
 				FlushBeforeCopy();
 				BlitFramebuffer(dstBuffer, dstX, dstY, srcBuffer, srcX, srcY, dstWidth, dstHeight, bpp);
 				RebindFramebuffer();
-				SetColorUpdated(dstBuffer);
+				SetColorUpdated(dstBuffer, gstate_c.skipDrawReason);
 				return true;  // No need to actually do the memory copy behind, probably.
 			}
 		}
@@ -764,7 +764,7 @@ void FramebufferManagerCommon::NotifyBlockTransferAfter(u32 dstBasePtr, int dstS
 				int dstBpp = dstBuffer->format == GE_FORMAT_8888 ? 4 : 2;
 				float dstXFactor = (float)bpp / dstBpp;
 				DrawPixels(dstBuffer, static_cast<int>(dstX * dstXFactor), dstY, srcBase, dstBuffer->format, static_cast<int>(srcStride * dstXFactor), static_cast<int>(dstWidth * dstXFactor), dstHeight);
-				SetColorUpdated(dstBuffer);
+				SetColorUpdated(dstBuffer, gstate_c.skipDrawReason);
 				RebindFramebuffer();
 			}
 		}
diff --git a/GPU/Common/FramebufferCommon.h b/GPU/Common/FramebufferCommon.h
index c42d11fc09..220fc2e30c 100644
--- a/GPU/Common/FramebufferCommon.h
+++ b/GPU/Common/FramebufferCommon.h
@@ -19,9 +19,10 @@
 
 #include <set>
 #include <vector>
+
 #include "Common/CommonTypes.h"
 #include "Core/MemMap.h"
-#include "GPU/GPUState.h"
+#include "GPU/GPU.h"
 #include "GPU/ge_constants.h"
 
 enum {
@@ -100,12 +101,12 @@ public:
 	void SetDisplayFramebuffer(u32 framebuf, u32 stride, GEBufferFormat format);
 
 	void DoSetRenderFrameBuffer();
-	void SetRenderFrameBuffer() {
+	void SetRenderFrameBuffer(bool framebufChanged, int skipDrawReason) {
 		// Inlining this part since it's so frequent.
-		if (!gstate_c.framebufChanged && currentRenderVfb_) {
+		if (!framebufChanged && currentRenderVfb_) {
 			currentRenderVfb_->last_frame_render = gpuStats.numFlips;
 			currentRenderVfb_->dirtyAfterDisplay = true;
-			if (!gstate_c.skipDrawReason)
+			if (!skipDrawReason)
 				currentRenderVfb_->reallyDirtyAfterDisplay = true;
 			return;
 		}
@@ -166,9 +167,9 @@ public:
 			currentRenderVfb_->depthUpdated = true;
 		}
 	}
-	void SetColorUpdated() {
+	void SetColorUpdated(int skipDrawReason) {
 		if (currentRenderVfb_) {
-			SetColorUpdated(currentRenderVfb_);
+			SetColorUpdated(currentRenderVfb_, skipDrawReason);
 		}
 	}
 	void SetRenderSize(VirtualFramebuffer *vfb);
@@ -198,13 +199,13 @@ protected:
 
 	void UpdateFramebufUsage(VirtualFramebuffer *vfb);
 
-	void SetColorUpdated(VirtualFramebuffer *dstBuffer) {
+	void SetColorUpdated(VirtualFramebuffer *dstBuffer, int skipDrawReason) {
 		dstBuffer->memoryUpdated = false;
 		dstBuffer->dirtyAfterDisplay = true;
 		dstBuffer->drawnWidth = dstBuffer->width;
 		dstBuffer->drawnHeight = dstBuffer->height;
 		dstBuffer->drawnFormat = dstBuffer->format;
-		if ((gstate_c.skipDrawReason & SKIPDRAW_SKIPFRAME) == 0)
+		if ((skipDrawReason & SKIPDRAW_SKIPFRAME) == 0)
 			dstBuffer->reallyDirtyAfterDisplay = true;
 	}
 
diff --git a/GPU/Common/GPUDebugInterface.h b/GPU/Common/GPUDebugInterface.h
index 0c7bb2afb6..c062777260 100644
--- a/GPU/Common/GPUDebugInterface.h
+++ b/GPU/Common/GPUDebugInterface.h
@@ -20,6 +20,7 @@
 #include <vector>
 #include <string>
 
+#include "GPU/GPU.h"
 #include "GPU/GPUInterface.h"
 #include "Core/MemMap.h"
 
diff --git a/GPU/Common/TextureDecoder.cpp b/GPU/Common/TextureDecoder.cpp
index 298b93954a..53d3968c0e 100644
--- a/GPU/Common/TextureDecoder.cpp
+++ b/GPU/Common/TextureDecoder.cpp
@@ -18,6 +18,9 @@
 #include "ext/xxhash.h"
 #include "Common/CPUDetect.h"
 #include "Common/ColorConv.h"
+
+#include "GPU/GPU.h"
+#include "GPU/GPUState.h"
 #include "GPU/Common/TextureDecoder.h"
 // NEON is in a separate file so that it can be compiled with a runtime check.
 #include "GPU/Common/TextureDecoderNEON.h"
@@ -64,6 +67,39 @@ u32 QuickTexHashSSE2(const void *checkp, u32 size) {
 }
 #endif
 
+// Masks to downalign bufw to 16 bytes, and wrap at 2048.
+static const u32 textureAlignMask16[16] = {
+	0x7FF & ~(((8 * 16) / 16) - 1),  //GE_TFMT_5650,
+	0x7FF & ~(((8 * 16) / 16) - 1),  //GE_TFMT_5551,
+	0x7FF & ~(((8 * 16) / 16) - 1),  //GE_TFMT_4444,
+	0x7FF & ~(((8 * 16) / 32) - 1),  //GE_TFMT_8888,
+	0x7FF & ~(((8 * 16) / 4) - 1),   //GE_TFMT_CLUT4,
+	0x7FF & ~(((8 * 16) / 8) - 1),   //GE_TFMT_CLUT8,
+	0x7FF & ~(((8 * 16) / 16) - 1),  //GE_TFMT_CLUT16,
+	0x7FF & ~(((8 * 16) / 32) - 1),  //GE_TFMT_CLUT32,
+	0x7FF, //GE_TFMT_DXT1,
+	0x7FF, //GE_TFMT_DXT3,
+	0x7FF, //GE_TFMT_DXT5,
+	0,   // INVALID,
+	0,   // INVALID,
+	0,   // INVALID,
+	0,   // INVALID,
+	0,   // INVALID,
+};
+
+u32 GetTextureBufw(int level, u32 texaddr, GETextureFormat format) {
+	// This is a hack to allow for us to draw the huge PPGe texture, which is always in kernel ram.
+	if (texaddr < PSP_GetKernelMemoryEnd())
+		return gstate.texbufwidth[level] & 0x1FFF;
+
+	u32 bufw = gstate.texbufwidth[level] & textureAlignMask16[format];
+	if (bufw == 0) {
+		// If it's less than 16 bytes, use 16 bytes.
+		bufw = (8 * 16) / textureBitsPerPixel[format];
+	}
+	return bufw;
+}
+
 u32 QuickTexHashNonSSE(const void *checkp, u32 size) {
 	u32 check = 0;
 
diff --git a/GPU/Common/TextureDecoder.h b/GPU/Common/TextureDecoder.h
index ec99c69d43..f10e35f5c0 100644
--- a/GPU/Common/TextureDecoder.h
+++ b/GPU/Common/TextureDecoder.h
@@ -27,7 +27,6 @@ enum CheckAlphaResult {
 #include "Common/Common.h"
 #include "Core/MemMap.h"
 #include "GPU/ge_constants.h"
-#include "GPU/GPUState.h"
 #include "GPU/Common/TextureDecoderNEON.h"
 
 void SetupTextureDecoder();
@@ -133,38 +132,7 @@ static const u8 textureBitsPerPixel[16] = {
 	0,   // INVALID,
 };
 
-// Masks to downalign bufw to 16 bytes, and wrap at 2048.
-static const u32 textureAlignMask16[16] = {
-	0x7FF & ~(((8 * 16) / 16) - 1),  //GE_TFMT_5650,
-	0x7FF & ~(((8 * 16) / 16) - 1),  //GE_TFMT_5551,
-	0x7FF & ~(((8 * 16) / 16) - 1),  //GE_TFMT_4444,
-	0x7FF & ~(((8 * 16) / 32) - 1),  //GE_TFMT_8888,
-	0x7FF & ~(((8 * 16) / 4) - 1),   //GE_TFMT_CLUT4,
-	0x7FF & ~(((8 * 16) / 8) - 1),   //GE_TFMT_CLUT8,
-	0x7FF & ~(((8 * 16) / 16) - 1),  //GE_TFMT_CLUT16,
-	0x7FF & ~(((8 * 16) / 32) - 1),  //GE_TFMT_CLUT32,
-	0x7FF, //GE_TFMT_DXT1,
-	0x7FF, //GE_TFMT_DXT3,
-	0x7FF, //GE_TFMT_DXT5,
-	0,   // INVALID,
-	0,   // INVALID,
-	0,   // INVALID,
-	0,   // INVALID,
-	0,   // INVALID,
-};
-
-static inline u32 GetTextureBufw(int level, u32 texaddr, GETextureFormat format) {
-	// This is a hack to allow for us to draw the huge PPGe texture, which is always in kernel ram.
-	if (texaddr < PSP_GetKernelMemoryEnd())
-		return gstate.texbufwidth[level] & 0x1FFF;
-
-	u32 bufw = gstate.texbufwidth[level] & textureAlignMask16[format];
-	if (bufw == 0) {
-		// If it's less than 16 bytes, use 16 bytes.
-		bufw = (8 * 16) / textureBitsPerPixel[format];
-	}
-	return bufw;
-}
+u32 GetTextureBufw(int level, u32 texaddr, GETextureFormat format);
 
 template <typename IndexT, typename ClutT>
 inline void DeIndexTexture(ClutT *dest, const IndexT *indexed, int length, const ClutT *clut) {
diff --git a/GPU/Directx9/FramebufferDX9.cpp b/GPU/Directx9/FramebufferDX9.cpp
index f98e27654f..b49ea0e04c 100644
--- a/GPU/Directx9/FramebufferDX9.cpp
+++ b/GPU/Directx9/FramebufferDX9.cpp
@@ -1218,7 +1218,7 @@ namespace DX9 {
 	void FramebufferManagerDX9::FlushBeforeCopy() {
 		// Flush anything not yet drawn before blitting, downloading, or uploading.
 		// This might be a stalled list, or unflushed before a block transfer, etc.
-		SetRenderFrameBuffer();
+		SetRenderFrameBuffer(gstate_c.framebufChanged, gstate_c.skipDrawReason);
 		transformDraw_->Flush();
 	}
 
diff --git a/GPU/Directx9/GPU_DX9.cpp b/GPU/Directx9/GPU_DX9.cpp
index 900f9873b0..9ecef98f75 100644
--- a/GPU/Directx9/GPU_DX9.cpp
+++ b/GPU/Directx9/GPU_DX9.cpp
@@ -741,7 +741,7 @@ void DIRECTX9_GPU::Execute_Prim(u32 op, u32 diff) {
 	}
 
 	// This also make skipping drawing very effective.
-	framebufferManager_.SetRenderFrameBuffer();
+	framebufferManager_.SetRenderFrameBuffer(gstate_c.framebufChanged, gstate_c.skipDrawReason);
 	if (gstate_c.skipDrawReason & (SKIPDRAW_SKIPFRAME | SKIPDRAW_NON_DISPLAYED_FB)) {
 		transformDraw_.SetupVertexDecoder(gstate.vertType);
 		// Rough estimate, not sure what's correct.
@@ -796,7 +796,7 @@ void DIRECTX9_GPU::Execute_Prim(u32 op, u32 diff) {
 
 void DIRECTX9_GPU::Execute_Bezier(u32 op, u32 diff) {
 	// This also make skipping drawing very effective.
-	framebufferManager_.SetRenderFrameBuffer();
+	framebufferManager_.SetRenderFrameBuffer(gstate_c.framebufChanged, gstate_c.skipDrawReason);
 	if (gstate_c.skipDrawReason & (SKIPDRAW_SKIPFRAME | SKIPDRAW_NON_DISPLAYED_FB)) {
 		// TODO: Should this eat some cycles?  Probably yes.  Not sure if important.
 		return;
@@ -839,7 +839,7 @@ void DIRECTX9_GPU::Execute_Bezier(u32 op, u32 diff) {
 
 void DIRECTX9_GPU::Execute_Spline(u32 op, u32 diff) {
 	// This also make skipping drawing very effective.
-	framebufferManager_.SetRenderFrameBuffer();
+	framebufferManager_.SetRenderFrameBuffer(gstate_c.framebufChanged, gstate_c.skipDrawReason);
 	if (gstate_c.skipDrawReason & (SKIPDRAW_SKIPFRAME | SKIPDRAW_NON_DISPLAYED_FB)) {
 		// TODO: Should this eat some cycles?  Probably yes.  Not sure if important.
 		return;
diff --git a/GPU/Directx9/TextureCacheDX9.h b/GPU/Directx9/TextureCacheDX9.h
index 25cb5d82b7..eb25632717 100644
--- a/GPU/Directx9/TextureCacheDX9.h
+++ b/GPU/Directx9/TextureCacheDX9.h
@@ -22,8 +22,8 @@
 #include "../Globals.h"
 #include "helper/global.h"
 #include "helper/fbo.h"
+#include "GPU/GPU.h"
 #include "GPU/GPUInterface.h"
-#include "GPU/GPUState.h"
 #include "GPU/Directx9/TextureScalerDX9.h"
 #include "GPU/Common/TextureCacheCommon.h"
 
diff --git a/GPU/Directx9/TransformPipelineDX9.cpp b/GPU/Directx9/TransformPipelineDX9.cpp
index 6b7eb88252..2856af7297 100644
--- a/GPU/Directx9/TransformPipelineDX9.cpp
+++ b/GPU/Directx9/TransformPipelineDX9.cpp
@@ -857,7 +857,7 @@ rotateVBO:
 				framebufferManager_->SetDepthUpdated();
 			}
 			if (mask & D3DCLEAR_TARGET) {
-				framebufferManager_->SetColorUpdated();
+				framebufferManager_->SetColorUpdated(gstate_c.skipDrawReason);
 			}
 
 			dxstate.colorMask.set((mask & D3DCLEAR_TARGET) != 0, (mask & D3DCLEAR_TARGET) != 0, (mask & D3DCLEAR_TARGET) != 0, (mask & D3DCLEAR_STENCIL) != 0);
@@ -876,7 +876,7 @@ rotateVBO:
 	dcid_ = 0;
 	prevPrim_ = GE_PRIM_INVALID;
 	gstate_c.vertexFullAlpha = true;
-	framebufferManager_->SetColorUpdated();
+	framebufferManager_->SetColorUpdated(gstate_c.skipDrawReason);
 
 	host->GPUNotifyDraw();
 }
diff --git a/GPU/Directx9/TransformPipelineDX9.h b/GPU/Directx9/TransformPipelineDX9.h
index fcb7bfe13f..c698cccfe7 100644
--- a/GPU/Directx9/TransformPipelineDX9.h
+++ b/GPU/Directx9/TransformPipelineDX9.h
@@ -21,6 +21,7 @@
 
 #include <d3d9.h>
 
+#include "GPU/GPUState.h"
 #include "GPU/Common/GPUDebugInterface.h"
 #include "GPU/Common/IndexGenerator.h"
 #include "GPU/Common/VertexDecoderCommon.h"
@@ -28,6 +29,7 @@
 #include "GPU/Directx9/PixelShaderGeneratorDX9.h"
 
 struct DecVtxFormat;
+struct UVScale;
 
 namespace DX9 {
 
diff --git a/GPU/GLES/Framebuffer.cpp b/GPU/GLES/Framebuffer.cpp
index e750787e82..1dd10ed02b 100644
--- a/GPU/GLES/Framebuffer.cpp
+++ b/GPU/GLES/Framebuffer.cpp
@@ -1857,7 +1857,7 @@ void FramebufferManager::DestroyAllFBOs() {
 void FramebufferManager::FlushBeforeCopy() {
 	// Flush anything not yet drawn before blitting, downloading, or uploading.
 	// This might be a stalled list, or unflushed before a block transfer, etc.
-	SetRenderFrameBuffer();
+	SetRenderFrameBuffer(gstate_c.framebufChanged, gstate_c.skipDrawReason);
 	transformDraw_->Flush();
 }
 
diff --git a/GPU/GLES/GLES_GPU.cpp b/GPU/GLES/GLES_GPU.cpp
index 4fab1d65fb..8d2edb8d4a 100644
--- a/GPU/GLES/GLES_GPU.cpp
+++ b/GPU/GLES/GLES_GPU.cpp
@@ -805,7 +805,7 @@ void GLES_GPU::Execute_Prim(u32 op, u32 diff) {
 	}
 
 	// This also makes skipping drawing very effective.
-	framebufferManager_.SetRenderFrameBuffer();
+	framebufferManager_.SetRenderFrameBuffer(gstate_c.framebufChanged, gstate_c.skipDrawReason);
 	if (gstate_c.skipDrawReason & (SKIPDRAW_SKIPFRAME | SKIPDRAW_NON_DISPLAYED_FB))	{
 		transformDraw_.SetupVertexDecoder(gstate.vertType);
 		// Rough estimate, not sure what's correct.
@@ -881,7 +881,7 @@ void GLES_GPU::Execute_VertexTypeSkinning(u32 op, u32 diff) {
 
 void GLES_GPU::Execute_Bezier(u32 op, u32 diff) {
 	// This also make skipping drawing very effective.
-	framebufferManager_.SetRenderFrameBuffer();
+	framebufferManager_.SetRenderFrameBuffer(gstate_c.framebufChanged, gstate_c.skipDrawReason);
 	if (gstate_c.skipDrawReason & (SKIPDRAW_SKIPFRAME | SKIPDRAW_NON_DISPLAYED_FB))	{
 		// TODO: Should this eat some cycles?  Probably yes.  Not sure if important.
 		return;
@@ -924,7 +924,7 @@ void GLES_GPU::Execute_Bezier(u32 op, u32 diff) {
 
 void GLES_GPU::Execute_Spline(u32 op, u32 diff) {
 	// This also make skipping drawing very effective.
-	framebufferManager_.SetRenderFrameBuffer();
+	framebufferManager_.SetRenderFrameBuffer(gstate_c.framebufChanged, gstate_c.skipDrawReason);
 	if (gstate_c.skipDrawReason & (SKIPDRAW_SKIPFRAME | SKIPDRAW_NON_DISPLAYED_FB))	{
 		// TODO: Should this eat some cycles?  Probably yes.  Not sure if important.
 		return;
diff --git a/GPU/GLES/ShaderManager.cpp b/GPU/GLES/ShaderManager.cpp
index 503e4a0ba8..e9a64d026f 100644
--- a/GPU/GLES/ShaderManager.cpp
+++ b/GPU/GLES/ShaderManager.cpp
@@ -828,9 +828,10 @@ LinkedShader *ShaderManager::ApplyFragmentShader(Shader *vs, int prim, u32 vertT
 	// Okay, we have both shaders. Let's see if there's a linked one.
 	LinkedShader *ls = NULL;
 
+	u32 switchDirty = shaderSwitchDirty_;
 	for (auto iter = linkedShaderCache_.begin(); iter != linkedShaderCache_.end(); ++iter) {
 		// Deferred dirtying! Let's see if we can make this even more clever later.
-		iter->ls->dirtyUniforms |= shaderSwitchDirty_;
+		iter->ls->dirtyUniforms |= switchDirty;
 
 		if (iter->vs == vs && iter->fs == fs) {
 			ls = iter->ls;
diff --git a/GPU/GLES/TransformPipeline.cpp b/GPU/GLES/TransformPipeline.cpp
index f34be585f9..2994fe26bd 100644
--- a/GPU/GLES/TransformPipeline.cpp
+++ b/GPU/GLES/TransformPipeline.cpp
@@ -876,7 +876,7 @@ rotateVBO:
 			// Stencil takes alpha.
 			glClearStencil(clearColor >> 24);
 			glClear(target);
-			framebufferManager_->SetColorUpdated();
+			framebufferManager_->SetColorUpdated(gstate_c.skipDrawReason);
 		}
 	}
 
@@ -891,7 +891,7 @@ rotateVBO:
 	dcid_ = 0;
 	prevPrim_ = GE_PRIM_INVALID;
 	gstate_c.vertexFullAlpha = true;
-	framebufferManager_->SetColorUpdated();
+	framebufferManager_->SetColorUpdated(gstate_c.skipDrawReason);
 
 #ifndef MOBILE_DEVICE
 	host->GPUNotifyDraw();
diff --git a/GPU/GLES/TransformPipeline.h b/GPU/GLES/TransformPipeline.h
index 3e18d5c930..b93a6c992a 100644
--- a/GPU/GLES/TransformPipeline.h
+++ b/GPU/GLES/TransformPipeline.h
@@ -18,6 +18,8 @@
 #pragma once
 
 #include <unordered_map>
+
+#include "GPU/GPUState.h"
 #include "GPU/Common/GPUDebugInterface.h"
 #include "GPU/Common/IndexGenerator.h"
 #include "GPU/Common/VertexDecoderCommon.h"
diff --git a/GPU/GPU.cpp b/GPU/GPU.cpp
new file mode 100644
index 0000000000..f4f4a54f6e
--- /dev/null
+++ b/GPU/GPU.cpp
@@ -0,0 +1,98 @@
+// Copyright (c) 2015- PPSSPP Project.
+
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU General Public License as published by
+// the Free Software Foundation, version 2.0 or later versions.
+
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU General Public License 2.0 for more details.
+
+// A copy of the GPL 2.0 should have been included with the program.
+// If not, see http://www.gnu.org/licenses/
+
+// Official git repository and contact information can be found at
+// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
+
+#include "Core/Core.h"
+
+#include "GPU/GPU.h"
+#include "GPU/GPUInterface.h"
+#include "GPU/GLES/GLES_GPU.h"
+#include "GPU/Null/NullGpu.h"
+#include "GPU/Software/SoftGpu.h"
+
+#if defined(_WIN32)
+#include "GPU/Directx9/helper/global.h"
+#include "GPU/Directx9/GPU_DX9.h"
+#endif
+
+GPUStatistics gpuStats;
+GPUInterface *gpu;
+GPUDebugInterface *gpuDebug;
+
+template <typename T>
+static void SetGPU(T *obj) {
+	gpu = obj;
+	gpuDebug = obj;
+}
+
+bool GPU_Init() {
+	switch (PSP_CoreParameter().gpuCore) {
+	case GPU_NULL:
+		SetGPU(new NullGPU());
+		break;
+	case GPU_GLES:
+		SetGPU(new GLES_GPU());
+		break;
+	case GPU_SOFTWARE:
+		SetGPU(new SoftGPU());
+		break;
+	case GPU_DIRECTX9:
+#if defined(_WIN32)
+		SetGPU(new DIRECTX9_GPU());
+#endif
+		break;
+	}
+
+	return gpu != NULL;
+}
+
+void GPU_Shutdown() {
+	delete gpu;
+	gpu = 0;
+	gpuDebug = 0;
+}
+
+void GPU_Reinitialize() {
+	if (gpu) {
+		gpu->Reinitialize();
+	}
+}
+
+void InitGfxState() {
+	memset(&gstate, 0, sizeof(gstate));
+	memset(&gstate_c, 0, sizeof(gstate_c));
+	for (int i = 0; i < 256; i++) {
+		gstate.cmdmem[i] = i << 24;
+	}
+
+	// Lighting is not enabled by default, matrices are zero initialized.
+	memset(gstate.worldMatrix, 0, sizeof(gstate.worldMatrix));
+	memset(gstate.viewMatrix, 0, sizeof(gstate.viewMatrix));
+	memset(gstate.projMatrix, 0, sizeof(gstate.projMatrix));
+	memset(gstate.tgenMatrix, 0, sizeof(gstate.tgenMatrix));
+	memset(gstate.boneMatrix, 0, sizeof(gstate.boneMatrix));
+}
+
+void ShutdownGfxState() {
+}
+
+// When you have changed state outside the psp gfx core,
+// or saved the context and has reloaded it, call this function.
+void ReapplyGfxState() {
+	if (!gpu)
+		return;
+	gpu->ReapplyGfxState();
+}
diff --git a/GPU/GPU.h b/GPU/GPU.h
new file mode 100644
index 0000000000..ff504ae47c
--- /dev/null
+++ b/GPU/GPU.h
@@ -0,0 +1,113 @@
+// Copyright (c) 2015- PPSSPP Project.
+
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU General Public License as published by
+// the Free Software Foundation, version 2.0 or later versions.
+
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU General Public License 2.0 for more details.
+
+// A copy of the GPL 2.0 should have been included with the program.
+// If not, see http://www.gnu.org/licenses/
+
+// Official git repository and contact information can be found at
+// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
+
+
+#pragma once
+
+#include <cstring>
+
+class GPUInterface;
+class GPUDebugInterface;
+
+enum SkipDrawReasonFlags {
+	SKIPDRAW_SKIPFRAME = 1,
+	SKIPDRAW_NON_DISPLAYED_FB = 2,   // Skip drawing to FBO:s that have not been displayed.
+	SKIPDRAW_BAD_FB_TEXTURE = 4,
+	SKIPDRAW_WINDOW_MINIMIZED = 8, // Don't draw when the host window is minimized.
+};
+
+// Global GPU-related utility functions. 
+// Nothing directly Ge-related in here.
+
+// PSP uses a curious 24-bit float - it's the top 24 bits of a regular IEEE754 32-bit float.
+// This is used for light positions, transform matrices, you name it.
+inline float getFloat24(unsigned int data) {
+	data <<= 8;
+	float f;
+	memcpy(&f, &data, 4);
+	return f;
+}
+
+// in case we ever want to generate PSP display lists...
+inline unsigned int toFloat24(float f) {
+	unsigned int i;
+	memcpy(&i, &f, 4);
+	return i >> 8;
+}
+
+struct GPUStatistics {
+	void Reset() {
+		// Never add a vtable :)
+		memset(this, 0, sizeof(*this));
+	}
+	void ResetFrame() {
+		numDrawCalls = 0;
+		numCachedDrawCalls = 0;
+		numVertsSubmitted = 0;
+		numCachedVertsDrawn = 0;
+		numUncachedVertsDrawn = 0;
+		numTrackedVertexArrays = 0;
+		numTextureInvalidations = 0;
+		numTextureSwitches = 0;
+		numShaderSwitches = 0;
+		numFlushes = 0;
+		numTexturesDecoded = 0;
+		numAlphaTestedDraws = 0;
+		numNonAlphaTestedDraws = 0;
+		msProcessingDisplayLists = 0;
+		vertexGPUCycles = 0;
+		otherGPUCycles = 0;
+		memset(gpuCommandsAtCallLevel, 0, sizeof(gpuCommandsAtCallLevel));
+	}
+
+	// Per frame statistics
+	int numDrawCalls;
+	int numCachedDrawCalls;
+	int numFlushes;
+	int numVertsSubmitted;
+	int numCachedVertsDrawn;
+	int numUncachedVertsDrawn;
+	int numTrackedVertexArrays;
+	int numTextureInvalidations;
+	int numTextureSwitches;
+	int numShaderSwitches;
+	int numTexturesDecoded;
+	double msProcessingDisplayLists;
+	int vertexGPUCycles;
+	int otherGPUCycles;
+	int gpuCommandsAtCallLevel[4];
+
+	int numAlphaTestedDraws;
+	int numNonAlphaTestedDraws;
+
+	// Total statistics, updated by the GPU core in UpdateStats
+	int numVBlanks;
+	int numFlips;
+	int numTextures;
+	int numVertexShaders;
+	int numFragmentShaders;
+	int numShaders;
+	int numFBOs;
+};
+
+extern GPUStatistics gpuStats;
+extern GPUInterface *gpu;
+extern GPUDebugInterface *gpuDebug;
+
+bool GPU_Init();
+void GPU_Shutdown();
+void GPU_Reinitialize();
diff --git a/GPU/GPU.vcxproj b/GPU/GPU.vcxproj
index 8870207a94..0e92a7c33d 100644
--- a/GPU/GPU.vcxproj
+++ b/GPU/GPU.vcxproj
@@ -227,6 +227,7 @@
     <ClInclude Include="GLES\TextureScaler.h" />
     <ClInclude Include="GLES\TransformPipeline.h" />
     <ClInclude Include="GLES\VertexShaderGenerator.h" />
+    <ClInclude Include="GPU.h" />
     <ClInclude Include="GPUCommon.h" />
     <ClInclude Include="GPUInterface.h" />
     <ClInclude Include="GPUState.h" />
@@ -300,6 +301,7 @@
     <ClCompile Include="GLES\TextureScaler.cpp" />
     <ClCompile Include="GLES\TransformPipeline.cpp" />
     <ClCompile Include="GLES\VertexShaderGenerator.cpp" />
+    <ClCompile Include="GPU.cpp" />
     <ClCompile Include="GPUCommon.cpp" />
     <ClCompile Include="GPUState.cpp" />
     <ClCompile Include="Math3D.cpp" />
diff --git a/GPU/GPU.vcxproj.filters b/GPU/GPU.vcxproj.filters
index 0348eb8683..1ef4f62305 100644
--- a/GPU/GPU.vcxproj.filters
+++ b/GPU/GPU.vcxproj.filters
@@ -183,6 +183,7 @@
     <ClInclude Include="Common\TextureScalerCommon.h">
       <Filter>Common</Filter>
     </ClInclude>
+    <ClInclude Include="GPU.h" />
   </ItemGroup>
   <ItemGroup>
     <ClCompile Include="Math3D.cpp">
@@ -350,8 +351,9 @@
     <ClCompile Include="Common\TextureScalerCommon.cpp">
       <Filter>Common</Filter>
     </ClCompile>
+    <ClCompile Include="GPU.cpp" />
   </ItemGroup>
   <ItemGroup>
     <None Include="CMakeLists.txt" />
   </ItemGroup>
-</Project>
+</Project>
\ No newline at end of file
diff --git a/GPU/GPUCommon.cpp b/GPU/GPUCommon.cpp
index 6822ed7471..cb97da1071 100644
--- a/GPU/GPUCommon.cpp
+++ b/GPU/GPUCommon.cpp
@@ -3,6 +3,7 @@
 #include "native/base/timeutil.h"
 #include "Common/ColorConv.h"
 #include "GPU/GeDisasm.h"
+#include "GPU/GPU.h"
 #include "GPU/GPUCommon.h"
 #include "GPU/GPUState.h"
 #include "ChunkFile.h"
@@ -24,6 +25,11 @@ GPUCommon::GPUCommon() :
 	Reinitialize();
 	SetupColorConv();
 	SetThreadEnabled(g_Config.bSeparateCPUThread);
+	InitGfxState();
+}
+
+GPUCommon::~GPUCommon() {
+	ShutdownGfxState();
 }
 
 void GPUCommon::Reinitialize() {
@@ -639,7 +645,7 @@ void GPUCommon::ReapplyGfxStateInternal() {
 	// Let's just skip the transfer size stuff, it's just values.
 }
 
-inline void GPUCommon::UpdateState(GPUState state) {
+inline void GPUCommon::UpdateState(GPURunState state) {
 	gpuState = state;
 	if (state != GPUSTATE_RUNNING)
 		downcount = 0;
diff --git a/GPU/GPUCommon.h b/GPU/GPUCommon.h
index fe09c08edd..cc60d15080 100644
--- a/GPU/GPUCommon.h
+++ b/GPU/GPUCommon.h
@@ -14,11 +14,10 @@
 
 typedef ThreadEventQueue<GPUInterface, GPUEvent, GPUEventType, GPU_EVENT_INVALID, GPU_EVENT_SYNC_THREAD, GPU_EVENT_FINISH_EVENT_LOOP> GPUThreadEventQueue;
 
-class GPUCommon : public GPUThreadEventQueue, public GPUDebugInterface
-{
+class GPUCommon : public GPUThreadEventQueue, public GPUDebugInterface {
 public:
 	GPUCommon();
-	virtual ~GPUCommon() {}
+	virtual ~GPUCommon();
 	virtual void Reinitialize();
 
 	virtual void InterruptStart(int listid);
@@ -128,7 +127,7 @@ protected:
 	void UpdatePC(u32 currentPC) {
 		UpdatePC(currentPC, currentPC);
 	}
-	void UpdateState(GPUState state);
+	void UpdateState(GPURunState state);
 	void PopDLQueue();
 	void CheckDrawSync();
 	int  GetNextListIndex();
@@ -163,7 +162,7 @@ protected:
 	recursive_mutex listLock;
 
 	bool interruptRunning;
-	GPUState gpuState;
+	GPURunState gpuState;
 	bool isbreak;
 	u64 drawCompleteTicks;
 	u64 busyTicks;
diff --git a/GPU/GPUInterface.h b/GPU/GPUInterface.h
index dbcbe8214e..15a57fed3c 100644
--- a/GPU/GPUInterface.h
+++ b/GPU/GPUInterface.h
@@ -21,6 +21,7 @@
 #include <string>
 #include <vector>
 
+#include "GPU/GPU.h"
 #include "Core/MemMap.h"
 #include "GPU/ge_constants.h"
 
@@ -54,8 +55,7 @@ enum DisplayListState {
   PSP_GE_DL_STATE_PAUSED = 4,
 };
 
-enum SignalBehavior
-{
+enum SignalBehavior {
 	PSP_GE_SIGNAL_NONE             = 0x00,
 	PSP_GE_SIGNAL_HANDLER_SUSPEND  = 0x01,
 	PSP_GE_SIGNAL_HANDLER_CONTINUE = 0x02,
@@ -91,8 +91,7 @@ enum SignalBehavior
 	PSP_GE_SIGNAL_BREAK2           = 0xFF,
 };
 
-enum GPUState
-{
+enum GPURunState {
 	GPUSTATE_RUNNING = 0,
 	GPUSTATE_DONE = 1,
 	GPUSTATE_STALL = 2,
@@ -100,15 +99,13 @@ enum GPUState
 	GPUSTATE_ERROR = 4,
 };
 
-enum GPUSyncType
-{
+enum GPUSyncType {
 	GPU_SYNC_DRAW,
 	GPU_SYNC_LIST,
 };
 
 // Used for debug
-struct FramebufferInfo
-{
+struct FramebufferInfo {
 	u32 fb_address;
 	u32 z_address;
 	int format;
@@ -117,15 +114,13 @@ struct FramebufferInfo
 	void* fbo;
 };
 
-struct DisplayListStackEntry
-{
+struct DisplayListStackEntry {
 	u32 pc;
 	u32 offsetAddr;
 	u32 baseAddr;
 };
 
-struct DisplayList
-{
+struct DisplayList {
 	int id;
 	u32 startpc;
 	u32 pc;
diff --git a/GPU/GPUState.cpp b/GPU/GPUState.cpp
index 8a6422bea7..50320303b3 100644
--- a/GPU/GPUState.cpp
+++ b/GPU/GPUState.cpp
@@ -18,20 +18,13 @@
 #include "GPU/ge_constants.h"
 #include "GPU/GPUState.h"
 #include "GPU/GLES/ShaderManager.h"
-#include "GPU/GLES/GLES_GPU.h"
-#include "GPU/Null/NullGpu.h"
-#include "GPU/Software/SoftGpu.h"
-
-#if defined(_WIN32)
-#include "GPU/Directx9/helper/global.h"
-#include "GPU/Directx9/GPU_DX9.h"
-#endif
 
 #include "Common/ChunkFile.h"
 #include "Core/CoreParameter.h"
 #include "Core/Config.h"
 #include "Core/System.h"
 #include "Core/MemMap.h"
+
 #ifdef _M_SSE
 #include <emmintrin.h>
 #endif
@@ -41,75 +34,6 @@ GPUgstate MEMORY_ALIGNED16(gstate);
 // Let's align this one too for good measure.
 GPUStateCache MEMORY_ALIGNED16(gstate_c);
 
-GPUInterface *gpu;
-GPUDebugInterface *gpuDebug;
-GPUStatistics gpuStats;
-
-template <typename T>
-static void SetGPU(T *obj) {
-	gpu = obj;
-	gpuDebug = obj;
-}
-
-bool GPU_Init() {
-	switch (PSP_CoreParameter().gpuCore) {
-	case GPU_NULL:
-		SetGPU(new NullGPU());
-		break;
-	case GPU_GLES:
-		SetGPU(new GLES_GPU());
-		break;
-	case GPU_SOFTWARE:
-		SetGPU(new SoftGPU());
-		break;
-	case GPU_DIRECTX9:
-#if defined(_WIN32)
-		SetGPU(new DIRECTX9_GPU());
-#endif
-		break;
-	}
-
-	return gpu != NULL;
-}
-
-void GPU_Shutdown() {
-	delete gpu;
-	gpu = 0;
-	gpuDebug = 0;
-}
-
-void GPU_Reinitialize() {
-	if (gpu) {
-		gpu->Reinitialize();
-	}
-}
-
-void InitGfxState() {
-	memset(&gstate, 0, sizeof(gstate));
-	memset(&gstate_c, 0, sizeof(gstate_c));
-	for (int i = 0; i < 256; i++) {
-		gstate.cmdmem[i] = i << 24;
-	}
-
-	// Lighting is not enabled by default, matrices are zero initialized.
-	memset(gstate.worldMatrix, 0, sizeof(gstate.worldMatrix));
-	memset(gstate.viewMatrix, 0, sizeof(gstate.viewMatrix));
-	memset(gstate.projMatrix, 0, sizeof(gstate.projMatrix));
-	memset(gstate.tgenMatrix, 0, sizeof(gstate.tgenMatrix));
-	memset(gstate.boneMatrix, 0, sizeof(gstate.boneMatrix));
-}
-
-void ShutdownGfxState() {
-}
-
-// When you have changed state outside the psp gfx core,
-// or saved the context and has reloaded it, call this function.
-void ReapplyGfxState() {
-	if (!gpu)
-		return;
-	gpu->ReapplyGfxState();
-}
-
 struct CmdRange {
 	u8 start;
 	u8 end;
diff --git a/GPU/GPUState.h b/GPU/GPUState.h
index 0e1f68f739..cc8e3d339d 100644
--- a/GPU/GPUState.h
+++ b/GPU/GPUState.h
@@ -19,38 +19,19 @@
 
 #include <cmath>
 
-#include "../Globals.h"
-#include "ge_constants.h"
+#include "Globals.h"
+#include "GPU/GPU.h"
+#include "GPU/ge_constants.h"
 #include "Common/Common.h"
 
 class PointerWrap;
 
-// PSP uses a curious 24-bit float - it's basically the top 24 bits of a regular IEEE754 32-bit float.
-// This is used for light positions, transform matrices, you name it.
-inline float getFloat24(unsigned int data)
-{
-	data <<= 8;
-	float f;
-	memcpy(&f, &data, 4);
-	return f;
-}
-
-// in case we ever want to generate PSP display lists...
-inline unsigned int toFloat24(float f) {
-	unsigned int i;
-	memcpy(&i, &f, 4);
-	return i >> 8;
-}
-
-struct GPUgstate
-{
+struct GPUgstate {
 	// Getting rid of this ugly union in favor of the accessor functions
 	// might be a good idea....
-	union
-	{
+	union {
 		u32 cmdmem[256];
-		struct
-		{
+		struct {
 			u32 nop,
 				vaddr,
 				iaddr,
@@ -439,20 +420,12 @@ struct GPUgstate
 	void Restore(u32_le *ptr);
 };
 
-enum SkipDrawReasonFlags {
-	SKIPDRAW_SKIPFRAME = 1,
-	SKIPDRAW_NON_DISPLAYED_FB = 2,   // Skip drawing to FBO:s that have not been displayed.
-	SKIPDRAW_BAD_FB_TEXTURE = 4,
-	SKIPDRAW_WINDOW_MINIMIZED = 8, // Don't draw when the host window is minimized.
-};
-
 bool vertTypeIsSkinningEnabled(u32 vertType);
 
 inline int vertTypeGetNumBoneWeights(u32 vertType) { return 1 + ((vertType & GE_VTYPE_WEIGHTCOUNT_MASK) >> GE_VTYPE_WEIGHTCOUNT_SHIFT); }
 inline int vertTypeGetWeightMask(u32 vertType) { return vertType & GE_VTYPE_WEIGHT_MASK; }
 inline int vertTypeGetTexCoordMask(u32 vertType) { return vertType & GE_VTYPE_TC_MASK; }
 
-
 // The rest is cached simplified/converted data for fast access.
 // Does not need to be saved when saving/restoring context.
 //
@@ -519,66 +492,6 @@ struct GPUStateCache {
 	void DoState(PointerWrap &p);
 };
 
-// TODO: Implement support for these.
-struct GPUStatistics {
-	void Reset() {
-		// Never add a vtable :)
-		memset(this, 0, sizeof(*this));
-	}
-	void ResetFrame() {
-		numDrawCalls = 0;
-		numCachedDrawCalls = 0;
-		numVertsSubmitted = 0;
-		numCachedVertsDrawn = 0;
-		numUncachedVertsDrawn = 0;
-		numTrackedVertexArrays = 0;
-		numTextureInvalidations = 0;
-		numTextureSwitches = 0;
-		numShaderSwitches = 0;
-		numFlushes = 0;
-		numTexturesDecoded = 0;
-		numAlphaTestedDraws = 0;
-		numNonAlphaTestedDraws = 0;
-		msProcessingDisplayLists = 0;
-		vertexGPUCycles = 0;
-		otherGPUCycles = 0;
-		memset(gpuCommandsAtCallLevel, 0, sizeof(gpuCommandsAtCallLevel));
-	}
-
-	// Per frame statistics
-	int numDrawCalls;
-	int numCachedDrawCalls;
-	int numFlushes;
-	int numVertsSubmitted;
-	int numCachedVertsDrawn;
-	int numUncachedVertsDrawn;
-	int numTrackedVertexArrays;
-	int numTextureInvalidations;
-	int numTextureSwitches;
-	int numShaderSwitches;
-	int numTexturesDecoded;
-	double msProcessingDisplayLists;
-	int vertexGPUCycles;
-	int otherGPUCycles;
-	int gpuCommandsAtCallLevel[4];
-
-	int numAlphaTestedDraws;
-	int numNonAlphaTestedDraws;
-
-	// Total statistics, updated by the GPU core in UpdateStats
-	int numVBlanks;
-	int numFlips;
-	int numTextures;
-	int numVertexShaders;
-	int numFragmentShaders;
-	int numShaders;
-	int numFBOs;
-};
-
-bool GPU_Init();
-void GPU_Shutdown();
-void GPU_Reinitialize();
-
 void InitGfxState();
 void ShutdownGfxState();
 void ReapplyGfxState();
@@ -588,9 +501,6 @@ class GPUDebugInterface;
 
 extern GPUgstate gstate;
 extern GPUStateCache gstate_c;
-extern GPUInterface *gpu;
-extern GPUDebugInterface *gpuDebug;
-extern GPUStatistics gpuStats;
 
 inline u32 GPUStateCache::getRelativeAddress(u32 data) const {
 	u32 baseExtended = ((gstate.base & 0x000F0000) << 8) | data;
diff --git a/GPU/GeDisasm.cpp b/GPU/GeDisasm.cpp
index deb00fb397..69d307a2b4 100644
--- a/GPU/GeDisasm.cpp
+++ b/GPU/GeDisasm.cpp
@@ -15,10 +15,11 @@
 // Official git repository and contact information can be found at
 // https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
 
-#include "../Core/MemMap.h"
+#include "Core/MemMap.h"
 
-#include "GPUState.h"
 #include "ge_constants.h"
+#include "GPU/GPU.h"
+#include "GPU/GPUState.h"
 
 void GeDescribeVertexType(u32 op, char *buffer, int len) {
 	bool through = (op & GE_VTYPE_THROUGH_MASK) == GE_VTYPE_THROUGH;
diff --git a/GPU/Null/NullGpu.h b/GPU/Null/NullGpu.h
index c9e8f493d5..db6f9000ed 100644
--- a/GPU/Null/NullGpu.h
+++ b/GPU/Null/NullGpu.h
@@ -17,7 +17,8 @@
 
 #pragma once
 
-#include "../GPUCommon.h"
+#include "GPU/GPUState.h"
+#include "GPU/GPUCommon.h"
 
 class ShaderManager;
 
diff --git a/Qt/GPU.pro b/Qt/GPU.pro
index 3c94228c6b..34729febf6 100644
--- a/Qt/GPU.pro
+++ b/Qt/GPU.pro
@@ -22,6 +22,7 @@ win32 {
 }
 
 SOURCES += $$P/GPU/GeDisasm.cpp \ # GPU
+	$$P/GPU/GPU.cpp \
 	$$P/GPU/GPUCommon.cpp \
 	$$P/GPU/GPUState.cpp \
 	$$P/GPU/Math3D.cpp \
diff --git a/Windows/WndMainWindow.cpp b/Windows/WndMainWindow.cpp
index 35f3945b99..ea470ea04c 100644
--- a/Windows/WndMainWindow.cpp
+++ b/Windows/WndMainWindow.cpp
@@ -66,7 +66,6 @@
 #include "Windows/RawInput.h"
 #include "Windows/TouchInputHandler.h"
 #include "GPU/GPUInterface.h"
-#include "GPU/GPUState.h"
 #include "gfx_es2/gpu_features.h"
 #include "GPU/GLES/TextureScaler.h"
 #include "GPU/GLES/TextureCache.h"
diff --git a/android/dbg.sh b/android/dbg.sh
new file mode 100644
index 0000000000..359d47d40e
--- /dev/null
+++ b/android/dbg.sh
@@ -0,0 +1,2 @@
+export NDK=/c/AndroidNDK
+$NDK/ndk-gdb --nowait
diff --git a/android/jni/Android.mk b/android/jni/Android.mk
index ac498f1bf6..b4361662d9 100644
--- a/android/jni/Android.mk
+++ b/android/jni/Android.mk
@@ -148,6 +148,7 @@ EXEC_AND_LIB_FILES := \
   $(SRC)/Common/Timer.cpp \
   $(SRC)/Common/Misc.cpp \
   $(SRC)/GPU/Math3D.cpp \
+  $(SRC)/GPU/GPU.cpp \
   $(SRC)/GPU/GPUCommon.cpp \
   $(SRC)/GPU/GPUState.cpp \
   $(SRC)/GPU/GeDisasm.cpp \