GS: Implement draw buffering

Account for triangles forming non-overlapping quads in triangles strips/fans.

.github/workflows/scripts/linux/pcsx2-qt.metainfo.xml.in

@@ -20,7 +20,7 @@
   <url type="donation">https://github.com/sponsors/PCSX2</url>
   <url type="faq">https://pcsx2.net/docs/</url>
   <url type="help">https://pcsx2.net/discord</url>
-  <url type="contribute">https://github.com/PCSX2/pcsx2/blob/master/.github/CONTRIBUTING.md</url>
+  <url type="contribute">https://pcsx2.net/docs/category/contributing</url>
   <url type="translate">https://crowdin.com/project/pcsx2-emulator</url>
   <url type="contact">https://mastodon.social/@PCSX2</url>
   <screenshots>

pcsx2-qt/Settings/MemoryCardSettingsWidget.cpp

@@ -40,7 +40,6 @@ MemoryCardSettingsWidget::MemoryCardSettingsWidget(SettingsWindow* settings_dial
 
 	SettingWidgetBinder::BindWidgetToFolderSetting(sif, m_ui.directory, m_ui.browse, m_ui.open, m_ui.reset, "Folders",
 		"MemoryCards", Path::Combine(EmuFolders::DataRoot, "memcards"));
-	SettingWidgetBinder::BindWidgetToBoolSetting(sif, m_ui.automaticManagement, "EmuCore", "McdFolderAutoManage", true);
 
 	setupAdditionalUi();
 
@@ -58,11 +57,6 @@ MemoryCardSettingsWidget::MemoryCardSettingsWidget(SettingsWindow* settings_dial
 	connect(m_ui.deleteCard, &QPushButton::clicked, this, &MemoryCardSettingsWidget::deleteCard);
 
 	refresh();
-
-	dialog()->registerWidgetHelp(m_ui.automaticManagement, tr("Automatically manage saves based on running game"),
-		tr("Checked"),
-		tr("(Folder type only / Card size: Auto) Loads only the relevant booted game saves, ignoring others. Avoids "
-		   "running out of space for saves."));
 }
 
 MemoryCardSettingsWidget::~MemoryCardSettingsWidget() = default;

pcsx2-qt/Settings/MemoryCardSettingsWidget.h

@@ -9,6 +9,7 @@
 #include <QtWidgets/QTreeWidget>
 #include <QtWidgets/QToolButton>
 #include <QtWidgets/QListWidget>
+#include <QtWidgets/QCheckBox>
 
 #include <array>
 #include <optional>

pcsx2-qt/Settings/MemoryCardSettingsWidget.ui

@@ -175,22 +175,6 @@
      </layout>
     </widget>
    </item>
-   <item>
-    <widget class="QGroupBox" name="settingsGroupBox">
-     <property name="title">
-      <string>Settings</string>
-     </property>
-     <layout class="QGridLayout" name="gridLayout_3">
-      <item row="0" column="0">
-       <widget class="QCheckBox" name="automaticManagement">
-        <property name="text">
-         <string>Automatically manage saves based on running game</string>
-        </property>
-       </widget>
-      </item>
-     </layout>
-    </widget>
-   </item>
   </layout>
  </widget>
  <customwidgets>
@@ -211,7 +195,6 @@
   <tabstop>renameCard</tabstop>
   <tabstop>convertCard</tabstop>
   <tabstop>deleteCard</tabstop>
-  <tabstop>automaticManagement</tabstop>
  </tabstops>
  <resources/>
  <connections/>

pcsx2-qt/Translations/pcsx2-qt_en.ts

@@ -11888,12 +11888,12 @@ This action cannot be undone.</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp" line="2788"/>
+        <location filename="../../pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp" line="2713"/>
         <source>Stencil buffers and texture barriers are both unavailable, this will break some graphical effects.</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp" line="5130"/>
+        <location filename="../../pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp" line="5055"/>
         <source>Spin GPU During Readbacks is enabled, but calibrated timestamps are unavailable.  This might be really slow.</source>
         <translation type="unfinished"></translation>
     </message>
@@ -11995,7 +11995,7 @@ Please see our official documentation for more information.</source>
 <context>
     <name>GSDeviceVK</name>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp" line="2153"/>
+        <location filename="../../pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp" line="2078"/>
         <source>Your GPU does not support the required Vulkan features.</source>
         <translation type="unfinished"></translation>
     </message>

pcsx2/Config.h

@@ -1326,7 +1326,6 @@ struct Pcsx2Config
 		UseSavestateSelector : 1,
 		InhibitScreensaver : 1,
 		BackupSavestate : 1,
-		McdFolderAutoManage : 1,
 		ManuallySetRealTimeClock : 1, // passes user-set real-time clock information to cdvd at startup
 		UseSystemLocaleFormat : 1, // presents OS time format instead of yyyy-MM-dd HH:mm:ss for manual RTC
 

pcsx2/GS/GSState.h

@@ -127,37 +127,41 @@ private:
 
 protected:
 	static constexpr int INVALID_ALPHA_MINMAX = 500;
+	static constexpr int MAX_DRAW_BUFFERS = 3;
 
 	GSVertex m_v = {};
 	float m_q = 1.0f;
 	GSVector4i m_scissor_cull_min = {};
 	GSVector4i m_scissor_cull_max = {};
 	GSVector4i m_xyof = {};
+	u32 m_used_buffers_idx = 0;
+	u32 m_current_buffer_idx = 0;
+	bool m_recent_buffer_switch = false;
 
-	struct
+	struct GSVertexBuff
 	{
 		GSVertex* buff;
-		GSVertex* buff_copy;            // same size buffer to copy/modify the original buffer
+		GSVertex* buff_copy; // same size buffer to copy/modify the original buffer
 		u32 head, tail, next, maxcount; // head: first vertex, tail: last vertex + 1, next: last indexed + 1
 		u32 xy_tail;
 		GSVector4i xy[4];
 		GSVector4i xyhead;
-	} m_vertex = {};
+	};
 
-	struct
+	GSVertexBuff m_vertex_buffers[MAX_DRAW_BUFFERS];
+	GSVertexBuff* m_vertex;
+
+	struct GSIndexBuff
 	{
 		u16* buff;
 		u32 tail;
-	} m_index = {};
+	};
 
-	struct
-	{
-		GSVertex* buff;
-		u32 head, tail, next, maxcount; // head: first vertex, tail: last vertex + 1, next: last indexed + 1
-		u32 xy_tail;
-		GSVector4i xy[4];
-		GSVector4i xyhead;
-	} m_draw_vertex = {};
+	GSIndexBuff m_index_buffers[MAX_DRAW_BUFFERS];
+
+	GSIndexBuff* m_index;
+
+	GSVertexBuff m_draw_vertex = {};
 
 	struct
 	{
@@ -165,6 +169,16 @@ protected:
 		u32 tail;
 	} m_draw_index = {};
 
+	struct GSDrawBufferEnv
+	{
+		GSDrawingEnvironment m_env;
+		int m_backed_up_ctx = 0;
+		u32 m_dirty_regs = 0;
+		GSVector4i draw_rect = GSVector4i::zero();
+	};
+
+	GSDrawBufferEnv m_env_buffers[MAX_DRAW_BUFFERS] = {};
+
 	void UpdateContext();
 	void UpdateScissor();
 
@@ -175,6 +189,7 @@ protected:
 	template<u32 prim> void HandleAutoFlush();
 	bool EarlyDetectShuffle(u32 prim);
 	void CheckCLUTValidity(u32 prim);
+	bool CheckOverlapVerts(u32 n);
 
 	template <u32 prim, bool auto_flush> void VertexKick(u32 skip);
 
@@ -241,9 +256,10 @@ public:
 	GSLocalMemory m_mem;
 	GSDrawingEnvironment m_env = {};
 	GSDrawingEnvironment m_prev_env = {};
+	GSDrawingEnvironment m_temp_env = {};
 	const GSDrawingEnvironment* m_draw_env = &m_env;
 	GSDrawingContext* m_context = nullptr;
-	GSVector4i temp_draw_rect = {};
+	GSVector4i temp_draw_rect;
 	std::unique_ptr<GSDumpBase> m_dump;
 	bool m_scissor_invalid = false;
 	bool m_quad_check_valid = false;
@@ -441,7 +457,15 @@ public:
 	virtual void Reset(bool hardware_reset);
 	virtual void UpdateSettings(const Pcsx2Config::GSOptions& old_config);
 
+	void ResetDrawBuffers();
+	void ResetDrawBufferIdx();
+	void FlushBuffers(bool use_flush_reason = false, GSFlushReason flush_reason = GSFlushReason::CONTEXTCHANGE);
+	void PushBuffer();
+	void SetDrawBufferEnv();
+	void SetDrawBuffDirty();
+	bool CanBufferNewDraw();
 	void Flush(GSFlushReason reason);
+	void FlushDraw(GSFlushReason reason);
 	u32 CalcMask(int exp, int max_exp);
 	void FlushPrim();
 	bool TestDrawChanged();
@@ -454,7 +478,7 @@ public:
 
 	virtual void Move();
 
-	GSVector4i GetTEX0Rect();
+	GSVector4i GetTEX0Rect(GSDrawingContext prev_ctx);
 	void CheckWriteOverlap(bool req_write, bool req_read);
 	void Write(const u8* mem, int len);
 	void Read(u8* mem, int len);

pcsx2/GS/Renderers/DX11/D3D.cpp

@@ -426,7 +426,8 @@ GSRendererType D3D::GetPreferredRenderer()
 			if (!feature_level.has_value())
 				return GSRendererType::DX11;
 			else if (feature_level == D3D_FEATURE_LEVEL_12_0)
-				return check_vulkan_supported() ? GSRendererType::VK : GSRendererType::OGL;
+				//return check_vulkan_supported() ? GSRendererType::VK : GSRendererType::OGL;
+				return GSRendererType::DX12;
 			else if (feature_level == D3D_FEATURE_LEVEL_11_0)
 				return GSRendererType::OGL;
 			else
@@ -439,7 +440,8 @@ GSRendererType D3D::GetPreferredRenderer()
 			if (!feature_level.has_value())
 				return GSRendererType::DX11;
 			else if (feature_level == D3D_FEATURE_LEVEL_12_0)
-				return check_vulkan_supported() ? GSRendererType::VK : GSRendererType::DX12;
+				//return check_vulkan_supported() ? GSRendererType::VK : GSRendererType::DX12;
+				return GSRendererType::DX12;
 			else if (feature_level == D3D_FEATURE_LEVEL_11_1)
 				return GSRendererType::DX12;
 			else

pcsx2/GS/Renderers/HW/GSHwHack.cpp

@@ -115,7 +115,7 @@ bool GSHwHack::GSC_IRem(GSRendererHW& r, int& skip)
 		// Detect the deswizzling shuffle from depth, copying the RG and BA separately on each half of the page (ignore the split).
 		if (RTME && RFBP != RTBP0 && RFPSM == PSMCT16S && RTPSM == PSMCT16S)
 		{
-			if (r.m_vt.m_max.p.x == 64 && r.m_vt.m_max.p.y == 64 && r.m_index.tail == 128)
+			if (r.m_vt.m_max.p.x == 64 && r.m_vt.m_max.p.y == 64 && r.m_index->tail == 128)
 			{
 				const GSVector4i draw_size(r.m_vt.m_min.p.x, r.m_vt.m_min.p.y/2, r.m_vt.m_max.p.x, r.m_vt.m_max.p.y/2);
 				const GSVector4i read_size(r.m_vt.m_min.t.x, r.m_vt.m_min.t.y/2, r.m_vt.m_max.t.x, r.m_vt.m_max.t.y/2);
@@ -126,7 +126,7 @@ bool GSHwHack::GSC_IRem(GSRendererHW& r, int& skip)
 		}
 
 		// Following the previous draw, it tries to copy everything read from depth and offset it by 2, for the alternate line channel shuffle (skipped above).
-		if (RTBP0 == (RFBP - 0x20) && r.m_vt.m_max.p.x == 64 && r.m_vt.m_max.p.y == 34 && r.m_index.tail == 2)
+		if (RTBP0 == (RFBP - 0x20) && r.m_vt.m_max.p.x == 64 && r.m_vt.m_max.p.y == 34 && r.m_index->tail == 2)
 		{
 			GSVector4i draw_size(r.m_vt.m_min.p.x, r.m_vt.m_min.p.y - 2.0f, r.m_vt.m_max.p.x, r.m_vt.m_max.p.y - 2.0f);
 			GSVector4i read_size(r.m_vt.m_min.t.x, r.m_vt.m_min.t.y, r.m_vt.m_max.t.x, r.m_vt.m_max.t.y);
@@ -264,10 +264,10 @@ bool GSHwHack::GSC_SFEX3(GSRendererHW& r, int& skip)
 			// Skipping is no good as the copy is used again later, and it causes a weird shimmer/echo effect every other frame.
 
 			// Add on the height from the second part of the draw to the first, to make it one big rect.
-			r.m_vertex.buff[1].XYZ.Y += r.m_vertex.buff[r.m_vertex.tail - 1].XYZ.Y - r.m_context->XYOFFSET.OFY;
-			r.m_vertex.buff[1].V = r.m_vertex.buff[r.m_vertex.tail - 1].V;
-			r.m_vertex.tail = 2;
-			r.m_index.tail = 2;
+			r.m_vertex->buff[1].XYZ.Y += r.m_vertex->buff[r.m_vertex->tail - 1].XYZ.Y - r.m_context->XYOFFSET.OFY;
+			r.m_vertex->buff[1].V = r.m_vertex->buff[r.m_vertex->tail - 1].V;
+			r.m_vertex->tail = 2;
+			r.m_index->tail = 2;
 		}
 	}
 
@@ -332,9 +332,9 @@ bool GSHwHack::GSC_NamcoGames(GSRendererHW& r, int& skip)
 {
 	if (skip == 0)
 	{
-		if (!s_nativeres && r.PRIM->PRIM == GS_SPRITE && RTME && RTEX0.TFX == 1 && RFPSM == RTPSM && RTPSM == PSMCT32 && RFBMSK == 0xFF000000 && r.m_index.tail > 2)
+		if (!s_nativeres && r.PRIM->PRIM == GS_SPRITE && RTME && RTEX0.TFX == 1 && RFPSM == RTPSM && RTPSM == PSMCT32 && RFBMSK == 0xFF000000 && r.m_index->tail > 2)
 		{
-			GSVertex* v = &r.m_vertex.buff[0];
+			GSVertex* v = &r.m_vertex->buff[0];
 			// Don't enable hack on native res.
 			// Fixes ghosting/blur effect and white lines appearing in stages: Moonfit Wilderness, Acid Rain - caused by upscaling.
 			// Game copies the framebuffer as individual page rects with slight offsets (like 1/16 of a pixel etc) which doesn't wokr well with upscaling.
@@ -348,7 +348,7 @@ bool GSHwHack::GSC_NamcoGames(GSRendererHW& r, int& skip)
 			else
 			{
 				// Fixes the alignment of the two halves for the heat haze on the temple stage.
-				for (u32 i = 0; i < r.m_index.tail; i+=2)
+				for (u32 i = 0; i < r.m_index->tail; i+=2)
 				{
 					v[i].XYZ.Y -= 0x8;
 				}
@@ -676,7 +676,7 @@ bool GSHwHack::GSC_NFSUndercover(GSRendererHW& r, int& skip)
 
 	if (RPRIM->TME && Frame.PSM == PSMCT16S && Frame.FBMSK != 0 && Frame.FBW == 10 && Texture.TBW == 1 && Texture.TBP0 == 0x02800 && Texture.PSM == PSMZ16S)
 	{
-		GSVertex* v = &r.m_vertex.buff[1];
+		GSVertex* v = &r.m_vertex->buff[1];
 		v[0].XYZ.X = static_cast<u16>(RCONTEXT->XYOFFSET.OFX + ((r.m_r.z * 2) << 4));
 		v[0].XYZ.Y = static_cast<u16>(RCONTEXT->XYOFFSET.OFY + (r.m_r.w << 4));
 		v[0].U = r.m_r.z << 4;
@@ -687,8 +687,8 @@ bool GSHwHack::GSC_NFSUndercover(GSRendererHW& r, int& skip)
 		r.m_vt.m_max.p.y = r.m_r.w;
 		r.m_vt.m_max.t.x = r.m_r.z;
 		r.m_vt.m_max.t.y = r.m_r.w;
-		r.m_vertex.head = r.m_vertex.tail = r.m_vertex.next = 2;
-		r.m_index.tail = 2;
+		r.m_vertex->head = r.m_vertex->tail = r.m_vertex->next = 2;
+		r.m_index->tail = 2;
 		skip = 79;
 	}
 	else
@@ -840,7 +840,7 @@ bool GSHwHack::GSC_Battlefield2(GSRendererHW& r, int& skip)
 
 			if (dst)
 			{
-				float dc = r.m_vertex.buff[1].XYZ.Z;
+				float dc = r.m_vertex->buff[1].XYZ.Z;
 				g_gs_device->ClearDepth(dst->m_texture, dc * std::exp2(-32.0f));
 			}
 		}
@@ -858,9 +858,9 @@ bool GSHwHack::GSC_BlueTongueGames(GSRendererHW& r, int& skip)
 	if (RPRIM->TME && RTEX0.TW == 3 && RTEX0.TH == 3 && RTEX0.PSM == 0 && RFRAME.FBMSK == 0x00FFFFFF && RFRAME.FBW == 8 && r.PCRTCDisplays.GetResolution().x > 512)
 	{
 		// Check we are drawing stripes
-		for (u32 i = 1; i < r.m_vertex.tail; i+=2)
+		for (u32 i = 1; i < r.m_vertex->tail; i+=2)
 		{
-			int value = (((r.m_vertex.buff[i].XYZ.X - r.m_vertex.buff[i - 1].XYZ.X) + 8) >> 4);
+			int value = (((r.m_vertex->buff[i].XYZ.X - r.m_vertex->buff[i - 1].XYZ.X) + 8) >> 4);
 			if (value != 32)
 				return false;
 		}
@@ -872,18 +872,18 @@ bool GSHwHack::GSC_BlueTongueGames(GSRendererHW& r, int& skip)
 
 		for (int vert = 32; vert < 40; vert+=2)
 		{
-			r.m_vertex.buff[vert].XYZ.X = context->XYOFFSET.OFX + (((vert * 16) << 4) - 8);
-			r.m_vertex.buff[vert].XYZ.Y = context->XYOFFSET.OFY;
-			r.m_vertex.buff[vert].U = (vert * 16) << 4;
-			r.m_vertex.buff[vert].V = 0;
-			r.m_vertex.buff[vert+1].XYZ.X = context->XYOFFSET.OFX + ((((vert * 16) + 32) << 4) - 8);
-			r.m_vertex.buff[vert+1].XYZ.Y = context->XYOFFSET.OFY + (r.PCRTCDisplays.GetResolution().y << 4) + 8;
-			r.m_vertex.buff[vert+1].U = ((vert * 16) + 32) << 4;
-			r.m_vertex.buff[vert+1].V = r.PCRTCDisplays.GetResolution().y << 4;
+			r.m_vertex->buff[vert].XYZ.X = context->XYOFFSET.OFX + (((vert * 16) << 4) - 8);
+			r.m_vertex->buff[vert].XYZ.Y = context->XYOFFSET.OFY;
+			r.m_vertex->buff[vert].U = (vert * 16) << 4;
+			r.m_vertex->buff[vert].V = 0;
+			r.m_vertex->buff[vert+1].XYZ.X = context->XYOFFSET.OFX + ((((vert * 16) + 32) << 4) - 8);
+			r.m_vertex->buff[vert+1].XYZ.Y = context->XYOFFSET.OFY + (r.PCRTCDisplays.GetResolution().y << 4) + 8;
+			r.m_vertex->buff[vert+1].U = ((vert * 16) + 32) << 4;
+			r.m_vertex->buff[vert+1].V = r.PCRTCDisplays.GetResolution().y << 4;
 		}
 
-		/*r.m_vertex.head = r.m_vertex.tail = r.m_vertex.next = 2;
-		r.m_index.tail = 2;*/
+		/*r.m_vertex->head = r.m_vertex->tail = r.m_vertex->next = 2;
+		r.m_index->tail = 2;*/
 
 		r.m_vt.m_max.p.x = r.m_r.z;
 		r.m_vt.m_max.p.y = r.m_r.w;
@@ -917,7 +917,7 @@ bool GSHwHack::GSC_BlueTongueGames(GSRendererHW& r, int& skip)
 
 	// This is the giant dither-like depth buffer. We need this on the CPU *and* the GPU for textures which are
 	// rendered on both.
-	if (context->FRAME.FBW == 8 && r.m_index.tail == 32 && r.PRIM->TME && context->TEX0.TBW == 1)
+	if (context->FRAME.FBW == 8 && r.m_index->tail == 32 && r.PRIM->TME && context->TEX0.TBW == 1)
 	{
 		r.SwPrimRender(r, false, false);
 		return false;
@@ -957,8 +957,8 @@ bool GSHwHack::GSC_MetalGearSolid3(GSRendererHW& r, int& skip)
 	GL_INS("OI_MetalGearSolid3(): %x -> %x, %dx%d, subtract %d", RFBP, RFBP + (RFBW / 2), r.m_r.width(), r.m_r.height(),
 		w_sub);
 
-	for (u32 i = 0; i < r.m_vertex.next; i++)
-		r.m_vertex.buff[i].XYZ.X -= w_sub_fp;
+	for (u32 i = 0; i < r.m_vertex->next; i++)
+		r.m_vertex->buff[i].XYZ.X -= w_sub_fp;
 
 	// No point adjusting the scissor, it just ends up expanding out anyway.. but we do have to fix up the draw rect.
 	r.m_r -= GSVector4i(w_sub);
@@ -971,7 +971,7 @@ bool GSHwHack::GSC_Turok(GSRendererHW& r, int& skip)
 	// Since we can't look in to the future to check this, the options are either rearrange all the pages in a target when the width changes
 	// (very slow, could break a ton of stuff which stores different things in the alpha channel), or this. I choose this.
 
-	if (r.m_index.tail == 6 && RPRIM->PRIM == 4 && !RTME && RFBMSK == 0x00FFFFFF && floor(r.m_vt.m_max.p.x) == 512 && r.m_env.CTXT[r.m_backed_up_ctx].FRAME.FBW == 10 && RFRAME.FBW == 8 && RFPSM == PSMCT32 && RTEST.ATE && RTEST.ATST == ATST_GEQUAL)
+	if (r.m_index->tail == 6 && RPRIM->PRIM == 4 && !RTME && RFBMSK == 0x00FFFFFF && floor(r.m_vt.m_max.p.x) == 512 && r.m_env.CTXT[r.m_backed_up_ctx].FRAME.FBW == 10 && RFRAME.FBW == 8 && RFPSM == PSMCT32 && RTEST.ATE && RTEST.ATST == ATST_GEQUAL)
 	{
 		int num_pages = r.m_cached_ctx.FRAME.FBW * ((floor(r.m_vt.m_max.p.y) + 31) / 32);
 		r.m_cached_ctx.FRAME.FBW = 10;
@@ -988,7 +988,7 @@ bool GSHwHack::GSC_Turok(GSRendererHW& r, int& skip)
 
 bool GSHwHack::OI_PointListPalette(GSRendererHW& r, GSTexture* rt, GSTexture* ds, GSTextureCache::Source* t)
 {
-	const u32 n_vertices = r.m_vertex.next;
+	const u32 n_vertices = r.m_vertex->next;
 	const int w = r.m_r.width();
 	const int h = r.m_r.height();
 	const bool is_copy = !r.PRIM->ABE || (
@@ -1021,7 +1021,7 @@ bool GSHwHack::OI_PointListPalette(GSRendererHW& r, GSTexture* rt, GSTexture* ds
 		const u32 FBP = r.m_cached_ctx.FRAME.Block();
 		const u32 FBW = r.m_cached_ctx.FRAME.FBW;
 		GL_INS("PointListPalette - m_r = <%d, %d => %d, %d>, n_vertices = %u, FBP = 0x%x, FBW = %u", r.m_r.x, r.m_r.y, r.m_r.z, r.m_r.w, n_vertices, FBP, FBW);
-		const GSVertex* RESTRICT v = r.m_vertex.buff;
+		const GSVertex* RESTRICT v = r.m_vertex->buff;
 		const int ox(r.m_context->XYOFFSET.OFX);
 		const int oy(r.m_context->XYOFFSET.OFY);
 		for (size_t i = 0; i < n_vertices; ++i)
@@ -1219,9 +1219,9 @@ bool GSHwHack::OI_ArTonelico2(GSRendererHW& r, GSTexture* rt, GSTexture* ds, GST
 	   buffer to adapt the page width properly.
 	 */
 
-	const GSVertex* v = &r.m_vertex.buff[0];
+	const GSVertex* v = &r.m_vertex->buff[0];
 
-	if (ds && r.m_vertex.next == 2 && !RPRIM->TME && RFRAME.FBW == 10 && v->XYZ.Z == 0 && RTEST.ZTST == ZTST_ALWAYS)
+	if (ds && r.m_vertex->next == 2 && !RPRIM->TME && RFRAME.FBW == 10 && v->XYZ.Z == 0 && RTEST.ZTST == ZTST_ALWAYS)
 	{
 		GL_INS("OI_ArTonelico2");
 		g_gs_device->ClearDepth(ds, 0.0f);

pcsx2/GS/Renderers/HW/GSRendererHWMultiISA.cpp

@@ -39,21 +39,21 @@ bool GSRendererHWFunctions::SwPrimRender(GSRendererHW& hw, bool invalidate_tc, b
 	GSRasterizerData data;
 	GSScanlineGlobalData& gd = data.global;
 
-	hw.m_sw_vertex_buffer.resize(((hw.m_vertex.next + 1) & ~1));
+	hw.m_sw_vertex_buffer.resize(((hw.m_vertex->next + 1) & ~1));
 
 	data.primclass = vt.m_primclass;
 	data.buff = nullptr;
 	data.vertex = hw.m_sw_vertex_buffer.data();
-	data.vertex_count = hw.m_vertex.next;
-	data.index = hw.m_index.buff;
-	data.index_count = hw.m_index.tail;
+	data.vertex_count = hw.m_vertex->next;
+	data.index = hw.m_index->buff;
+	data.index_count = hw.m_index->tail;
 	data.scanmsk_value = env.SCANMSK.MSK;
 
 	// Skip per pixel division if q is constant.
 	// Optimize the division by 1 with a nop. It also means that GS_SPRITE_CLASS must be processed when !vt.m_eq.q.
 	// If you have both GS_SPRITE_CLASS && vt.m_eq.q, it will depends on the first part of the 'OR'.
 	const u32 q_div = !hw.IsMipMapActive() && ((vt.m_eq.q && vt.m_min.t.z != 1.0f) || (!vt.m_eq.q && vt.m_primclass == GS_SPRITE_CLASS));
-	GSVertexSW::s_cvb[vt.m_primclass][PRIM->TME][PRIM->FST][q_div](context, data.vertex, hw.m_vertex.buff, hw.m_vertex.next);
+	GSVertexSW::s_cvb[vt.m_primclass][PRIM->TME][PRIM->FST][q_div](context, data.vertex, hw.m_vertex->buff, hw.m_vertex->next);
 
 	GSVector4i scissor = context->scissor.in;
 	GSVector4i bbox = GSVector4i(vt.m_min.p.floor().xyxy(vt.m_max.p.ceil())).rintersect(scissor);
@@ -524,12 +524,12 @@ bool GSRendererHWFunctions::SwPrimRender(GSRendererHW& hw, bool invalidate_tc, b
 
 		const u32 ofx = context->XYOFFSET.OFX;
 
-		for (int i = 0, j = hw.m_vertex.tail; i < j; i++)
+		for (int i = 0, j = hw.m_vertex->tail; i < j; i++)
 		{
 #if _M_SSE >= 0x501
-			if ((((hw.m_vertex.buff[i].XYZ.X - ofx) + 15) >> 4) & 7) // aligned to 8
+			if ((((hw.m_vertex->buff[i].XYZ.X - ofx) + 15) >> 4) & 7) // aligned to 8
 #else
-			if ((((hw.m_vertex.buff[i].XYZ.X - ofx) + 15) >> 4) & 3) // aligned to 4
+			if ((((hw.m_vertex->buff[i].XYZ.X - ofx) + 15) >> 4) & 3) // aligned to 4
 #endif
 			{
 				gd.sel.notest = 0;

pcsx2/GS/Renderers/HW/GSTextureCache.cpp

@@ -1788,8 +1788,10 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 				         (GSLocalMemory::m_psm[color_psm].bpp >= 16 || (/*possible_shuffle &&*/ GSLocalMemory::m_psm[color_psm].bpp == 8 && GSLocalMemory::m_psm[t->m_TEX0.PSM].bpp >= 16)) && // Channel shuffles or non indexed lookups.
 				         t->m_age <= 1 && (!found_t || t->m_last_draw > dst->m_last_draw) /*&& CanTranslate(bp, bw, psm, block_boundary_rect, t->m_TEX0.TBP0, t->m_TEX0.PSM, t->m_TEX0.TBW)*/)
 				{
+					const u32 end_block = GSLocalMemory::GetEndBlockAddress(bp, TEX0.TBW, TEX0.PSM, r);
+					const u32 adj_bp = (end_block < t->m_TEX0.TBP0 && t->UnwrappedEndBlock() > GS_MAX_BLOCKS) ? (bp + GS_MAX_BLOCKS) : bp;
 					u32 rt_tbw = std::max(1U, t->m_TEX0.TBW);
-					u32 horz_page_offset = ((bp - t->m_TEX0.TBP0) >> 5) % rt_tbw;
+					u32 horz_page_offset = ((adj_bp - t->m_TEX0.TBP0) >> 5) % rt_tbw;
 
 					if (GSLocalMemory::m_psm[psm].bpp == GSLocalMemory::m_psm[t->m_TEX0.PSM].bpp && bw != rt_tbw && block_boundary_rect.height() > GSLocalMemory::m_psm[psm].pgs.y)
 						continue;
@@ -1802,7 +1804,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 					    ((t->m_TEX0.TBW < (horz_page_offset + ((block_boundary_rect.z + GSLocalMemory::m_psm[psm].pgs.x - 1) / GSLocalMemory::m_psm[psm].pgs.x)) ||
 					      (t->m_TEX0.TBW != bw && block_boundary_rect.w > GSLocalMemory::m_psm[psm].pgs.y))))
 					{
-						DbgCon.Warning("BP %x - 16bit bad match for target bp %x bw %d src %d format %d", bp, t->m_TEX0.TBP0, t->m_TEX0.TBW, bw, t->m_TEX0.PSM);
+						DbgCon.Warning("BP %x - 16bit bad match for target bp %x bw %d src %d format %d", adj_bp, t->m_TEX0.TBP0, t->m_TEX0.TBW, bw, t->m_TEX0.PSM);
 						continue;
 					}
 					// Keep note that 2 bw is basically 1 normal page, as bw is in 64 pixels, and 8bit pages are 128 pixels wide, aka 2 bw.
@@ -1814,7 +1816,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 					              ((GSLocalMemory::m_psm[psm].bpp == 32) ? bw : ((bw + 1) / 2)) <= t->m_TEX0.TBW) &&
 					            !(((GSLocalMemory::m_psm[psm].bpp == 32) ? bw : ((bw + 1) / 2)) == rt_tbw)))))
 					{
-						DbgCon.Warning("BP %x - 8bit bad match for target bp %x bw %d src %d format %d", bp, t->m_TEX0.TBP0, t->m_TEX0.TBW, bw, t->m_TEX0.PSM);
+						DbgCon.Warning("BP %x - 8bit bad match for target bp %x bw %d src %d format %d", adj_bp, t->m_TEX0.TBP0, t->m_TEX0.TBW, bw, t->m_TEX0.PSM);
 						continue;
 					}
 					else if (!possible_shuffle && GSLocalMemory::m_psm[psm].bpp <= 8 && TEX0.TBW == 1)
@@ -1852,16 +1854,16 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 						else // Formats are not compatible for normal draws, only shuffles.
 							continue;
 					}
-					if (bp > t->m_TEX0.TBP0)
+					if (adj_bp > t->m_TEX0.TBP0)
 					{
-						if (!region.HasEither() && GSLocalMemory::m_psm[psm].bpp == 32 && (t->m_TEX0.TBW - (((bp - t->m_TEX0.TBP0) >> 5) % rt_tbw)) < static_cast<u32>((block_boundary_rect.width() + 63) / 64))
+						if (!region.HasEither() && GSLocalMemory::m_psm[psm].bpp == 32 && (t->m_TEX0.TBW - (((adj_bp - t->m_TEX0.TBP0) >> 5) % rt_tbw)) < static_cast<u32>((block_boundary_rect.width() + 63) / 64))
 						{
 							DbgCon.Warning("Bad alignmenet");
 							continue;
 						}
 
 						// Make sure it's inside if not a shuffle, sometimes valid areas can get messy, like TOCA Race Driver 2 where it goes over to 480, but it's rounded up to 512 in the shuffle.
-						if (!possible_shuffle && !t->Inside(bp, bw, psm, block_boundary_rect))
+						if (!possible_shuffle && !t->Inside(adj_bp, bw, psm, block_boundary_rect))
 							continue;
 
 						GSVector4i new_rect = (GSLocalMemory::m_psm[color_psm].bpp != GSLocalMemory::m_psm[t->m_TEX0.PSM].bpp && (psm & 0x7) != PSMCT16) ? block_boundary_rect : rect;
@@ -1871,7 +1873,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 						// Hitman Blood Money is an example of this in the theatre.
 						const u32 rt_tbw = (possible_shuffle || bw == 1 || GSUtil::GetChannelMask(psm) != 0x8 || frame.FBW <= bw || frame.FBW == t->m_TEX0.TBW || bw == t->m_TEX0.TBW) ? t->m_TEX0.TBW : frame.FBW;
 
-						const bool can_translate = CanTranslate(bp, bw, src_psm, new_rect, t->m_TEX0.TBP0, t->m_TEX0.PSM, rt_tbw);
+						const bool can_translate = CanTranslate(adj_bp, bw, src_psm, new_rect, t->m_TEX0.TBP0, t->m_TEX0.PSM, rt_tbw);
 						if (can_translate)
 						{
 							const bool swizzle_match = GSLocalMemory::m_psm[src_psm].depth == GSLocalMemory::m_psm[t->m_TEX0.PSM].depth;
@@ -1881,7 +1883,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 
 							if (swizzle_match)
 							{
-								rect = TranslateAlignedRectByPage(t->m_TEX0.TBP0, t->m_end_block, rt_tbw, t->m_TEX0.PSM, t->m_valid, bp, src_psm, bw, new_rect);
+								rect = TranslateAlignedRectByPage(t->m_TEX0.TBP0, t->m_end_block, rt_tbw, t->m_TEX0.PSM, t->m_valid, adj_bp, src_psm, bw, new_rect);
 								rect.x -= new_rect.x;
 								rect.y -= new_rect.y;
 							}
@@ -1901,7 +1903,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 									new_rect.z = (new_rect.z + (page_size.x - 1)) & ~(page_size.x - 1);
 									new_rect.w = (new_rect.w + (page_size.y - 1)) & ~(page_size.y - 1);
 								}
-								rect = TranslateAlignedRectByPage(t, bp & ~((1 << 5) - 1), src_psm, bw, new_rect);
+								rect = TranslateAlignedRectByPage(t, adj_bp & ~((1 << 5) - 1), src_psm, bw, new_rect);
 								rect.x -= new_rect.x & ~(page_size.x - 1);
 								rect.y -= new_rect.y & ~(page_size.y - 1);
 							}
@@ -1933,7 +1935,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 						}
 						else
 						{
-							SurfaceOffset so = ComputeSurfaceOffset(bp, bw, src_psm, new_rect, t);
+							SurfaceOffset so = ComputeSurfaceOffset(adj_bp, bw, src_psm, new_rect, t);
 							if (!so.is_valid && t->Wraps())
 							{
 								// Improves Beyond Good & Evil shadow.
@@ -2692,10 +2694,12 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 			for (auto i = list.begin(); i != list.end(); ++i)
 			{
 				Target* t = *i;
+				const u32 end_block = GSLocalMemory::GetEndBlockAddress(bp, TEX0.TBW, TEX0.PSM, GSVector4i(0, size.y, size.x, size.y + 1));
+				const u32 bp_adj = (end_block < t->m_TEX0.TBP0 && t->UnwrappedEndBlock() > GS_MAX_BLOCKS) ? (bp + GS_MAX_BLOCKS) : bp;
 				const bool half_buffer_match = GSConfig.UserHacks_TextureInsideRt >= GSTextureInRtMode::InsideTargets && TEX0.TBW == t->m_TEX0.TBW && TEX0.PSM == t->m_TEX0.PSM &&
 												bp == GSLocalMemory::GetStartBlockAddress(t->m_TEX0.TBP0, t->m_TEX0.TBW, t->m_TEX0.PSM, GSVector4i(0, size.y, size.x, size.y + 1));
 				// Make sure the target is inside the texture
-				if (t->m_TEX0.TBP0 <= bp && bp <= t->m_end_block && (half_buffer_match || t->Inside(bp, TEX0.TBW, TEX0.PSM, GSVector4i::loadh(size))))
+				if (t->m_TEX0.TBP0 <= bp_adj && bp_adj <= t->UnwrappedEndBlock() && (half_buffer_match || t->Inside(bp_adj, TEX0.TBW, TEX0.PSM, GSVector4i::loadh(size))))
 				{
 					if (dst && (GSState::s_n - dst->m_last_draw) < (GSState::s_n - t->m_last_draw))
 						continue;
@@ -4306,23 +4310,27 @@ bool GSTextureCache::PrepareDownloadTexture(u32 width, u32 height, GSTexture::Fo
 		}
 	}
 }*/
-void GSTextureCache::InvalidateContainedTargets(u32 start_bp, u32 end_bp, u32 write_psm, u32 write_bw)
+void GSTextureCache::InvalidateContainedTargets(u32 start_bp, u32 end_bp, u32 write_psm, u32 write_bw, u32 fb_mask, bool ignore_exact)
 {
-	const bool preserve_alpha = (GSLocalMemory::m_psm[write_psm].trbpp == 24);
-	for (int type = 0; type < 2; type++)
+	const bool preserve_alpha = (GSLocalMemory::m_psm[write_psm].trbpp == 24) || (fb_mask & 0xFF000000);
+	for (int type = 0; type < (ignore_exact ? 1 : 2); type++)
 	{
 		auto& list = m_dst[type];
 		for (auto i = list.begin(); i != list.end();)
 		{
 			Target* const t = *i;
-			if (start_bp != t->m_TEX0.TBP0 && (t->m_TEX0.TBP0 > end_bp || t->UnwrappedEndBlock() < start_bp))
+
+			if ((ignore_exact && start_bp == t->m_TEX0.TBP0) || (start_bp != t->m_TEX0.TBP0 && (t->m_TEX0.TBP0 > end_bp || t->UnwrappedEndBlock() < start_bp)))
 			{
 				++i;
 				continue;
 			}
 
+			const bool compatible_fmt = GSUtil::HasCompatibleBits(t->m_TEX0.PSM, write_psm);
+			const bool compatible_width = std::max(t->m_TEX0.TBW, 1U) == std::max(write_bw, 1U);
+
 			// If not fully contained but they are aligned and or clean, just dirty the area.
-			if (type != DepthStencil && start_bp != t->m_TEX0.TBP0 && (t->m_TEX0.TBP0 < start_bp || t->UnwrappedEndBlock() > end_bp))
+			if ((type != DepthStencil || !compatible_fmt || !compatible_width) && start_bp != t->m_TEX0.TBP0 && (t->m_TEX0.TBP0 < start_bp || t->UnwrappedEndBlock() > end_bp))
 			{
 				const u32 offset = (std::abs(static_cast<int>(start_bp - t->m_TEX0.TBP0)) >> 5) % std::max(1U, t->m_TEX0.TBW);
 				const GSVector4i dirty_rect = t->m_dirty.GetTotalRect(t->m_TEX0, t->m_unscaled_size).rintersect(t->m_valid);
@@ -4338,7 +4346,7 @@ void GSTextureCache::InvalidateContainedTargets(u32 start_bp, u32 end_bp, u32 wr
 					{
 
 						RGBAMask mask;
-						mask._u32 = GSUtil::GetChannelMask(write_psm);
+						mask._u32 = GSUtil::GetChannelMask(write_psm, fb_mask);
 						AddDirtyRectTarget(t, invalidate_r, t->m_TEX0.PSM, t->m_TEX0.TBW, mask, false);
 					}
 
@@ -4368,7 +4376,7 @@ void GSTextureCache::InvalidateContainedTargets(u32 start_bp, u32 end_bp, u32 wr
 
 			t->m_valid_alpha_low &= preserve_alpha;
 			t->m_valid_alpha_high &= preserve_alpha;
-			t->m_valid_rgb = false;
+			t->m_valid_rgb &= (fb_mask & 0x00FFFFFF) != 0;
 
 			// Don't keep partial depth buffers around.
 			if ((!t->m_valid_alpha_low && !t->m_valid_alpha_high && !t->m_valid_rgb) || type == DepthStencil)
@@ -4390,6 +4398,16 @@ void GSTextureCache::InvalidateContainedTargets(u32 start_bp, u32 end_bp, u32 wr
 				delete t;
 				continue;
 			}
+			else if (ignore_exact && GSUtil::HasCompatibleBits(t->m_TEX0.PSM, write_psm))
+			{
+				RGBAMask mask;
+				mask._u32 = GSUtil::GetChannelMask(write_psm, fb_mask);
+
+				AddDirtyRectTarget(t, t->m_valid, t->m_TEX0.PSM, t->m_TEX0.TBW, mask, false);
+				t->m_valid_rgb |= !!(mask._u32 & 0x7);
+				t->m_valid_alpha_low |= mask.c.a;
+				t->m_valid_alpha_high |= mask.c.a;
+			}
 
 			GL_CACHE("TC: InvalidateContainedTargets: Clear RGB valid on %s[%x, %s]", to_string(type), t->m_TEX0.TBP0, GSUtil::GetPSMName(t->m_TEX0.PSM));
 			++i;

pcsx2/GS/Renderers/HW/GSTextureCache.h

@@ -32,8 +32,18 @@ public:
 
 	constexpr static bool CheckOverlap(const u32 a_bp, const u32 a_bp_end, const u32 b_bp, const u32 b_bp_end) noexcept
 	{
-		const bool valid = a_bp <= a_bp_end && b_bp <= b_bp_end;
-		const bool overlap = a_bp <= b_bp_end && a_bp_end >= b_bp;
+		u32 b_bp_start_synced = b_bp;
+		u32 b_bp_end_synced = b_bp_end;
+
+		// Check for wrapping
+		if (a_bp_end > GS_MAX_BLOCKS && b_bp_end < a_bp)
+		{
+			b_bp_start_synced += GS_MAX_BLOCKS;
+			b_bp_end_synced += GS_MAX_BLOCKS;
+		}
+
+		const bool valid = a_bp <= a_bp_end && b_bp_start_synced <= b_bp_end_synced;
+		const bool overlap = a_bp <= b_bp_end_synced && a_bp_end >= b_bp_start_synced;
 		return valid && overlap;
 	}
 
@@ -522,7 +532,7 @@ public:
 	bool HasTargetInHeightCache(u32 bp, u32 fbw, u32 psm, u32 max_age = std::numeric_limits<u32>::max(), bool move_front = true);
 	bool Has32BitTarget(u32 bp);
 
-	void InvalidateContainedTargets(u32 start_bp, u32 end_bp, u32 write_psm = PSMCT32, u32 write_bw = 1);
+	void InvalidateContainedTargets(u32 start_bp, u32 end_bp, u32 write_psm = PSMCT32, u32 write_bw = 1, u32 fb_mask = 0x00000000, bool ignore_exact = false);
 	void InvalidateVideoMemType(int type, u32 bp, u32 write_psm = PSMCT32, u32 write_fbmsk = 0, bool dirty_only = false);
 	void InvalidateVideoMemSubTarget(GSTextureCache::Target* rt);
 	void InvalidateVideoMem(const GSOffset& off, const GSVector4i& r, bool target = true);

pcsx2/GS/Renderers/SW/GSRendererSW.cpp

@@ -320,14 +320,14 @@ void GSRendererSW::RewriteVerticesIfSTOverflow()
 			constexpr int n = GSUtil::GetClassVertexCount(primclass);
 
 			// Make sure the copy buffer is large enough.
-			while (m_vertex.maxcount < m_index.tail)
+			while (m_vertex->maxcount < m_index->tail)
 				GrowVertexBuffer();
 
-			GSVertex* RESTRICT vertex = m_vertex.buff;
-			GSVertex* RESTRICT vertex_copy = m_vertex.buff_copy;
-			u16* RESTRICT index = m_index.buff;
+			GSVertex* RESTRICT vertex = m_vertex->buff;
+			GSVertex* RESTRICT vertex_copy = m_vertex->buff_copy;
+			u16* RESTRICT index = m_index->buff;
 
-			for (int i = 0; i < static_cast<int>(m_index.tail); i += n)
+			for (int i = 0; i < static_cast<int>(m_index->tail); i += n)
 			{
 				GSVector4 stcq[n];
 
@@ -381,18 +381,18 @@ void GSRendererSW::RewriteVerticesIfSTOverflow()
 			}
 
 			// Swap the buffers and fix the counts.
-			std::swap(m_vertex.buff, m_vertex.buff_copy);
-			m_vertex.head = m_vertex.next = m_vertex.tail = m_index.tail;
+			std::swap(m_vertex->buff, m_vertex->buff_copy);
+			m_vertex->head = m_vertex->next = m_vertex->tail = m_index->tail;
 			
 			// Recalculate ST min/max/eq in the vertex trace.
 			GSVector4 tmin = GSVector4::cxpr(FLT_MAX);
 			GSVector4 tmax = GSVector4::cxpr(-FLT_MAX);
-			for (int i = 0; i < static_cast<int>(m_index.tail); i += n)
+			for (int i = 0; i < static_cast<int>(m_index->tail); i += n)
 			{
 				for (int j = 0; j < n; j++)
 				{
-					GSVector4 stcq = GSVector4::cast(GSVector4i(m_vertex.buff[i + j].m[0]));
-					const float Q = (primclass == GS_SPRITE_CLASS) ? stcq.w : m_vertex.buff[i + 1].RGBAQ.Q;
+					GSVector4 stcq = GSVector4::cast(GSVector4i(m_vertex->buff[i + j].m[0]));
+					const float Q = (primclass == GS_SPRITE_CLASS) ? stcq.w : m_vertex->buff[i + 1].RGBAQ.Q;
 					stcq = (stcq / Q).xyzw(stcq);
 					
 					tmin = tmin.min(stcq);
@@ -451,11 +451,11 @@ void GSRendererSW::Draw()
 	SharedData* sd = static_cast<SharedData*>(data.get());
 
 	sd->primclass = m_vt.m_primclass;
-	sd->buff = (u8*)m_vertex_heap.alloc(sizeof(GSVertexSW) * ((m_vertex.next + 1) & ~1) + sizeof(u32) * m_index.tail, 64);
+	sd->buff = (u8*)m_vertex_heap.alloc(sizeof(GSVertexSW) * ((m_vertex->next + 1) & ~1) + sizeof(u32) * m_index->tail, 64);
 	sd->vertex = (GSVertexSW*)sd->buff;
-	sd->vertex_count = m_vertex.next;
-	sd->index = (u16*)(sd->buff + sizeof(GSVertexSW) * ((m_vertex.next + 1) & ~1));
-	sd->index_count = m_index.tail;
+	sd->vertex_count = m_vertex->next;
+	sd->index = (u16*)(sd->buff + sizeof(GSVertexSW) * ((m_vertex->next + 1) & ~1));
+	sd->index_count = m_index->tail;
 	sd->scanmsk_value = m_draw_env->SCANMSK.MSK;
 
 	// skip per pixel division if q is constant.
@@ -463,9 +463,9 @@ void GSRendererSW::Draw()
 	// If you have both GS_SPRITE_CLASS && m_vt.m_eq.q, it will depends on the first part of the 'OR'
 	u32 q_div = !IsMipMapActive() && ((m_vt.m_eq.q && m_vt.m_min.t.z != 1.0f) || (!m_vt.m_eq.q && m_vt.m_primclass == GS_SPRITE_CLASS));
 
-	GSVertexSW::s_cvb[m_vt.m_primclass][PRIM->TME][PRIM->FST][q_div](m_context, sd->vertex, m_vertex.buff, m_vertex.next);
+	GSVertexSW::s_cvb[m_vt.m_primclass][PRIM->TME][PRIM->FST][q_div](m_context, sd->vertex, m_vertex->buff, m_vertex->next);
 
-	std::memcpy(sd->index, m_index.buff, sizeof(u16) * m_index.tail);
+	std::memcpy(sd->index, m_index->buff, sizeof(u16) * m_index->tail);
 
 	GSVector4i scissor = context->scissor.in;
 	GSVector4i bbox = GSVector4i(m_vt.m_min.p.floor().upld(m_vt.m_max.p.floor())) + GSVector4i(0, 0, 1, 1); // right/bottom should be exclusive so +1
@@ -485,12 +485,12 @@ void GSRendererSW::Draw()
 	{
 		int n = GSUtil::GetVertexCount(PRIM->PRIM);
 
-		for (u32 i = 0, j = 0; i < m_index.tail; i += n, j++)
+		for (u32 i = 0, j = 0; i < m_index->tail; i += n, j++)
 		{
 			for (int k = 0; k < n; k++)
 			{
-				GSVertex* v = &m_vertex.buff[m_index.buff[i + k]];
-				GSVertex* vn = &m_vertex.buff[m_index.buff[i + n - 1]];
+				GSVertex* v = &m_vertex->buff[m_index->buff[i + k]];
+				GSVertex* vn = &m_vertex->buff[m_index->buff[i + n - 1]];
 
 				fprintf(s_fp, "%d:%d %f %f %f %f\n",
 					j, k,
@@ -1505,12 +1505,12 @@ bool GSRendererSW::GetScanlineGlobalData(SharedData* data)
 
 		u32 ofx = context->XYOFFSET.OFX;
 
-		for (int i = 0, j = m_vertex.tail; i < j; i++)
+		for (int i = 0, j = m_vertex->tail; i < j; i++)
 		{
 #if _M_SSE >= 0x501
-			if ((((m_vertex.buff[i].XYZ.X - ofx) + 15) >> 4) & 7) // aligned to 8
+			if ((((m_vertex->buff[i].XYZ.X - ofx) + 15) >> 4) & 7) // aligned to 8
 #else
-			if ((((m_vertex.buff[i].XYZ.X - ofx) + 15) >> 4) & 3) // aligned to 4
+			if ((((m_vertex->buff[i].XYZ.X - ofx) + 15) >> 4) & 3) // aligned to 4
 #endif
 			{
 				gd.sel.notest = 0;

pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp

@@ -481,72 +481,26 @@ bool GSDeviceVK::CreateDevice(VkSurfaceKHR surface, bool enable_validation_layer
 	vkGetPhysicalDeviceQueueFamilyProperties(m_physical_device, &queue_family_count, queue_family_properties.data());
 	DevCon.WriteLn("%u vulkan queue families", queue_family_count);
 
-	std::vector<uint32_t> queue_family_users(queue_family_count, 0);
-
+	// Find graphics and present queues.
 	m_graphics_queue_family_index = queue_family_count;
 	m_present_queue_family_index = queue_family_count;
-	u32 present_queue_index = 0;
 	m_spin_queue_family_index = queue_family_count;
 	u32 spin_queue_index = 0;
-
-	// Graphics Queue
 	for (uint32_t i = 0; i < queue_family_count; i++)
 	{
-		if (queue_family_properties[i].queueFlags & VK_QUEUE_GRAPHICS_BIT)
+		VkBool32 graphics_supported = queue_family_properties[i].queueFlags & VK_QUEUE_GRAPHICS_BIT;
+		if (graphics_supported)
 		{
 			m_graphics_queue_family_index = i;
-			queue_family_users[i]++;
-			break;
+			// Quit now, no need for a present queue.
+			if (!surface)
+			{
+				break;
+			}
 		}
-	}
 
-	// Spinwait Queue
-	for (uint32_t i = 0; i < queue_family_count; i++)
-	{
-		if (queue_family_properties[i].queueCount == queue_family_users[i])
-			continue;
-		if (!(queue_family_properties[i].queueFlags & VK_QUEUE_COMPUTE_BIT))
-			continue;
-		if (queue_family_properties[i].timestampValidBits == 0)
-			continue; // We need timing
-
-		if (!(queue_family_properties[i].queueFlags & VK_QUEUE_GRAPHICS_BIT))
+		if (surface)
 		{
-			m_spin_queue_family_index = i;
-			break;
-		}
-		else if (m_spin_queue_family_index == queue_family_count)
-			m_spin_queue_family_index = i;
-	}
-
-	if (m_spin_queue_family_index != queue_family_count)
-	{
-		spin_queue_index = queue_family_users[m_spin_queue_family_index];
-		queue_family_users[m_spin_queue_family_index]++;
-		m_spin_queue_is_graphics_queue = false;
-	}
-	else
-	{
-		// No spare queue? Try the graphics queue.
-		if ((queue_family_properties[m_graphics_queue_family_index].queueFlags & VK_QUEUE_COMPUTE_BIT) &&
-			(queue_family_properties[m_graphics_queue_family_index].timestampValidBits != 0))
-		{
-			m_spin_queue_family_index = m_graphics_queue_family_index;
-			spin_queue_index = 0;
-			m_spin_queue_is_graphics_queue = true;
-		}
-		else
-			m_spin_queue_is_graphics_queue = false;
-	}
-
-	// Present Queue
-	if (surface)
-	{
-		for (uint32_t i = 0; i < queue_family_count; i++)
-		{
-			if (queue_family_properties[i].queueCount == queue_family_users[i])
-				continue;
-
 			VkBool32 present_supported;
 			VkResult res = vkGetPhysicalDeviceSurfaceSupportKHR(m_physical_device, i, surface, &present_supported);
 			if (res != VK_SUCCESS)
@@ -555,48 +509,35 @@ bool GSDeviceVK::CreateDevice(VkSurfaceKHR surface, bool enable_validation_layer
 				return false;
 			}
 
-			if (!present_supported)
-				continue;
-
-			// Perfer aync compute queue
-			if ((queue_family_properties[i].queueFlags & VK_QUEUE_COMPUTE_BIT) &&
-				!(queue_family_properties[i].queueFlags & VK_QUEUE_GRAPHICS_BIT))
-			{
-				m_present_queue_family_index = i;
-				break;
-			}
-			else if (m_present_queue_family_index == queue_family_count)
-				m_present_queue_family_index = i;
-		}
-
-		if (m_present_queue_family_index != queue_family_count)
-		{
-			present_queue_index = queue_family_users[m_present_queue_family_index];
-			queue_family_users[m_present_queue_family_index]++;
-		}
-		else
-		{
-			// No spare queue? Try the graphics queue.
-			VkBool32 present_supported;
-			VkResult res = vkGetPhysicalDeviceSurfaceSupportKHR(m_physical_device, m_graphics_queue_family_index, surface, &present_supported);
-			if (res != VK_SUCCESS)
-			{
-				LOG_VULKAN_ERROR(res, "vkGetPhysicalDeviceSurfaceSupportKHR failed: ");
-				return false;
-			}
-
 			if (present_supported)
 			{
-				m_present_queue_family_index = m_graphics_queue_family_index;
-				present_queue_index = 0;
+				m_present_queue_family_index = i;
+			}
+
+			// Prefer one queue family index that does both graphics and present.
+			if (graphics_supported && present_supported)
+			{
+				break;
 			}
 		}
 	}
-
-	// Swap spin and present to simplify queue priorities logic.
-	if (!m_spin_queue_is_graphics_queue && m_present_queue_family_index == m_spin_queue_family_index)
-		std::swap(spin_queue_index, present_queue_index);
-
+	for (uint32_t i = 0; i < queue_family_count; i++)
+	{
+		// Pick a queue for spinning
+		if (!(queue_family_properties[i].queueFlags & VK_QUEUE_COMPUTE_BIT))
+			continue; // We need compute
+		if (queue_family_properties[i].timestampValidBits == 0)
+			continue; // We need timing
+		const bool queue_is_used = i == m_graphics_queue_family_index || i == m_present_queue_family_index;
+		if (queue_is_used && m_spin_queue_family_index != queue_family_count)
+			continue; // Found a non-graphics queue to use
+		spin_queue_index = 0;
+		m_spin_queue_family_index = i;
+		if (queue_is_used && queue_family_properties[i].queueCount > 1)
+			spin_queue_index = 1;
+		if (!(queue_family_properties[i].queueFlags & VK_QUEUE_GRAPHICS_BIT))
+			break; // Async compute queue, definitely pick this one
+	}
 	if (m_graphics_queue_family_index == queue_family_count)
 	{
 		Console.Error("VK: Failed to find an acceptable graphics queue.");
@@ -614,16 +555,14 @@ bool GSDeviceVK::CreateDevice(VkSurfaceKHR surface, bool enable_validation_layer
 	device_info.flags = 0;
 	device_info.queueCreateInfoCount = 0;
 
-	// Low priority for the spin queue
-	static constexpr float queue_priorities[] = {1.0f, 1.0f, 0.0f}; 
-
+	static constexpr float queue_priorities[] = {1.0f, 0.0f}; // Low priority for the spin queue
 	std::array<VkDeviceQueueCreateInfo, 3> queue_infos;
 	VkDeviceQueueCreateInfo& graphics_queue_info = queue_infos[device_info.queueCreateInfoCount++];
 	graphics_queue_info.sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO;
 	graphics_queue_info.pNext = nullptr;
 	graphics_queue_info.flags = 0;
 	graphics_queue_info.queueFamilyIndex = m_graphics_queue_family_index;
-	graphics_queue_info.queueCount = queue_family_users[m_graphics_queue_family_index];
+	graphics_queue_info.queueCount = 1;
 	graphics_queue_info.pQueuePriorities = queue_priorities;
 
 	if (surface != VK_NULL_HANDLE && m_graphics_queue_family_index != m_present_queue_family_index)
@@ -633,19 +572,19 @@ bool GSDeviceVK::CreateDevice(VkSurfaceKHR surface, bool enable_validation_layer
 		present_queue_info.pNext = nullptr;
 		present_queue_info.flags = 0;
 		present_queue_info.queueFamilyIndex = m_present_queue_family_index;
-		present_queue_info.queueCount = queue_family_users[m_present_queue_family_index];
+		present_queue_info.queueCount = 1;
 		present_queue_info.pQueuePriorities = queue_priorities;
 	}
 
 	if (m_spin_queue_family_index == m_graphics_queue_family_index)
 	{
-		if (spin_queue_index == 1)
-			graphics_queue_info.pQueuePriorities = queue_priorities + 1;
+		if (spin_queue_index != 0)
+			graphics_queue_info.queueCount = 2;
 	}
 	else if (m_spin_queue_family_index == m_present_queue_family_index)
 	{
-		if (spin_queue_index == 1)
-			queue_infos[1].pQueuePriorities = queue_priorities + 1;
+		if (spin_queue_index != 0)
+			queue_infos[1].queueCount = 2; // present queue
 	}
 	else if (m_spin_queue_family_index != queue_family_count)
 	{
@@ -655,7 +594,7 @@ bool GSDeviceVK::CreateDevice(VkSurfaceKHR surface, bool enable_validation_layer
 		spin_queue_info.flags = 0;
 		spin_queue_info.queueFamilyIndex = m_spin_queue_family_index;
 		spin_queue_info.queueCount = 1;
-		spin_queue_info.pQueuePriorities = queue_priorities + 2;
+		spin_queue_info.pQueuePriorities = queue_priorities + 1;
 	}
 
 	device_info.pQueueCreateInfos = queue_infos.data();
@@ -744,11 +683,13 @@ bool GSDeviceVK::CreateDevice(VkSurfaceKHR surface, bool enable_validation_layer
 	vkGetDeviceQueue(m_device, m_graphics_queue_family_index, 0, &m_graphics_queue);
 	if (surface)
 	{
-		vkGetDeviceQueue(m_device, m_present_queue_family_index, present_queue_index, &m_present_queue);
+		vkGetDeviceQueue(m_device, m_present_queue_family_index, 0, &m_present_queue);
 	}
 	m_spinning_supported = m_spin_queue_family_index != queue_family_count &&
 	                       queue_family_properties[m_graphics_queue_family_index].timestampValidBits > 0 &&
 	                       m_device_properties.limits.timestampPeriod > 0;
+	m_spin_queue_is_graphics_queue =
+		m_spin_queue_family_index == m_graphics_queue_family_index && spin_queue_index == 0;
 
 	m_gpu_timing_supported = (m_device_properties.limits.timestampComputeAndGraphics != 0 &&
 							  queue_family_properties[m_graphics_queue_family_index].timestampValidBits > 0 &&
@@ -1348,24 +1289,8 @@ void GSDeviceVK::SubmitCommandBuffer(VKSwapChain* present_swap_chain)
 
 	if (present_swap_chain)
 	{
-		// vkQueuePresentKHR on NVidia dosn't seem to properly wait on the passed semaphore, causing artifacts.
-		// OBS capture with BPM encouters issues, but can apparently occur on the presented image aswell.
-		// Instead, wait on the RenderingFinished semaphore with vkQueueSubmit.
-		const uint32_t present_wait_bits = VK_PIPELINE_STAGE_ALL_COMMANDS_BIT;
-		const VkSubmitInfo submit_present_wait_info = {VK_STRUCTURE_TYPE_SUBMIT_INFO, nullptr, 1,
-			present_swap_chain->GetRenderingFinishedSemaphorePtr(), &present_wait_bits, 0,
-			nullptr, 1, present_swap_chain->GetPresentReadySemaphorePtr()};
-
-		res = vkQueueSubmit(m_present_queue, 1, &submit_present_wait_info, nullptr);
-		if (res != VK_SUCCESS)
-		{
-			LOG_VULKAN_ERROR(res, "vkQueueSubmit failed: ");
-			m_last_submit_failed = true;
-			return;
-		}
-
 		const VkPresentInfoKHR present_info = {VK_STRUCTURE_TYPE_PRESENT_INFO_KHR, nullptr, 1,
-			present_swap_chain->GetPresentReadySemaphorePtr(), 1, present_swap_chain->GetSwapChainPtr(),
+			present_swap_chain->GetRenderingFinishedSemaphorePtr(), 1, present_swap_chain->GetSwapChainPtr(),
 			present_swap_chain->GetCurrentImageIndexPtr(), nullptr};
 
 		present_swap_chain->ResetImageAcquireResult();

pcsx2/GS/Renderers/Vulkan/VKSwapChain.cpp

@@ -490,18 +490,6 @@ bool VKSwapChain::CreateSwapChain()
 			sema.available_semaphore = VK_NULL_HANDLE;
 			return false;
 		}
-
-		res = vkCreateSemaphore(
-			GSDeviceVK::GetInstance()->GetDevice(), &semaphore_info, nullptr, &sema.present_ready_semaphore);
-		if (res != VK_SUCCESS)
-		{
-			LOG_VULKAN_ERROR(res, "vkCreateSemaphore failed: ");
-			vkDestroySemaphore(GSDeviceVK::GetInstance()->GetDevice(), sema.rendering_finished_semaphore, nullptr);
-			vkDestroySemaphore(GSDeviceVK::GetInstance()->GetDevice(), sema.available_semaphore, nullptr);
-			sema.rendering_finished_semaphore = VK_NULL_HANDLE;
-			sema.available_semaphore = VK_NULL_HANDLE;
-			return false;
-		}
 	}
 
 	return true;
@@ -517,8 +505,6 @@ void VKSwapChain::DestroySwapChainImages()
 	m_images.clear();
 	for (auto& it : m_semaphores)
 	{
-		if (it.present_ready_semaphore != VK_NULL_HANDLE)
-			vkDestroySemaphore(GSDeviceVK::GetInstance()->GetDevice(), it.present_ready_semaphore, nullptr);
 		if (it.rendering_finished_semaphore != VK_NULL_HANDLE)
 			vkDestroySemaphore(GSDeviceVK::GetInstance()->GetDevice(), it.rendering_finished_semaphore, nullptr);
 		if (it.available_semaphore != VK_NULL_HANDLE)

pcsx2/GS/Renderers/Vulkan/VKSwapChain.h

@@ -64,14 +64,6 @@ public:
 	{
 		return &m_semaphores[m_current_semaphore].rendering_finished_semaphore;
 	}
-	__fi VkSemaphore GetPresentReadySemaphore() const
-	{
-		return m_semaphores[m_current_semaphore].present_ready_semaphore;
-	}
-	__fi const VkSemaphore* GetPresentReadySemaphorePtr() const
-	{
-		return &m_semaphores[m_current_semaphore].present_ready_semaphore;
-	}
 
 	VkFormat GetTextureFormat() const;
 	VkResult AcquireNextImage();
@@ -100,7 +92,6 @@ private:
 	{
 		VkSemaphore available_semaphore;
 		VkSemaphore rendering_finished_semaphore;
-		VkSemaphore present_ready_semaphore;
 	};
 
 	WindowInfo m_window_info;

pcsx2/Pcsx2Config.cpp

@@ -1923,7 +1923,6 @@ Pcsx2Config::Pcsx2Config()
 {
 	bitset = 0;
 	// Set defaults for fresh installs / reset settings
-	McdFolderAutoManage = true;
 	EnablePatches = true;
 	EnableFastBoot = true;
 	EnableRecordingTools = true;
@@ -1979,7 +1978,6 @@ void Pcsx2Config::LoadSaveCore(SettingsWrapper& wrap)
 	SettingsWrapBitBool(HostFs);
 
 	SettingsWrapBitBool(BackupSavestate);
-	SettingsWrapBitBool(McdFolderAutoManage);
 
 	SettingsWrapBitBool(WarnAboutUnsafeSettings);
 

pcsx2/SIO/Memcard/MemoryCardFile.cpp

@@ -611,7 +611,7 @@ void FileMcd_EmuOpen()
 
 
 	Mcd::impl.Open();
-	Mcd::implFolder.SetFiltering(EmuConfig.McdFolderAutoManage);
+	Mcd::implFolder.SetFiltering(true);
 	Mcd::implFolder.Open();
 }
 
@@ -823,7 +823,7 @@ int FileMcd_ReIndex(uint port, uint slot, const std::string& filter)
 		//	return Mcd::impl.ReIndex( combinedSlot, filter );
 		//	break;
 		case MemoryCardType::Folder:
-			if (!Mcd::implFolder.ReIndex(combinedSlot, EmuConfig.McdFolderAutoManage, filter))
+			if (!Mcd::implFolder.ReIndex(combinedSlot, true, filter))
 				return -1;
 			break;
 		default:
@@ -930,7 +930,7 @@ std::vector<AvailableMcdInfo> FileMcd_GetAvailableCards(bool include_in_use_card
 			Pcsx2Config::McdOptions config;
 			config.Enabled = true;
 			config.Type = MemoryCardType::Folder;
-			sourceFolderMemoryCard.Open(fd.FileName, config, (8 * 1024 * 1024) / FolderMemoryCard::ClusterSize, EmuConfig.McdFolderAutoManage, "");
+			sourceFolderMemoryCard.Open(fd.FileName, config, (8 * 1024 * 1024) / FolderMemoryCard::ClusterSize, true, "");
 
 			mcds.push_back({std::move(basename), std::move(fd.FileName), fd.ModificationTime,
 				MemoryCardType::Folder, MemoryCardFileType::Unknown, 0u, sourceFolderMemoryCard.IsFormatted()});

pcsx2/VMManager.cpp

@@ -2993,8 +2993,6 @@ void VMManager::CheckForMemoryCardConfigChanges(const Pcsx2Config& old_config)
 		}
 	}
 
-	changed |= (EmuConfig.McdFolderAutoManage != old_config.McdFolderAutoManage);
-
 	if (!changed)
 		return;