GS/HW: Add new method of rounding sprites

Idea:
PSSetShaderResource binds local state only (ps_sr_views).
PSUnbindConflictingSRVs binds/updates local and gpu state (ps_sr_views, ps_cached_sr_views and srv gpu state)

What we can do is before checking for conflicts we can preemptively bind new tex on the local state if it's not already bound to the current render target, depth target, then PSUnbindConflictingSRVs will update the gpu state when it checks for conflicts.
Then the following update that happens during the Draw which calls for a gpu upgrade will be ignored since it's already been updated.

bin/resources/GameIndex.yaml

@@ -24781,6 +24781,11 @@ SLES-53820:
 SLES-53821:
   name: "Radio Helicopter II"
   region: "PAL-E"
+  patches:
+    9A695202:
+      content: |-
+        comment=Patch that nops a branch instruction causing a freeze.
+        patch=1,EE,001799AC,word,00000000
 SLES-53824:
   name: "Trapt"
   region: "PAL-E"
@@ -40089,6 +40094,11 @@ SLPM-62624:
   name-sort: "ぷちこぷたー2"
   name-en: "Petit Copter 2"
   region: "NTSC-J"
+  patches:
+    9A695202:
+      content: |-
+        comment=Patch that nops a branch instruction causing a freeze.
+        patch=1,EE,001799B8,word,00000000
 SLPM-62625:
   name: "鬼浜爆走愚連隊 激闘編"
   name-sort: "おにはまばくそうぐれんたい げきとうへん"

bin/resources/shaders/dx11/tfx.fx

@@ -139,8 +139,12 @@ struct PS_OUTPUT
 #endif
 #endif
 #endif
-#if PS_ZCLAMP || PS_ZFLOOR
-	float depth : SV_DepthLessEqual;
+#if (PS_ZFLOOR || PS_ZCLAMP)
+	#if PS_HAS_CONSERVATIVE_DEPTH
+		float depth : SV_DepthLessEqual;
+	#else
+		float depth : SV_Depth;
+	#endif
 #endif
 };
 

bin/resources/shaders/opengl/tfx_fs.glsl

@@ -113,7 +113,7 @@ layout(binding = 3) uniform sampler2D img_prim_min;
 //layout(pixel_center_integer) in vec4 gl_FragCoord;
 #endif
 
-#if (PS_ZFLOOR || PS_ZCLAMP) && HAS_CONSERVATIVE_DEPTH
+#if (PS_ZFLOOR || PS_ZCLAMP) && PS_HAS_CONSERVATIVE_DEPTH
 layout(depth_less) out float gl_FragDepth;
 #endif
 

common/Darwin/DarwinMisc.cpp

@@ -11,6 +11,7 @@
 #include "common/Threading.h"
 #include "common/WindowInfo.h"
 #include "common/HostSys.h"
+#include "fmt/format.h"
 
 #include <csignal>
 #include <cstring>
@@ -19,12 +20,15 @@
 #include <sys/mman.h>
 #include <sys/types.h>
 #include <sys/sysctl.h>
+#include <thread>
 #include <time.h>
 #include <mach/mach_init.h>
 #include <mach/mach_port.h>
 #include <mach/mach_time.h>
 #include <mach/mach_vm.h>
+#include <mach/message.h>
 #include <mach/task.h>
+#include <mach/thread_state.h>
 #include <mach/vm_map.h>
 #include <mutex>
 #include <ApplicationServices/ApplicationServices.h>
@@ -294,7 +298,7 @@ static CPUInfo CalcCPUInfo()
 	std::vector<DarwinMisc::CPUClass> classes = DarwinMisc::GetCPUClasses();
 	out.num_clusters = static_cast<u32>(classes.size());
 	out.num_big_cores = classes.empty() ? 0 : classes[0].num_physical;
-	out.num_threads   = classes.empty() ? 0 : classes[0].num_logical;
+	out.num_threads = classes.empty() ? 0 : classes[0].num_logical;
 	out.num_small_cores = 0;
 	for (std::size_t i = 1; i < classes.size(); i++)
 	{
@@ -568,15 +572,206 @@ void HostSys::EndCodeWrite()
 
 #endif // _M_ARM64
 
+#define USE_MACH_EXCEPTION_PORTS
+
 namespace PageFaultHandler
 {
+#ifdef USE_MACH_EXCEPTION_PORTS
+	static void SignalHandler(mach_port_t port);
+#else
 	static void SignalHandler(int sig, siginfo_t* info, void* ctx);
+#endif
 
 	static std::recursive_mutex s_exception_handler_mutex;
 	static bool s_in_exception_handler = false;
 	static bool s_installed = false;
 } // namespace PageFaultHandler
 
+#ifdef USE_MACH_EXCEPTION_PORTS
+
+#if defined(_M_X86)
+#define THREAD_STATE64_COUNT x86_THREAD_STATE64_COUNT
+#define THREAD_STATE64 x86_THREAD_STATE64
+#define thread_state64_t x86_thread_state64_t
+#elif defined(_M_ARM64)
+#define THREAD_STATE64_COUNT ARM_THREAD_STATE64_COUNT
+#define THREAD_STATE64 ARM_THREAD_STATE64
+#define thread_state64_t arm_thread_state64_t
+#else
+#error Unknown Darwin Platform
+#endif
+
+void PageFaultHandler::SignalHandler(mach_port_t port)
+{
+	Threading::SetNameOfCurrentThread("Mach Exception Thread");
+
+#pragma pack(4)
+	struct
+	{
+		mach_msg_header_t Head;
+		NDR_record_t NDR;
+		exception_type_t exception;
+		mach_msg_type_number_t codeCnt;
+		int64_t code[2];
+		int flavor;
+		mach_msg_type_number_t old_stateCnt;
+		natural_t old_state[THREAD_STATE64_COUNT];
+		mach_msg_trailer_t trailer;
+	} msg_in;
+
+	struct
+	{
+		mach_msg_header_t Head;
+		NDR_record_t NDR;
+		kern_return_t RetCode;
+		int flavor;
+		mach_msg_type_number_t new_stateCnt;
+		natural_t new_state[THREAD_STATE64_COUNT];
+	} msg_out;
+#pragma pack()
+	memset(&msg_in, 0xee, sizeof(msg_in));
+	memset(&msg_out, 0xee, sizeof(msg_out));
+	mach_msg_size_t send_size = 0;
+	mach_msg_option_t option = MACH_RCV_MSG;
+	while (true)
+	{
+		kern_return_t r;
+		if ((r = mach_msg_overwrite(&msg_out.Head, option, send_size, sizeof(msg_in), port,
+				 MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL, &msg_in.Head, 0)))
+		{
+			pxFail(fmt::format("CRITICAL: mach_msg_overwrite: {:x}", r).c_str());
+		}
+
+		if (msg_in.Head.msgh_id == MACH_NOTIFY_NO_SENDERS)
+		{
+			// the other thread exited
+			mach_port_deallocate(mach_task_self(), port);
+			return;
+		}
+
+		if (msg_in.Head.msgh_id != 2406)
+		{
+			pxFailRel("unknown message received");
+			return;
+		}
+
+		if (msg_in.flavor != THREAD_STATE64)
+		{
+			pxFailRel(fmt::format("unknown flavour {}, expected {}", msg_in.flavor, THREAD_STATE64).c_str());
+			return;
+		}
+
+		s_exception_handler_mutex.lock();
+		thread_state64_t* state = (thread_state64_t*)msg_in.old_state;
+
+		HandlerResult result = HandlerResult::ExecuteNextHandler;
+		if (!s_in_exception_handler)
+		{
+			s_in_exception_handler = true;
+
+#ifdef _M_ARM64
+			result = HandlePageFault(reinterpret_cast<void*>(state->__pc), reinterpret_cast<void*>(msg_in.code[1]), (msg_in.code[0] & 2) != 0);
+#else
+			result = HandlePageFault(reinterpret_cast<void*>(state->__rip), reinterpret_cast<void*>(msg_in.code[1]), (msg_in.code[0] & 2) != 0);
+#endif
+			s_in_exception_handler = false;
+		}
+
+		// Set up the reply.
+		msg_out.Head.msgh_bits = MACH_MSGH_BITS(MACH_MSGH_BITS_REMOTE(msg_in.Head.msgh_bits), 0);
+		msg_out.Head.msgh_remote_port = msg_in.Head.msgh_remote_port;
+		msg_out.Head.msgh_local_port = MACH_PORT_NULL;
+		msg_out.Head.msgh_id = msg_in.Head.msgh_id + 100;
+		msg_out.NDR = msg_in.NDR;
+
+		if (result != HandlerResult::ContinueExecution) // cooked
+		{
+			msg_out.RetCode = KERN_FAILURE;
+			msg_out.flavor = 0;
+			msg_out.new_stateCnt = 0;
+
+			// The crash handler on macOS or Linux doesn't use context passed to it
+			// Stubbing it here is fine
+			CrashHandler::CrashSignalHandler(-1, nullptr, nullptr);
+
+			pxFailRel("CrashSignalHandler returned when it should have terminated us!");
+		}
+		else
+		{
+			// Resumes execution right where we left off (re-executes instruction that caused the SIGSEGV)
+			msg_out.RetCode = KERN_SUCCESS;
+			msg_out.flavor = THREAD_STATE64;
+			msg_out.new_stateCnt = THREAD_STATE64_COUNT;
+			memcpy(msg_out.new_state, msg_in.old_state, THREAD_STATE64_COUNT * sizeof(natural_t));
+		}
+
+		msg_out.Head.msgh_size =
+			offsetof(__typeof__(msg_out), new_state) + msg_out.new_stateCnt * sizeof(natural_t);
+		send_size = msg_out.Head.msgh_size;
+		option |= MACH_SEND_MSG;
+
+		s_exception_handler_mutex.unlock();
+	}
+}
+
+bool PageFaultHandler::Install(Error* error)
+{
+	exception_mask_t masks[EXC_TYPES_COUNT];
+	mach_port_t ports[EXC_TYPES_COUNT];
+	exception_behavior_t behaviors[EXC_TYPES_COUNT];
+	thread_state_flavor_t flavors[EXC_TYPES_COUNT];
+	mach_msg_type_number_t count = EXC_TYPES_COUNT;
+
+	kern_return_t r = task_get_exception_ports(mach_task_self(), EXC_MASK_ALL,
+		masks, &count, ports, behaviors, flavors);
+
+	mach_port_t port;
+	if ((r = mach_port_allocate(mach_task_self(), MACH_PORT_RIGHT_RECEIVE, &port)))
+	{
+		pxFailRel(fmt::format("mach_port_allocate: {:x}", r).c_str());
+		return false;
+	}
+
+	std::thread sig_thread(PageFaultHandler::SignalHandler, port);
+	sig_thread.detach();
+
+	if ((r = mach_port_insert_right(mach_task_self(), port, port, MACH_MSG_TYPE_MAKE_SEND)))
+	{
+		mach_port_deallocate(mach_task_self(), port);
+		pxFailRel(fmt::format("mach_port_insert_right: {:x}", r).c_str());
+		return false;
+	}
+
+	task_set_exception_ports(mach_task_self(), EXC_MASK_BAD_ACCESS, MACH_PORT_NULL, EXCEPTION_DEFAULT, THREAD_STATE_NONE);
+
+	if ((r = thread_set_exception_ports(mach_thread_self(), EXC_MASK_BAD_ACCESS, port, EXCEPTION_STATE | MACH_EXCEPTION_CODES, THREAD_STATE64)))
+	{
+		mach_port_deallocate(mach_task_self(), port);
+		pxFailRel(fmt::format("thread_set_exception_ports: {:x}", r).c_str());
+		return false;
+	}
+
+	if ((r = mach_port_mod_refs(mach_task_self(), port, MACH_PORT_RIGHT_SEND, -1)))
+	{
+		mach_port_deallocate(mach_task_self(), port);
+		pxFailRel(fmt::format("mach_port_mod_refs: {:x}", r).c_str());
+		return false;
+	}
+
+	mach_port_t previous;
+	if ((r = mach_port_request_notification(mach_task_self(), port, MACH_NOTIFY_NO_SENDERS, 0, port, MACH_MSG_TYPE_MAKE_SEND_ONCE, &previous)))
+	{
+		mach_port_deallocate(mach_task_self(), port);
+		pxFailRel(fmt::format("mach_port_mod_refs: {:x}", r).c_str());
+		return false;
+	}
+
+	s_installed = true;
+	return true;
+}
+
+#else
+
 void PageFaultHandler::SignalHandler(int sig, siginfo_t* info, void* ctx)
 {
 #if defined(_M_X86)
@@ -644,3 +839,4 @@ bool PageFaultHandler::Install(Error* error)
 	s_installed = true;
 	return true;
 }
+#endif

pcsx2-gsrunner/test_check_dumps.py

@@ -184,7 +184,7 @@ if __name__ == "__main__":
     args = parser.parse_args()
     MAX_DIFF_FRAMES = args.maxframes
 
-    outfile = open(args.outfile, "w")
+    outfile = open(args.outfile, "w", encoding="utf-8")
     write(FILE_HEADER)
 
     if not check_regression_tests(os.path.realpath(args.baselinedir), os.path.realpath(args.testdir)):

pcsx2/CMakeLists.txt

@@ -830,6 +830,7 @@ set(pcsx2HostHeaders
 
 set(pcsx2ImGuiSources
 	ImGui/FullscreenUI.cpp
+	ImGui/FullscreenUI_Settings.cpp
 	ImGui/ImGuiFullscreen.cpp
 	ImGui/ImGuiManager.cpp
 	ImGui/ImGuiOverlays.cpp
@@ -837,6 +838,7 @@ set(pcsx2ImGuiSources
 
 set(pcsx2ImGuiHeaders
 	ImGui/FullscreenUI.h
+	ImGui/FullscreenUI_Internal.h
 	ImGui/ImGuiAnimated.h
 	ImGui/ImGuiFullscreen.h
 	ImGui/ImGuiManager.h

pcsx2/Docs/GameIndex.md

@@ -139,12 +139,15 @@ The clamp modes are also numerically based.
 
 ### GS Hardware Mipmap Fixes
 
-* mipmap                     [`0` or `1` or `2`]    {Off, Basic, Full}                    Default: Automatic (No value, looks up GameDB)
+* mipmap                     [`0` or `1`]    {Off, On}                   Default: On (looks up GameDB)
 * trilinearFiltering         [`0` or `1` or `2`]    {None, Trilinear, Trilinear Ultra}    Default: None (`0`)
 
 ### GS Hardware General Fixes
 
 * beforeDraw                {`OI` with suffix }  {None unless specific game GSC}         Default: Automatic (No value, looks up GameDB) with valid variable name (ex. OI_BurnoutGames)
+
+* moveHandler                {`MV` with suffix }  {None unless specific game GSC}         Default: Automatic (No value, looks up GameDB) with valid variable name (ex. MV_Ico)
+
 * afterDraw                 {`OO` with suffix }  {None unless specific game GSC}         Default: Automatic (No value, looks up GameDB) with valid variable name
 * conservativeFramebuffer   [`0` or `1`]               {Off or On}                             Default: On (`1`)
 * texturePreloading         [`0` or `1` or `2`]        {None, Partial or Full Hash Cache}     Default: None (`0`)
@@ -153,11 +156,17 @@ The clamp modes are also numerically based.
 ### GS Hardware Renderer Fixes
 
 * autoFlush                   [`0` or `1` or `2`]          {Disabled, Enabled (Sprites Only), Enabled (All Primitives)}                            Default: Off (`0`)
+* partialTargetInvalidation         [`0` or `1`]          {Off, On}                               Default: Off (`0`)
+* PCRTCOffsets    [`0` or `1`]          {Off, On}                               Default: Off (`0`)
+
+* PCRTCOverscan    [`0` or `1`]          {Off, On}                               Default: Off (`0`)
+
 * disableDepthSupport         [`0` or `1`]          {Off, On}                               Default: Off (`0`)
 * disablePartialInvalidation  [`0` or `1`]          {Off, On}                               Default: Off (`0`)
 * cpuFramebufferConversion    [`0` or `1`]          {Off, On}                               Default: Off (`0`)
 * preloadFrameData            [`0` or `1`]          {Off, On}                               Default: Off (`0`)
-* textureInsideRT             [`0` or `1`]          {Disabled, Inside Targets, Merge Targets}                                            Default: Off (`0`)
+* textureInsideRT             [`0` or `1`or `2`]          {Disabled, Inside Targets, Merge Targets}                                            Default: Off (`0`)
+* PCRTCOverscan               [`0` or `1`]          {Off, On}                               Default: Off (`0`)
 * PCRTCOverscan               [`0` or `1`]          {Off, On}                               Default: Off (`0`)
 * cpuCLUTRender               [`0` or `1` or `2`]   {Disabled, Normal, Aggressive}           Default: Disabled (`0`)
 * cpuSpriteRenderBW           [Value between `0` to `10`]   {Disabled, 1 (64), 2 (128), 3 (192), 4 (256), 5 (320), 6 (384), 7 (448), 8 (512), 9 (576), 10 (640)} Default: Off (`0`)

pcsx2/Docs/gamedb-schema.json

@@ -229,11 +229,6 @@
               "minimum": 0,
               "maximum": 100000
             },
-            "halfBottomOverride": {
-              "type": "integer",
-              "minimum": 0,
-              "maximum": 1
-            },
             "halfPixelOffset": {
               "type": "integer",
               "minimum": 0,

pcsx2/GS/GSState.cpp

@@ -3560,7 +3560,7 @@ __forceinline bool AreTrianglesQuad(const GSVertex* RESTRICT vin, const u16* RES
 	return are_quad;
 }
 
-__forceinline bool AreTrianglesQuadNonAA(const GSVertex* RESTRICT vin, const u16* RESTRICT index0, const u16* RESTRICT index1)
+bool GSState::AreTrianglesQuadNonAA(const GSVertex* RESTRICT vin, const u16* RESTRICT index0, const u16* RESTRICT index1)
 {
 	u32 v0[3] = {
 		vin[index0[0]].XYZ.U32[0],

pcsx2/GS/GSState.h

@@ -478,6 +478,7 @@ public:
 	
 	template<bool shuffle_check>
 	bool TrianglesAreQuadsImpl();
+	bool AreTrianglesQuadNonAA(const GSVertex* RESTRICT vin, const u16* RESTRICT index0, const u16* RESTRICT index1);
 	bool TrianglesAreQuads(bool shuffle_check = false);
 	template <u32 primclass>
 	PRIM_OVERLAP GetPrimitiveOverlapDrawlistImpl(bool save_drawlist = false, bool save_bbox = false, float bbox_scale = 1.0f);

pcsx2/GS/Renderers/DX11/GSDevice11.cpp

@@ -38,6 +38,15 @@ static bool SupportsTextureFormat(ID3D11Device* dev, DXGI_FORMAT format)
 	return (support & D3D11_FORMAT_SUPPORT_TEXTURE2D) != 0;
 }
 
+static bool IsTextureFormatHWBlendable(ID3D11Device* dev, DXGI_FORMAT format)
+{
+	UINT support;
+	if (FAILED(dev->CheckFormatSupport(format, &support)))
+		return false;
+
+	return (support & D3D11_FORMAT_SUPPORT_RENDER_TARGET) && (support & D3D11_FORMAT_SUPPORT_BLENDABLE);
+}
+
 GSDevice11::GSDevice11()
 {
 	memset(&m_state, 0, sizeof(m_state));
@@ -599,6 +608,15 @@ void GSDevice11::SetFeatures(IDXGIAdapter1* adapter)
 	m_max_texture_size = (m_feature_level >= D3D_FEATURE_LEVEL_11_0) ?
 	                         D3D11_REQ_TEXTURE2D_U_OR_V_DIMENSION :
 	                         D3D10_REQ_TEXTURE2D_U_OR_V_DIMENSION;
+
+	m_conservative_depth = (m_feature_level >= D3D_FEATURE_LEVEL_11_0);
+	m_rgba16_unorm_hw_blend = IsTextureFormatHWBlendable(m_dev.get(), DXGI_FORMAT_R16G16B16A16_UNORM);
+
+	// Let the user know if said features are available.
+	Console.WriteLnFmt("D3D11: DXTn Texture Compression: {}", m_features.dxt_textures ? "Supported" : "Not Supported");
+	Console.WriteLnFmt("D3D11: BC6/7 Texture Compression: {}", m_features.bptc_textures ? "Supported" : "Not Supported");
+	Console.WriteLnFmt("D3D11: Conservative Depth: {}", m_conservative_depth ? "Supported" : "Not Supported");
+	Console.WriteLnFmt("D3D11: RGBA16 UNORM Hardware Blending: {}", m_rgba16_unorm_hw_blend ? "Supported" : "Not Supported");
 }
 
 bool GSDevice11::HasSurface() const
@@ -972,15 +990,15 @@ void GSDevice11::EndPresent()
 	if (m_vsync_mode != GSVSyncMode::FIFO && m_gpu_timing_enabled)
 		PopTimestampQuery();
 
+	// clear out the swap chain view, it might get resized..
+	OMSetRenderTargets(nullptr, nullptr, nullptr, nullptr);
+
 	const UINT sync_interval = static_cast<UINT>(m_vsync_mode == GSVSyncMode::FIFO);
 	const UINT flags = (m_vsync_mode == GSVSyncMode::Disabled && m_using_allow_tearing) ? DXGI_PRESENT_ALLOW_TEARING : 0;
 	m_swap_chain->Present(sync_interval, flags);
 
 	if (m_gpu_timing_enabled)
 		KickTimestampQuery();
-
-	// clear out the swap chain view, it might get resized..
-	OMSetRenderTargets(nullptr, nullptr, nullptr, nullptr);
 }
 
 bool GSDevice11::CreateTimestampQueries()
@@ -1296,6 +1314,8 @@ void GSDevice11::DoStretchRect(GSTexture* sTex, const GSVector4& sRect, GSTextur
 
 void GSDevice11::DoStretchRect(GSTexture* sTex, const GSVector4& sRect, GSTexture* dTex, const GSVector4& dRect, ID3D11PixelShader* ps, ID3D11Buffer* ps_cb, ID3D11BlendState* bs, bool linear)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	CommitClear(sTex);
 
 	const bool draw_in_depth = dTex && dTex->IsDepthStencil();
@@ -1303,6 +1323,10 @@ void GSDevice11::DoStretchRect(GSTexture* sTex, const GSVector4& sRect, GSTextur
 	GSVector2i ds;
 	if (dTex)
 	{
+		// preemptively bind svr if possible
+		if (m_state.cached_rt_view != sTex && m_state.cached_dsv != sTex)
+			PSSetShaderResource(0, sTex);
+
 		// ps unbind conflicting srvs
 		PSUnbindConflictingSRVs(dTex);
 
@@ -1363,11 +1387,17 @@ void GSDevice11::DoStretchRect(GSTexture* sTex, const GSVector4& sRect, GSTextur
 
 void GSDevice11::PresentRect(GSTexture* sTex, const GSVector4& sRect, GSTexture* dTex, const GSVector4& dRect, PresentShader shader, float shaderTime, bool linear)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	CommitClear(sTex);
 
 	GSVector2i ds;
 	if (dTex)
 	{
+		// preemptively bind svr if possible
+		if (m_state.cached_rt_view != sTex && m_state.cached_dsv != sTex)
+			PSSetShaderResource(0, sTex);
+
 		// ps unbind conflicting srvs
 		PSUnbindConflictingSRVs(dTex);
 
@@ -1522,6 +1552,8 @@ void GSDevice11::DrawMultiStretchRects(const MultiStretchRect* rects, u32 num_re
 
 void GSDevice11::DoMultiStretchRects(const MultiStretchRect* rects, u32 num_rects, const GSVector2& ds)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	// Don't use primitive restart here, it ends up slower on some drivers.
 	const u32 vertex_reserve_size = num_rects * 4;
 	const u32 index_reserve_size = num_rects * 6;
@@ -1721,6 +1753,7 @@ void GSDevice11::SetupPS(const PSSelector& sel, const GSHWDrawConfig::PSConstant
 		ShaderMacro sm;
 
 		sm.AddMacro("PIXEL_SHADER", 1);
+		sm.AddMacro("PS_HAS_CONSERVATIVE_DEPTH", m_conservative_depth);
 		sm.AddMacro("PS_FST", sel.fst);
 		sm.AddMacro("PS_WMS", sel.wms);
 		sm.AddMacro("PS_WMT", sel.wmt);
@@ -2162,6 +2195,8 @@ void GSDevice11::RenderImGui()
 
 void GSDevice11::SetupDATE(GSTexture* rt, GSTexture* ds, SetDATM datm, const GSVector4i& bbox)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	// sfex3 (after the capcom logo), vf4 (first menu fading in), ffxii shadows, rumble roses shadows, persona4 shadows
 
 	CommitClear(rt);
@@ -2169,6 +2204,11 @@ void GSDevice11::SetupDATE(GSTexture* rt, GSTexture* ds, SetDATM datm, const GSV
 
 	m_ctx->ClearDepthStencilView(*static_cast<GSTexture11*>(ds), D3D11_CLEAR_STENCIL, 0.0f, 0);
 
+	// preemptively bind svr if possible
+
+	if (m_state.cached_rt_view != rt && m_state.cached_dsv != rt)
+		PSSetShaderResource(0, rt);
+
 	// ps unbind conflicting srvs
 
 	PSUnbindConflictingSRVs(ds);
@@ -2382,6 +2422,7 @@ void GSDevice11::VSSetShader(ID3D11VertexShader* vs, ID3D11Buffer* vs_cb)
 
 void GSDevice11::PSSetShaderResource(int i, GSTexture* sr)
 {
+	// Update local state only, PSUpdateShaderState updates gpu state.
 	m_state.ps_sr_views[i] = *static_cast<GSTexture11*>(sr);
 }
 
@@ -2418,7 +2459,7 @@ void GSDevice11::PSUpdateShaderState(const bool sr_update, const bool ss_update)
 	if (sr_update)
 	{
 		bool sr_changed = false;
-		for (size_t i = 0; i < m_state.ps_sr_views.size(); ++i)
+		for (size_t i = 0; i < MAX_TEXTURES; ++i)
 		{
 			if (m_state.ps_cached_sr_views[i] != m_state.ps_sr_views[i])
 			{
@@ -2430,14 +2471,14 @@ void GSDevice11::PSUpdateShaderState(const bool sr_update, const bool ss_update)
 		if (sr_changed)
 		{
 			m_state.ps_cached_sr_views = m_state.ps_sr_views;
-			m_ctx->PSSetShaderResources(0, m_state.ps_sr_views.size(), m_state.ps_sr_views.data());
+			m_ctx->PSSetShaderResources(0, MAX_TEXTURES, m_state.ps_sr_views.data());
 		}
 	}
 
 	if (ss_update)
 	{
 		bool ss_changed = false;
-		for (size_t i = 0; i < m_state.ps_ss.size(); ++i)
+		for (size_t i = 0; i < MAX_SAMPLERS; ++i)
 		{
 			if (m_state.ps_cached_ss[i] != m_state.ps_ss[i])
 			{
@@ -2449,7 +2490,7 @@ void GSDevice11::PSUpdateShaderState(const bool sr_update, const bool ss_update)
 		if (ss_changed)
 		{
 			m_state.ps_cached_ss = m_state.ps_ss;
-			m_ctx->PSSetSamplers(0, m_state.ps_ss.size(), m_state.ps_ss.data());
+			m_ctx->PSSetSamplers(0, MAX_SAMPLERS, m_state.ps_ss.data());
 		}
 	}
 }
@@ -2458,9 +2499,10 @@ void GSDevice11::PSUnbindConflictingSRVs(GSTexture* tex1, GSTexture* tex2)
 {
 	// Make sure no SRVs are bound using the same texture before binding it to a RTV.
 	bool changed = false;
-	for (size_t i = 0; i < m_state.ps_sr_views.size(); i++)
+	for (size_t i = 0; i < MAX_TEXTURES; i++)
 	{
-		if ((tex1 && m_state.ps_sr_views[i] == *static_cast<GSTexture11*>(tex1)) || (tex2 && m_state.ps_sr_views[i] == *static_cast<GSTexture11*>(tex2)))
+		// We chech against what's currently bound (cached_sr_views), then update local state (ps_sr_views) which calls PSUpdateShaderState to update gpu state.
+		if ((tex1 && m_state.ps_cached_sr_views[i] == *static_cast<GSTexture11*>(tex1)) || (tex2 && m_state.ps_cached_sr_views[i] == *static_cast<GSTexture11*>(tex2)))
 		{
 			m_state.ps_sr_views[i] = nullptr;
 			changed = true;
@@ -2604,7 +2646,6 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 			const GSVector4 dRect(config.colclip_update_area);
 			const GSVector4 sRect = dRect / GSVector4(size.x, size.y).xyxy();
 			StretchRect(colclip_rt, sRect, config.rt, dRect, ShaderConvert::COLCLIP_RESOLVE, false);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			Recycle(colclip_rt);
 
 			g_gs_device->SetColorClipTexture(nullptr);
@@ -2621,7 +2662,7 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 		{
 			config.colclip_update_area = config.drawarea;
 
-			colclip_rt = CreateRenderTarget(rtsize.x, rtsize.y, GSTexture::Format::ColorClip, false);
+			colclip_rt = CreateRenderTarget(rtsize.x, rtsize.y, m_rgba16_unorm_hw_blend ? GSTexture::Format::ColorClip : GSTexture::Format::ColorHDR, false);
 			if (!colclip_rt)
 			{
 				Console.Warning("D3D11: Failed to allocate ColorClip render target, aborting draw.");
@@ -2633,7 +2674,6 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 			const GSVector4 dRect = GSVector4((config.colclip_mode == GSHWDrawConfig::ColClipMode::ConvertOnly) ? GSVector4i::loadh(rtsize) : config.drawarea);
 			const GSVector4 sRect = dRect / GSVector4(rtsize.x, rtsize.y).xyxy();
 			StretchRect(config.rt, sRect, colclip_rt, dRect, ShaderConvert::COLCLIP_INIT, false);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 		}
 	}
 
@@ -2704,20 +2744,29 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 	if (config.tex && config.tex == config.ds)
 		read_only_dsv = static_cast<GSTexture11*>(config.ds)->ReadOnlyDepthStencilView();
 
-	// Should be called before changing local srv state.
-	PSUnbindConflictingSRVs(colclip_rt ? colclip_rt : config.rt, read_only_dsv ? nullptr : config.ds);
-
+	// Preemptively bind svr if possible.
+	// We update the local state, then if there are srv conflicts PSUnbindConflictingSRVs will update the gpu state.
 	if (config.tex)
 	{
 		CommitClear(config.tex);
-		PSSetShaderResource(0, config.tex);
+		if (m_state.cached_rt_view != config.tex && m_state.cached_dsv != config.tex)
+			PSSetShaderResource(0, config.tex);
 	}
 	if (config.pal)
 	{
 		CommitClear(config.pal);
-		PSSetShaderResource(1, config.pal);
+		if (m_state.cached_rt_view != config.pal && m_state.cached_dsv != config.pal)
+			PSSetShaderResource(1, config.pal);
 	}
 
+	// Should be called before changing current gpu state.
+	PSUnbindConflictingSRVs(colclip_rt ? colclip_rt : config.rt, read_only_dsv ? nullptr : config.ds);
+
+	if (config.tex)
+		PSSetShaderResource(0, config.tex);
+	if (config.pal)
+		PSSetShaderResource(1, config.pal);
+
 	SetupVS(config.vs, &config.cb_vs);
 	SetupPS(config.ps, &config.cb_ps, config.sampler);
 
@@ -2818,7 +2867,6 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 			const GSVector4 dRect(config.colclip_update_area);
 			const GSVector4 sRect = dRect / GSVector4(size.x, size.y).xyxy();
 			StretchRect(colclip_rt, sRect, config.rt, dRect, ShaderConvert::COLCLIP_RESOLVE, false);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			Recycle(colclip_rt);
 
 			g_gs_device->SetColorClipTexture(nullptr);
@@ -2847,6 +2895,7 @@ void GSDevice11::SendHWDraw(const GSHWDrawConfig& config, GSTexture* draw_rt_clo
 		{
 			const u32 draw_list_size = static_cast<u32>(config.drawlist->size());
 			const u32 indices_per_prim = config.indices_per_prim;
+			const GSVector4i rtsize(0, 0, draw_rt->GetWidth(), draw_rt->GetHeight());
 
 			pxAssert(config.drawlist && !config.drawlist->empty());
 			pxAssert(config.drawlist_bbox && static_cast<u32>(config.drawlist_bbox->size()) == draw_list_size);
@@ -2855,10 +2904,28 @@ void GSDevice11::SendHWDraw(const GSHWDrawConfig& config, GSTexture* draw_rt_clo
 			{
 				const u32 count = (*config.drawlist)[n] * indices_per_prim;
 
-				GSVector4i bbox = (*config.drawlist_bbox)[n].rintersect(config.drawarea);
+				const GSVector4i original_bbox = (*config.drawlist_bbox)[n].rintersect(config.drawarea);
+				GSVector4i snapped_bbox = original_bbox;
 
-				// Copy only the part needed by the draw.
-				CopyAndBind(bbox);
+				// We don't want the snapped box adjustments when the rect is empty as it might make the copy to pass.
+				// The empty rect itself needs to be handled in renderer properly.
+				if (!snapped_bbox.rempty())
+				{
+					// Aligning bbox to 4 pixel boundaries so copies will be faster using Direct Memory Access,
+					// otherwise it may stall as more commands need to be issued.
+					snapped_bbox.left &= ~3;
+					snapped_bbox.top &= ~3;
+					snapped_bbox.right = (snapped_bbox.right + 3) & ~3;
+					snapped_bbox.bottom = (snapped_bbox.bottom + 3) & ~3;
+
+					// Ensure the new sizes are within bounds.
+					snapped_bbox.left = std::max(0, snapped_bbox.left);
+					snapped_bbox.top = std::max(0, snapped_bbox.top);
+					snapped_bbox.right = std::min(snapped_bbox.right, rtsize.right);
+					snapped_bbox.bottom = std::min(snapped_bbox.bottom, rtsize.bottom);
+				}
+
+				CopyAndBind(snapped_bbox);
 
 				DrawIndexedPrimitive(p, count);
 				p += count;

pcsx2/GS/Renderers/DX11/GSDevice11.h

@@ -136,6 +136,8 @@ private:
 	bool m_using_flip_model_swap_chain = true;
 	bool m_using_allow_tearing = false;
 	bool m_is_exclusive_fullscreen = false;
+	bool m_conservative_depth = false;
+	bool m_rgba16_unorm_hw_blend = false;
 
 	struct
 	{

pcsx2/GS/Renderers/DX12/GSDevice12.cpp

@@ -1407,6 +1407,11 @@ bool GSDevice12::CheckFeatures(const u32& vendor_id)
 		DXGI_FEATURE_PRESENT_ALLOW_TEARING, &allow_tearing_supported, sizeof(allow_tearing_supported));
 	m_allow_tearing_supported = (SUCCEEDED(hr) && allow_tearing_supported == TRUE);
 
+	D3D12_FEATURE_DATA_D3D12_OPTIONS3 device_options3 = {};
+	hr = m_device->CheckFeatureSupport(D3D12_FEATURE_D3D12_OPTIONS3, &device_options3, sizeof(device_options3));
+	m_typed_casting_supported = SUCCEEDED(hr) && device_options3.CastingFullyTypedFormatSupported;
+	Console.WriteLnFmt("D3D12: Casting Fully Typed Formats: {}", m_typed_casting_supported ? "Supported" : "Not Supported");
+
 	D3D12_FEATURE_DATA_D3D12_OPTIONS12 device_options12 = {};
 	hr = m_device->CheckFeatureSupport(D3D12_FEATURE_D3D12_OPTIONS12, &device_options12, sizeof(device_options12));
 	if (SUCCEEDED(hr))
@@ -1470,7 +1475,7 @@ void GSDevice12::LookupNativeFormat(GSTexture::Format format, DXGI_FORMAT* d3d_f
 
 	const auto& mapping = s_format_mapping[static_cast<int>(format)];
 	if (d3d_format)
-		*d3d_format = mapping[0];
+		*d3d_format = (format == GSTexture::Format::DepthStencil && !m_typed_casting_supported) ? DXGI_FORMAT_R32G8X24_TYPELESS : mapping[0];
 	if (srv_format)
 		*srv_format = mapping[1];
 	if (rtv_format)
@@ -1743,6 +1748,8 @@ void GSDevice12::DrawMultiStretchRects(
 void GSDevice12::DoMultiStretchRects(
 	const MultiStretchRect* rects, u32 num_rects, GSTexture12* dTex, ShaderConvert shader)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	// Set up vertices first.
 	const u32 vertex_reserve_size = num_rects * 4 * sizeof(GSVertexPT1);
 	const u32 index_reserve_size = num_rects * 6 * sizeof(u16);
@@ -1880,6 +1887,8 @@ void GSDevice12::DoStretchRect(GSTexture12* sTex, const GSVector4& sRect, GSText
 
 void GSDevice12::DrawStretchRect(const GSVector4& sRect, const GSVector4& dRect, const GSVector2i& ds)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	// ia
 	const float left = dRect.x * 2 / ds.x - 1.0f;
 	const float top = 1.0f - dRect.y * 2 / ds.y;
@@ -3048,6 +3057,7 @@ const ID3DBlob* GSDevice12::GetTFXPixelShader(const GSHWDrawConfig::PSSelector&
 
 	ShaderMacro sm;
 	sm.AddMacro("PIXEL_SHADER", 1);
+	sm.AddMacro("PS_HAS_CONSERVATIVE_DEPTH", 1);
 	sm.AddMacro("PS_FST", sel.fst);
 	sm.AddMacro("PS_WMS", sel.wms);
 	sm.AddMacro("PS_WMT", sel.wmt);
@@ -3900,6 +3910,8 @@ void GSDevice12::SetPSConstantBuffer(const GSHWDrawConfig::PSConstantBuffer& cb)
 
 void GSDevice12::SetupDATE(GSTexture* rt, GSTexture* ds, SetDATM datm, const GSVector4i& bbox)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	GL_PUSH("SetupDATE {%d,%d} %dx%d", bbox.left, bbox.top, bbox.width(), bbox.height());
 
 	const GSVector2i size(ds->GetSize());
@@ -3932,6 +3944,8 @@ void GSDevice12::SetupDATE(GSTexture* rt, GSTexture* ds, SetDATM datm, const GSV
 
 GSTexture12* GSDevice12::SetupPrimitiveTrackingDATE(GSHWDrawConfig& config, PipelineSelector& pipe)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	// How this is done:
 	// - can't put a barrier for the image in the middle of the normal render pass, so that's out
 	// - so, instead of just filling the int texture with INT_MAX, we sample the RT and use -1 for failing values
@@ -4078,7 +4092,6 @@ void GSDevice12::RenderHW(GSHWDrawConfig& config)
 			SetPipeline(m_colclip_finish_pipelines[pipe.ds].get());
 			SetUtilityTexture(colclip_rt, m_point_sampler_cpu);
 			DrawStretchRect(sRect, GSVector4(config.colclip_update_area), rtsize);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 
 			Recycle(colclip_rt);
 			g_gs_device->SetColorClipTexture(nullptr);
@@ -4280,7 +4293,6 @@ void GSDevice12::RenderHW(GSHWDrawConfig& config)
 		const GSVector4 drawareaf = GSVector4((config.colclip_mode == GSHWDrawConfig::ColClipMode::ConvertOnly) ? GSVector4i::loadh(rtsize) : config.drawarea);
 		const GSVector4 sRect(drawareaf / GSVector4(rtsize.x, rtsize.y).xyxy());
 		DrawStretchRect(sRect, GSVector4(drawareaf), rtsize);
-		g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 
 		GL_POP();
 
@@ -4358,7 +4370,6 @@ void GSDevice12::RenderHW(GSHWDrawConfig& config)
 			SetPipeline(m_colclip_finish_pipelines[pipe.ds].get());
 			SetUtilityTexture(colclip_rt, m_point_sampler_cpu);
 			DrawStretchRect(sRect, GSVector4(config.colclip_update_area), rtsize);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 
 			Recycle(colclip_rt);
 			g_gs_device->SetColorClipTexture(nullptr);

pcsx2/GS/Renderers/DX12/GSDevice12.h

@@ -314,7 +314,8 @@ private:
 	bool m_allow_tearing_supported = false;
 	bool m_using_allow_tearing = false;
 	bool m_is_exclusive_fullscreen = false;
-	bool m_enhanced_barriers = true;
+	bool m_enhanced_barriers = false;
+	bool m_typed_casting_supported = false;
 	bool m_device_lost = false;
 
 	ComPtr<ID3D12RootSignature> m_tfx_root_signature;

pcsx2/GS/Renderers/HW/GSHwHack.cpp

@@ -756,7 +756,7 @@ bool GSHwHack::GSC_PolyphonyDigitalGames(GSRendererHW& r, int& skip)
 		config.ps.channel = ChannelFetch_RGB;
 		config.colormask.wrgba = 1 | 2 | 4;
 		r.EndHLEHardwareDraw(false);
-
+		src->m_last_draw = r.s_n;
 		return true;
 	}
 	else
@@ -814,6 +814,7 @@ bool GSHwHack::GSC_PolyphonyDigitalGames(GSRendererHW& r, int& skip)
 			config.ps.channel = ChannelFetch_RED + channel;
 			config.colormask.wrgba = 8;
 			r.EndHLEHardwareDraw(false);
+			dst->m_last_draw = r.s_n;
 		}
 
 		return true;

pcsx2/GS/Renderers/HW/GSRendererHW.cpp

@@ -3846,7 +3846,6 @@ void GSRendererHW::Draw()
 
 						GL_CACHE("HW: RT in RT Z copy back draw %d z_vert_offset %d z_offset %d", s_n, z_vertical_offset, vertical_offset);
 						g_gs_device->StretchRect(g_texture_cache->GetTemporaryZ(), sRect, ds->m_texture, GSVector4(dRect), ShaderConvert::DEPTH_COPY, false);
-						g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 					}
 
 					g_texture_cache->InvalidateTemporaryZ();
@@ -3863,7 +3862,6 @@ void GSRendererHW::Draw()
 					dRect = dRect.min_u32(GSVector4i(ds->m_unscaled_size.x * ds->m_scale, ds->m_unscaled_size.y * ds->m_scale).xyxy());
 
 					g_gs_device->StretchRect(ds->m_texture, sRect, g_texture_cache->GetTemporaryZ(), GSVector4(dRect), ShaderConvert::DEPTH_COPY, false);
-					g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 					z_address_info.rect_since = GSVector4i::zero();
 					g_texture_cache->SetTemporaryZInfo(z_address_info);
 				}
@@ -3911,7 +3909,6 @@ void GSRendererHW::Draw()
 					if (m_cached_ctx.TEST.ZTST > ZTST_ALWAYS || !dRect.rintersect(GSVector4i(GSVector4(m_r) * ds->m_scale)).eq(dRect))
 					{
 						g_gs_device->StretchRect(ds->m_texture, sRect, tex, GSVector4(dRect), ShaderConvert::DEPTH_COPY, false);
-						g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 					}
 					g_texture_cache->SetTemporaryZ(tex);
 					g_texture_cache->SetTemporaryZInfo(ds->m_TEX0.TBP0, page_offset, rt_page_offset);
@@ -4523,7 +4520,6 @@ void GSRendererHW::Draw()
 					{
 						const GSVector4i dRect = GSVector4i(0, 0, g_texture_cache->GetTemporaryZ()->GetWidth(), g_texture_cache->GetTemporaryZ()->GetHeight());
 						g_gs_device->StretchRect(g_texture_cache->GetTemporaryZ(), GSVector4(0.0f, 0.0f, 1.0f, 1.0f), tex, GSVector4(dRect), ShaderConvert::DEPTH_COPY, false);
-						g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 						g_texture_cache->InvalidateTemporaryZ();
 						g_texture_cache->SetTemporaryZ(tex);
 					}
@@ -4715,31 +4711,260 @@ void GSRendererHW::Draw()
 	// Note: second hack corrects only the texture coordinate
 	// Be careful to not correct downscaled targets, this can get messy and break post processing
 	// but it still needs to adjust native stuff from memory as it's not been compensated for upscaling (Dragon Quest 8 font for example).
-	if (CanUpscale() && (m_vt.m_primclass == GS_SPRITE_CLASS) && rt && rt->GetScale() > 1.0f)
+	if (CanUpscale() && ((m_vt.m_primclass == GS_SPRITE_CLASS) || ((m_index.tail % 6 == 0) && m_vt.m_primclass == GS_TRIANGLE_CLASS && m_vt.m_eq.z)) && rt && rt->GetScale() > 1.0f)
 	{
-		const u32 count = m_vertex.next;
-		GSVertex* v = &m_vertex.buff[0];
+		bool valid_format = true;
 
-		// Hack to avoid vertical black line in various games (ace combat/tekken)
-		if (GSConfig.UserHacks_AlignSpriteX)
+		if (m_vt.m_primclass == GS_TRIANGLE_CLASS)
 		{
-			// Note for performance reason I do the check only once on the first
-			// primitive
-			const int win_position = v[1].XYZ.X - context->XYOFFSET.OFX;
-			const bool unaligned_position = ((win_position & 0xF) == 8);
-			const bool unaligned_texture = ((v[1].U & 0xF) == 0) && PRIM->FST; // I'm not sure this check is useful
-			const bool hole_in_vertex = (count < 4) || (v[1].XYZ.X != v[2].XYZ.X);
-			if (hole_in_vertex && unaligned_position && (unaligned_texture || !PRIM->FST))
+			const GSVertex* RESTRICT v = m_vertex.buff;
+			const u16* RESTRICT index = m_index.buff;
+			const size_t count = m_index.tail;
+
+			for (int i = 0; i < count; i += 6)
 			{
-				// Normaly vertex are aligned on full pixels and texture in half
-				// pixels. Let's extend the coverage of an half-pixel to avoid
-				// hole after upscaling
-				for (u32 i = 0; i < count; i += 2)
+				// Non-axis aligned check when only two triangles
+				if (!AreTrianglesQuadNonAA(v, &index[i], &index[i + 3]))
 				{
-					v[i + 1].XYZ.X += 8;
-					// I really don't know if it is a good idea. Neither what to do for !PRIM->FST
-					if (unaligned_texture)
-						v[i + 1].U += 8;
+					valid_format = false;
+					break;
+				}
+			}
+		}
+		if (valid_format)
+		{
+			const u32 count = m_vertex.next;
+			GSVertex* v = &m_vertex.buff[0];
+			u16* idx = &m_index.buff[0];
+			const bool indexed_texture = src && src->m_scale == 1.0f && GSLocalMemory::m_psm[src->m_TEX0.PSM].pal > 0;
+
+			// Shortcut for when the sprites are right up against each other.
+			if (GSConfig.UserHacks_AlignSpriteX && m_vt.m_primclass == GS_SPRITE_CLASS && indexed_texture && PRIM->FST && m_index.tail > 2 &&
+				((v[idx[1]].U == v[idx[2]].U && (v[idx[1]].XYZ.X == v[idx[2]].XYZ.X) && v[idx[0]].V == v[idx[2]].V) || 
+				  (v[idx[1]].V == v[idx[2]].V && (v[idx[1]].XYZ.Y == v[idx[2]].XYZ.Y) && v[idx[0]].U == v[idx[2]].U)))
+			{
+				for (int i = 0; i < m_index.tail; i += 2)
+				{
+					if (!((v[idx[i + 1]].U == v[idx[i + 2]].U && (v[idx[i + 1]].XYZ.X == v[idx[i + 2]].XYZ.X) && v[idx[i]].V == v[idx[i + 2]].V)))
+					{
+						if (v[idx[i + 1]].U & 0x8)
+							v[idx[i + 1]].U -= 8;
+					}
+					if (!((v[idx[i + 1]].V == v[idx[i + 2]].V && (v[idx[i + 1]].XYZ.Y == v[idx[i + 2]].XYZ.Y) && v[idx[i]].U == v[idx[i + 2]].U)))
+					{
+						if (v[idx[i + 1]].V & 0x8)
+							v[idx[i + 11]].V -= 8;
+					}
+				}
+				if (v[idx[m_index.tail - 1]].U & 0x8)
+					v[idx[m_index.tail - 1]].U -= 8;
+
+				if (v[idx[m_index.tail - 1]].V & 0x8)
+					v[idx[m_index.tail - 1]].V -= 8;
+			}
+			else
+			// Hack to avoid vertical black line in various games (ace combat/tekken)
+			if (GSConfig.UserHacks_AlignSpriteX)
+			{
+				// Note for performance reason I do the check only once on the first
+				// primitive
+				const int win_position0 = v[idx[0]].XYZ.X - context->XYOFFSET.OFX;
+				const int win_position1 = v[idx[1]].XYZ.X - context->XYOFFSET.OFX;
+				const bool unaligned_position = ((win_position0 & 0xf) != 0) || ((win_position1 & 0xF) != 0);
+				const int first_s = (v[idx[0]].ST.S / v[idx[0]].RGBAQ.Q) * static_cast<int>(1 << m_cached_ctx.TEX0.TW);
+				const bool unaligned_texture = (PRIM->FST && ((v[1].U & 0xF) == 0)) || (!PRIM->FST && (first_s & 0xF) == 0); // I'm not sure this check is useful
+				const int gap = std::min(static_cast<int>(v[2].XYZ.X), static_cast<int>(v[3].XYZ.X)) - std::max(static_cast<int>(v[0].XYZ.X), static_cast<int>(v[1].XYZ.X));
+				bool hole_in_vertex = (count < 4) || (v[1].XYZ.X != v[2].XYZ.X && ((v[1].XYZ.X ^ v[2].XYZ.X) & 0xF) != 0);
+				const bool is_tri = m_vt.m_primclass == GS_TRIANGLE_CLASS;
+				const int skip = is_tri ? 3 : 2;
+
+				GSVector2 gradient = GSVector2(1.0f, 1.0f);
+
+				if (m_lod.y == 0)
+				{
+					bool can_disable_linear = true;
+					for (u32 i = 0; i < m_index.tail; i += skip)
+					{
+						const int x_offset = (is_tri && v[idx[i]].XYZ.X == v[idx[i + 1]].XYZ.X) ? 2 : 1;
+						const int y_offset = (is_tri && v[idx[i]].XYZ.Y == v[idx[i + 1]].XYZ.Y) ? 2 : 1;
+
+						GSVector2 vert = GSVector2(std::abs(static_cast<float>(v[idx[i]].XYZ.X - v[idx[i + x_offset]].XYZ.X)), std::abs(static_cast<float>(v[idx[i]].XYZ.Y - v[idx[i + y_offset]].XYZ.Y)));
+						GSVector2 tex;
+
+						if (!PRIM->FST)
+						{
+							const int s_offset = (is_tri && (v[idx[i]].ST.S / v[idx[i]].RGBAQ.Q) == (v[idx[i + 1]].ST.S / v[idx[i + 1]].RGBAQ.Q)) ? 2 : 1;
+							const int t_offset = (is_tri && (v[idx[i]].ST.T / v[idx[i]].RGBAQ.Q) == (v[idx[i + 1]].ST.T / v[idx[i + 1]].RGBAQ.Q)) ? 2 : 1;
+							GSVector2 v0, v1;
+							float s = std::min((v[idx[i]].ST.S / v[idx[i]].RGBAQ.Q), 1.0f);
+							float t = std::min((v[idx[i]].ST.T / v[idx[i]].RGBAQ.Q), 1.0f);
+							v0.x = static_cast<int>((1 << m_cached_ctx.TEX0.TW) * s * 16.0f);
+							v0.y = static_cast<int>((1 << m_cached_ctx.TEX0.TH) * t * 16.0f);
+
+							s = std::min((v[idx[i + s_offset]].ST.S / v[idx[i + s_offset]].RGBAQ.Q), 1.0f);
+							t = std::min((v[idx[i + t_offset]].ST.T / v[idx[i + t_offset]].RGBAQ.Q), 1.0f);
+							v1.x = static_cast<int>((1 << m_cached_ctx.TEX0.TW) * s * 16.0f);
+							v1.y = static_cast<int>((1 << m_cached_ctx.TEX0.TH) * t * 16.0f);
+
+							tex = GSVector2(std::abs(v1.x - v0.x), std::abs(v1.y - v0.y));
+						}
+						else
+						{
+							const int u_offset = (is_tri && v[idx[i]].U == v[idx[i + 1]].U) ? 2 : 1;
+							const int v_offset = (is_tri && v[idx[i]].V == v[idx[i + 1]].V) ? 2 : 1;
+							tex = GSVector2(std::abs(static_cast<float>(v[idx[i]].U - v[idx[i + u_offset]].U)), std::abs(static_cast<float>(v[idx[i]].V - v[idx[i + v_offset]].V)));
+						}
+
+						GSVector2 grad = tex / vert;
+						if ((grad.x != gradient.x && grad.x != floor(grad.x)) || (grad.y != gradient.y && grad.y != floor(grad.y)))
+						{
+							can_disable_linear = false;
+							gradient = grad;
+							break;
+						}
+					}
+					if (can_disable_linear)
+					{
+						m_vt.m_filter.linear = 0;
+						m_vt.m_filter.opt_linear = 0;
+					}
+				}
+
+				/*if ((m_vt.m_primclass == GS_SPRITE_CLASS) && hole_in_vertex && unaligned_position && (unaligned_texture || !PRIM->FST))
+				{
+					// Normaly vertex are aligned on full pixels and texture in half
+					// pixels. Let's extend the coverage of an half-pixel to avoid
+					// hole after upscaling
+					for (u32 i = 0; i < count; i += 2)
+					{
+						v[i + 1].XYZ.X += 8;
+						// I really don't know if it is a good idea. Neither what to do for !PRIM->FST
+						if (unaligned_texture)
+							v[i + 1].U += 8;
+					}
+				}
+				else */if (indexed_texture && (gradient.x > (1.0f / 16.0f) || gradient.y > (1.0f / 16.0f)))
+				{
+					const int comparitor = unaligned_position ? 0 : 8;
+
+					if (!PRIM->FST)
+					{
+						int s_offset = (is_tri && (v[idx[0]].ST.S / v[idx[0]].RGBAQ.Q) == (v[idx[1]].ST.S / v[idx[1]].RGBAQ.Q)) ? 2 : 1;
+						int t_offset = (is_tri && (v[idx[0]].ST.T / v[idx[0]].RGBAQ.Q) == (v[idx[1]].ST.T / v[idx[1]].RGBAQ.Q)) ? 2 : 1;
+						GSVector2i v0, v1;
+						float q = v[idx[0]].RGBAQ.Q == 0 ? FLT_MIN : v[idx[0]].RGBAQ.Q;
+						float s = v[idx[0]].ST.S / q;
+						float t = v[idx[0]].ST.T / q;
+						v0.x = static_cast<int>((1 << m_cached_ctx.TEX0.TW) * s * 16.0f);
+						v0.y = static_cast<int>((1 << m_cached_ctx.TEX0.TH) * t * 16.0f);
+
+						q = v[idx[s_offset]].RGBAQ.Q == 0 ? FLT_MIN : v[idx[s_offset]].RGBAQ.Q;
+						s = v[idx[s_offset]].ST.S / q;
+						t = v[idx[t_offset]].ST.T / q;
+						v1.x = static_cast<int>((1 << m_cached_ctx.TEX0.TW) * s * 16.0f);
+						v1.y = static_cast<int>((1 << m_cached_ctx.TEX0.TH) * t * 16.0f);
+						bool small_texture = std::abs(v1.x - v0.x) <= (64 * 16) || std::abs(v1.y - v0.y) <= (64 * 16);
+						bool offset_texture_x = (m_vt.IsLinear() || ((v0.x & 0xF) && (v1.x & 0xF))) && small_texture; // Keep them relatively small to avoid full screen stuff.
+						bool offset_texture_y = (m_vt.IsLinear() || ((v0.y & 0xF) && (v1.y & 0xF))) && small_texture;
+
+						//if (offset_texture_x && offset_texture_y)
+						{
+							for (u32 i = m_index.buff[0]; i < m_index.tail; i += skip)
+							{
+								GSVector2 st;
+
+								if (gradient.x > (1.0f / 16.0f))
+								{
+									float largest_s = std::max(is_tri ? (v[i + 2].ST.S / v[i + 2].RGBAQ.Q) : static_cast<float>(0), std::max((v[i].ST.S / v[i].RGBAQ.Q), (v[i + 1].ST.S / v[i + 1].RGBAQ.Q)));
+									float smallest_s = std::min(is_tri ? (v[i + 2].ST.S / v[i + 2].RGBAQ.Q) : static_cast<float>(0), std::min((v[i].ST.S / v[i].RGBAQ.Q), (v[i + 1].ST.S / v[i + 1].RGBAQ.Q)));
+
+									for (int j = 0; j < skip; j++)
+									{
+										q = v[i + j].RGBAQ.Q == 0 ? FLT_MIN : v[i + j].RGBAQ.Q;
+										float s = v[i + j].ST.S / q;
+										st.x = static_cast<float>(1 << m_cached_ctx.TEX0.TW) * s;
+										if ((v[i + j].ST.S / q) == largest_s)
+											v[i + j].ST.S = ((st.x - 0.5f) / static_cast<float>(1 << m_cached_ctx.TEX0.TW)) * q;
+										// Check the minimap in Persona 3.
+										if ((static_cast<int>(st.x * 16) & 0x8) == comparitor && (v[i + j].ST.S / q) == smallest_s)
+											v[i + j].ST.S = (std::max(0.0f, (st.x - 0.5f)) / static_cast<float>(1 << m_cached_ctx.TEX0.TW)) * q;
+									}
+								}
+
+								if (gradient.y > (1.0f / 16.0f))
+								{
+									float largest_t = std::max(is_tri ? (v[i + 2].ST.T / v[i + 2].RGBAQ.Q) : static_cast<float>(0), std::max((v[i].ST.T / v[i].RGBAQ.Q), (v[i + 1].ST.T / v[i + 1].RGBAQ.Q)));
+									float smallest_t = std::min(is_tri ? (v[i + 2].ST.T / v[i + 2].RGBAQ.Q) : static_cast<float>(0), std::min((v[i].ST.T / v[i].RGBAQ.Q), (v[i + 1].ST.T / v[i + 1].RGBAQ.Q)));
+									for (int j = 0; j < skip; j++)
+									{
+										q = v[i + j].RGBAQ.Q == 0 ? FLT_MIN : v[i + j].RGBAQ.Q;
+										float t = v[i + j].ST.T / q;
+										st.y = static_cast<float>(1 << m_cached_ctx.TEX0.TH) * t;
+										if ((v[i + j].ST.T / q) == largest_t)
+											v[i + j].ST.T = ((st.y - 0.5f) / static_cast<float>(1 << m_cached_ctx.TEX0.TH)) * q;
+										// Check the minimap in Persona 3.
+										if ((static_cast<int>(st.y * 16) & 0x8) == comparitor && (v[i + j].ST.T / q) == smallest_t)
+											v[i + j].ST.T = (std::max(0.0f, (st.y - 0.5f)) / static_cast<float>(1 << m_cached_ctx.TEX0.TH)) * q;
+									}
+								}
+							}
+						}
+					}
+					else
+					{
+						int u_offset = (is_tri && (v[0].U == v[1].U)) ? 2 : 1;
+						int v_offset = (is_tri && (v[0].V == v[1].V)) ? 2 : 1;
+						bool small_texture = std::abs(static_cast<int>(v[idx[u_offset]].U) - static_cast<int>(v[idx[0]].U)) <= (64 * 16) || std::abs(static_cast<int>(v[idx[v_offset]].V) - static_cast<int>(v[idx[0]].V)) <= (64 * 16);
+						bool offset_texture_x = (m_vt.IsLinear() || ((v[0].U & 0xF) && (v[idx[u_offset]].U & 0xF))) && small_texture;
+						bool offset_texture_y = (m_vt.IsLinear() || ((v[0].V & 0xF) && (v[idx[v_offset]].V & 0xF))) && small_texture;
+
+						if (offset_texture_x && offset_texture_y)
+						{
+							for (u32 i = m_index.buff[0]; i < m_index.tail; i += skip)
+							{
+								if (gradient.x > (1.0f / 16.0f))
+								{
+									const u16 largest_u = std::max(is_tri ? v[idx[i + 2]].U : static_cast<u16>(0), std::max(v[idx[i]].U, v[idx[i + 1]].U));
+
+									if ((v[idx[i]].U & 0x8) == comparitor && v[idx[i]].U == largest_u)
+										v[idx[i]].U = std::max(static_cast<int>(v[idx[i]].U) - 8, 0);
+
+									if ((v[idx[i + 1]].U & 0x8) == comparitor && v[idx[i + 1]].U == largest_u)
+										v[idx[i + 1]].U = std::max(static_cast<int>(v[idx[i + 1]].U) - 8, 0);
+
+									if (is_tri && (v[idx[i + 2]].U & 0x8) == comparitor && v[idx[i + 2]].U == largest_u)
+										v[idx[i + 2]].U = std::max(static_cast<int>(v[idx[i + 2]].U) - 8, 0);
+								}
+
+								if (gradient.y > (1.0f / 16.0f))
+								{
+									const u16 largest_v = std::max(is_tri ? v[idx[i + 2]].V : static_cast<u16>(0), std::max(v[idx[i]].V, v[idx[i + 1]].V));
+
+									if ((v[idx[i]].V & 0x8) == comparitor && v[idx[i]].V == largest_v)
+										v[idx[i]].V = std::max(static_cast<int>(v[idx[i]].V) - 8, 0);
+
+									if ((v[idx[i + 1]].V & 0x8) == comparitor && v[idx[i + 1]].V == largest_v)
+										v[idx[i + 1]].V = std::max(static_cast<int>(v[idx[i + 1]].V) - 8, 0);
+
+									if (is_tri && (v[idx[i + 2]].V & 0x8) == comparitor && v[idx[i + 2]].V == largest_v)
+										v[idx[i + 2]].V = std::max(static_cast<int>(v[idx[i + 2]].V) - 8, 0);
+								}
+							}
+						}
+					}
+				}
+				else if ((m_vt.m_primclass == GS_SPRITE_CLASS) && hole_in_vertex && unaligned_position && (unaligned_texture || !PRIM->FST))
+				{
+					// Normaly vertex are aligned on full pixels and texture in half
+					// pixels. Let's extend the coverage of an half-pixel to avoid
+					// hole after upscaling
+					for (u32 i = 0; i < count; i += 2)
+					{
+						v[i + 1].XYZ.X += 8;
+						// I really don't know if it is a good idea. Neither what to do for !PRIM->FST
+						if (unaligned_texture)
+							v[i + 1].U += 8;
+					}
 				}
 			}
 		}
@@ -4830,7 +5055,6 @@ void GSRendererHW::Draw()
 
 					GL_CACHE("HW: RT in RT Z copy back draw %d z_vert_offset %d rt_vert_offset %d z_horz_offset %d rt_horz_offset %d", s_n, z_vertical_offset, vertical_offset, z_horizontal_offset, horizontal_offset);
 					g_gs_device->StretchRect(g_texture_cache->GetTemporaryZ(), sRect, ds->m_texture, GSVector4(dRect), ShaderConvert::DEPTH_COPY, false);
-					g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 				}
 				else if (m_temp_z_full_copy)
 				{
@@ -4843,7 +5067,6 @@ void GSRendererHW::Draw()
 
 					GL_CACHE("HW: RT in RT Z copy back draw %d z_vert_offset %d z_offset %d", s_n, z_vertical_offset, vertical_offset);
 					g_gs_device->StretchRect(g_texture_cache->GetTemporaryZ(), sRect, ds->m_texture, GSVector4(dRect), ShaderConvert::DEPTH_COPY, false);
-					g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 				}
 
 				m_temp_z_full_copy = false;
@@ -5910,6 +6133,7 @@ void GSRendererHW::EmulateBlending(int rt_alpha_min, int rt_alpha_max, const boo
 	{
 		case AccBlendLevel::Maximum:
 			sw_blending |= true;
+			accumulation_blend &= (GSLocalMemory::m_psm[m_cached_ctx.FRAME.PSM].bpp == 32);
 			[[fallthrough]];
 		case AccBlendLevel::Full:
 			sw_blending |= m_conf.ps.blend_a != m_conf.ps.blend_b && alpha_c0_high_max_one;
@@ -7091,8 +7315,6 @@ __ri void GSRendererHW::HandleTextureHazards(const GSTextureCache::Target* rt, c
 
 	if (m_downscale_source)
 	{
-		g_perfmon.Put(GSPerfMon::TextureCopies, 1);
-
 		// Can't use box filtering on depth (yet), or fractional scales.
 		if (src_target->m_texture->IsDepthStencil() || std::floor(src_target->GetScale()) != src_target->GetScale())
 		{
@@ -7120,9 +7342,7 @@ __ri void GSRendererHW::HandleTextureHazards(const GSTextureCache::Target* rt, c
 	}
 	else
 	{
-		g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 		const GSVector4i offset = copy_range - GSVector4i(copy_dst_offset).xyxy();
-
 		// Adjust for bilinear, must be done after calculating offset.
 		copy_range.x -= 1;
 		copy_range.y -= 1;
@@ -9229,11 +9449,7 @@ bool GSRendererHW::OI_BlitFMV(GSTextureCache::Target* _rt, GSTextureCache::Sourc
 				const GSVector4 sRect(m_vt.m_min.t.x / tw, m_vt.m_min.t.y / th, m_vt.m_max.t.x / tw, m_vt.m_max.t.y / th);
 				const GSVector4i r_full_new(0, 0, tw, th);
 				g_gs_device->StretchRect(tex->m_texture, sRect, rt, dRect, ShaderConvert::COPY, m_vt.IsRealLinear());
-				g_perfmon.Put(GSPerfMon::TextureCopies, 1);
-
 				g_gs_device->CopyRect(rt, tex->m_texture, r_full_new, 0, 0);
-				g_perfmon.Put(GSPerfMon::TextureCopies, 1);
-
 				g_gs_device->Recycle(rt);
 			}
 

pcsx2/GS/Renderers/HW/GSTextureCache.cpp

@@ -2404,250 +2404,289 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 	};
 
 	Target* dst = nullptr;
-	auto& list = m_dst[type];
+	Target* dst_match = nullptr;
+	auto* list = &m_dst[type];
 
 	const GSVector4i min_rect = draw_rect.max_u32(GSVector4i(0, 0, draw_rect.x, draw_rect.y));
 	// TODO: Move all frame stuff to its own routine too.
 	if (!is_frame)
 	{
-		for (auto i = list.begin(); i != list.end();)
+		for (int iteration = 0; iteration < 2; iteration++)
 		{
-			Target* t = *i;
-			if (bp == t->m_TEX0.TBP0)
-			{
-				bool can_use = true;
+			if (dst != nullptr)
+				break;
 
-				if (dst && ((GSState::s_n - dst->m_last_draw) < (GSState::s_n - t->m_last_draw) && dst->m_TEX0.TBP0 <= bp))
+			auto& new_dst = iteration == 0 ? dst : dst_match;
+			list = &m_dst[iteration == 0 ? type : (1 - type)];
+			for (auto i = list->begin(); i != list->end();)
+			{
+				Target* t = *i;
+				if (bp == t->m_TEX0.TBP0)
 				{
-					DevCon.Warning("Ignoring target at %x as one at %x is newer", t->m_TEX0.TBP0, dst->m_TEX0.TBP0);
-					i++;
-					continue;
-				}
-				// if It's an old target and it's being completely overwritten, kill it.
-				// Dragon Quest 8 reuses a render-target sized buffer as a single-page buffer, without clearing it. But,
-				// it does dirty it by writing over the 64x64 region. So while we can't use this heuristic for tossing
-				// targets at BW=1 because it breaks other games, we can when the *new* buffer area is completely dirty.
-				if (((!preserve_rgb && !preserve_alpha) || (t->m_was_dst_matched && fbmask == 0xffffff)) && TEX0.TBW != t->m_TEX0.TBW)
-				{
-					// Old targets or shrunk targets where Y draw height goes outside the page.
-					if (TEX0.TBW > 1 && (t->m_age >= 1 || (type == RenderTarget && draw_rect.w > GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y && TEX0.TBW < t->m_TEX0.TBW)))
+					bool can_use = true;
+
+					if (new_dst && ((GSState::s_n - new_dst->m_last_draw) < (GSState::s_n - t->m_last_draw) && new_dst->m_TEX0.TBP0 <= bp))
+					{
+						DevCon.Warning("Ignoring target at %x as one at %x is newer", t->m_TEX0.TBP0, new_dst->m_TEX0.TBP0);
+						i++;
+						continue;
+					}
+					
+					// If there's no valid RGB, it can't be interchanged between RT and Depth
+					if (iteration == 1)
+					{
+						const u32 valid_mask = (t->m_valid_rgb ? 0x7 : 0x0) | ((t->m_valid_alpha_low || t->m_valid_alpha_high) ? 0x8 : 0x0);
+						if ((!(valid_mask & GSUtil::GetChannelMask(TEX0.PSM)) || (!is_shuffle && TEX0.TBW < (t->m_TEX0.TBW / 2))) || 
+							(!is_shuffle && GSLocalMemory::m_psm[t->m_TEX0.PSM].bpp != GSLocalMemory::m_psm[TEX0.PSM].bpp))
+						{
+							if (!preserve_rgb && !preserve_alpha && (!src || src->m_from_target != t) && (valid_mask & GSUtil::GetChannelMask(TEX0.PSM)))
+							{
+								InvalidateSourcesFromTarget(t);
+								i = list->erase(i);
+								delete t;
+							}
+							else
+								i++;
+							continue;
+						}
+					}
+
+					// if It's an old target and it's being completely overwritten, kill it.
+					// Dragon Quest 8 reuses a render-target sized buffer as a single-page buffer, without clearing it. But,
+					// it does dirty it by writing over the 64x64 region. So while we can't use this heuristic for tossing
+					// targets at BW=1 because it breaks other games, we can when the *new* buffer area is completely dirty.
+					if (((!preserve_rgb && !preserve_alpha) || (t->m_was_dst_matched && fbmask == 0xffffff)) && TEX0.TBW != t->m_TEX0.TBW)
+					{
+						// Old targets or shrunk targets where Y draw height goes outside the page.
+						if (TEX0.TBW > 1 && (t->m_age >= 1 || (type == RenderTarget && draw_rect.w > GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y && TEX0.TBW < t->m_TEX0.TBW)))
+						{
+							can_use = false;
+						}
+						else if (!t->m_dirty.empty())
+						{
+							const GSVector4i size_rect = GSVector4i::loadh(size);
+							can_use = !t->m_dirty.GetTotalRect(TEX0, size).rintersect(size_rect).eq(size_rect);
+						}
+					}
+					else if (type == RenderTarget && (fbmask == 0xffffff && !t->m_was_dst_matched && TEX0.TBW != t->m_TEX0.TBW))
+					{
+						// When returning to being matched with the Z buffer in width, we need to make sure the RGB is up to date as it could get used later (Hitman Contracts).
+						auto& rev_list = m_dst[1 - type];
+						for (auto j = rev_list.begin(); j != rev_list.end(); ++j)
+						{
+							Target* ds = *j;
+
+							if (t->m_TEX0.TBP0 != ds->m_TEX0.TBP0 || !ds->m_valid_rgb || TEX0.TBW != ds->m_TEX0.TBW)
+								continue;
+
+							t->m_was_dst_matched = true;
+							t->m_valid_rgb = false;
+							break;
+						}
+					}
+					// TODO: What might be a nicer solution than this, is to rearrange the targets to match the new layout, however this comes with some caviets:
+					// 1. They can draw wider than the FBW
+					// 2. The dirty+valid rects will need to also be rearranged
+					// 3. This could mean larger targets hanging around more
+					// 4. Sources which reference a target may become invalid and will need to be removed
+					// 5. Potential performance implications from additional render passes/copying
+					//
+					// But the bonuses are:
+					// 1. Rearranging the page layout will fix quite a few games which do this
+					// 2. Preserved data will be in the correct place (in most cases)
+					// 3. Less deleting sources/targets
+					// 4. We can basically do clears in hardware, if they aren't insane ones
+					bool dirtied_area = t->m_dirty.size() >= 1;
+
+					// Check it covers the whole area of the new draw
+					if (!is_shuffle && dirtied_area)
+					{
+						const u32 draw_start = GSLocalMemory::GetStartBlockAddress(TEX0.TBP0, TEX0.TBW, TEX0.PSM, draw_rect);
+						const u32 draw_end = GSLocalMemory::GetEndBlockAddress(TEX0.TBP0, TEX0.TBW, TEX0.PSM, draw_rect);
+
+						const GSVector4i dirty_rect = t->m_dirty.GetTotalRect(t->m_TEX0, t->m_unscaled_size);
+						const u32 dirty_start = GSLocalMemory::GetStartBlockAddress(t->m_TEX0.TBP0, t->m_TEX0.TBW, t->m_TEX0.PSM, dirty_rect);
+						const u32 dirty_end = GSLocalMemory::GetEndBlockAddress(t->m_TEX0.TBP0, t->m_TEX0.TBW, t->m_TEX0.PSM, dirty_rect);
+
+						if (dirty_end < draw_end || dirty_start > draw_start)
+							dirtied_area = false;
+					}
+
+					if (can_use && ((!is_shuffle && dirtied_area) || (is_shuffle && src && GSLocalMemory::m_psm[src->m_TEX0.PSM].bpp == 8 && GSLocalMemory::m_psm[t->m_TEX0.PSM].bpp == 16)) && ((preserve_alpha && preserve_rgb) || (draw_rect.w > GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y && !possible_clear)) && TEX0.TBW != t->m_TEX0.TBW)
 					{
 						can_use = false;
 					}
-					else if (!t->m_dirty.empty())
+
+					if (can_use)
 					{
-						const GSVector4i size_rect = GSVector4i::loadh(size);
-						can_use = !t->m_dirty.GetTotalRect(TEX0, size).rintersect(size_rect).eq(size_rect);
-					}
-				}
-				else if (type == RenderTarget && (fbmask == 0xffffff && !t->m_was_dst_matched && TEX0.TBW != t->m_TEX0.TBW))
-				{
-					// When returning to being matched with the Z buffer in width, we need to make sure the RGB is up to date as it could get used later (Hitman Contracts).
-					auto& rev_list = m_dst[1 - type];
-					for (auto j = rev_list.begin(); j != rev_list.end(); ++j)
-					{
-						Target* ds = *j;
-
-						if (t->m_TEX0.TBP0 != ds->m_TEX0.TBP0 || !ds->m_valid_rgb || TEX0.TBW != ds->m_TEX0.TBW)
-							continue;
-
-						t->m_was_dst_matched = true;
-						t->m_valid_rgb = false;
-						break;
-					}
-				}
-				// TODO: What might be a nicer solution than this, is to rearrange the targets to match the new layout, however this comes with some caviets:
-				// 1. They can draw wider than the FBW
-				// 2. The dirty+valid rects will need to also be rearranged
-				// 3. This could mean larger targets hanging around more
-				// 4. Sources which reference a target may become invalid and will need to be removed
-				// 5. Potential performance implications from additional render passes/copying
-				//
-				// But the bonuses are:
-				// 1. Rearranging the page layout will fix quite a few games which do this
-				// 2. Preserved data will be in the correct place (in most cases)
-				// 3. Less deleting sources/targets
-				// 4. We can basically do clears in hardware, if they aren't insane ones
-				bool dirtied_area = t->m_dirty.size() >= 1;
-
-				// Check it covers the whole area of the new draw
-				if (!is_shuffle && dirtied_area)
-				{
-					const u32 draw_start = GSLocalMemory::GetStartBlockAddress(TEX0.TBP0, TEX0.TBW, TEX0.PSM, draw_rect);
-					const u32 draw_end = GSLocalMemory::GetEndBlockAddress(TEX0.TBP0, TEX0.TBW, TEX0.PSM, draw_rect);
-
-					const GSVector4i dirty_rect = t->m_dirty.GetTotalRect(t->m_TEX0, t->m_unscaled_size);
-					const u32 dirty_start = GSLocalMemory::GetStartBlockAddress(t->m_TEX0.TBP0, t->m_TEX0.TBW, t->m_TEX0.PSM, dirty_rect);
-					const u32 dirty_end = GSLocalMemory::GetEndBlockAddress(t->m_TEX0.TBP0, t->m_TEX0.TBW, t->m_TEX0.PSM, dirty_rect);
-
-					if (dirty_end < draw_end || dirty_start > draw_start)
-						dirtied_area = false;
-				}
-
-				if (can_use && ((!is_shuffle && dirtied_area) || (is_shuffle && src && GSLocalMemory::m_psm[src->m_TEX0.PSM].bpp == 8 && GSLocalMemory::m_psm[t->m_TEX0.PSM].bpp == 16)) && ((preserve_alpha && preserve_rgb) || (draw_rect.w > GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y && !possible_clear)) && TEX0.TBW != t->m_TEX0.TBW)
-				{
-					can_use = false;
-				}
-
-				if (can_use)
-				{
-					if (used)
-						list.MoveFront(i.Index());
-					dst = t;
-
-					dst->m_32_bits_fmt |= (psm_s.bpp != 16);
-					break;
-				}
-				else if (!(src && src->m_from_target == t))
-				{
-					GL_INS("TC: Deleting RT BP 0x%x BW %d PSM %s due to change in target", t->m_TEX0.TBP0, t->m_TEX0.TBW, GSUtil::GetPSMName(t->m_TEX0.PSM));
-					InvalidateSourcesFromTarget(t);
-					i = list.erase(i);
-					delete t;
-
-					continue;
-				}
-			}
-			// Probably pointing to half way through the target
-			else if (!min_rect.rempty() && GSConfig.UserHacks_TextureInsideRt >= GSTextureInRtMode::InsideTargets)
-			{
-				// Some games misuse the scissor so it ends up valid 1 pixel over, which causes hell for us. So check if it still overlaps without the extra pixel.
-				const GSVector4i adjusted_valid = GSVector4i(t->m_valid.x, t->m_valid.y, std::min(t->m_valid.z, static_cast<int>(t->m_TEX0.TBW) * 64), t->m_valid.w - 1);
-				const u32 adjusted_endblock = GSLocalMemory::GetEndBlockAddress(t->m_TEX0.TBP0, t->m_TEX0.TBW, t->m_TEX0.PSM, adjusted_valid);
-				if (adjusted_endblock <= bp)
-				{
-					i++;
-					continue;
-				}
-
-				const u32 widthpage_offset = (std::abs(static_cast<int>(bp - t->m_TEX0.TBP0)) >> 5) % std::max(t->m_TEX0.TBW, 1U);
-				const bool is_aligned_ok = widthpage_offset == 0 || ((min_rect.width() <= static_cast<int>((t->m_TEX0.TBW - widthpage_offset) * 64) && (t->m_TEX0.TBW == TEX0.TBW || TEX0.TBW == 1)) && bp >= t->m_TEX0.TBP0);
-				const bool no_target_or_newer = (!dst || ((GSState::s_n - dst->m_last_draw) < (GSState::s_n - t->m_last_draw)));
-				const bool width_match = (t->m_TEX0.TBW == TEX0.TBW || (TEX0.TBW == 1 && draw_rect.w <= GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y));
-				const bool ds_offset = !ds || offset != 0;
-				const bool is_double_buffer = TEX0.TBP0 == ((((t->m_end_block + 1) - t->m_TEX0.TBP0) / 2) + t->m_TEX0.TBP0);
-				const bool source_match = src && src->m_TEX0.TBP0 <= bp && src->m_end_block > bp && src->m_TEX0.TBW == TEX0.TBW && src->m_from_target && src->m_from_target == t && t->Inside(bp, TEX0.TBW, TEX0.PSM, min_rect);
-				const bool was_used_last_draw = t->m_last_draw == (GSState::s_n - 1);
-				// if it's a shuffle, some games tend to offset back by a page, such as Tomb Raider, for no disernable reason, but it then causes problems.
-				// This can also happen horizontally (Catwoman moves everything one page left with shuffles), but this is too messy to deal with right now.
-				const bool overlaps = t->Overlaps(bp, TEX0.TBW, TEX0.PSM, min_rect) || (is_shuffle && src && GSLocalMemory::m_psm[src->m_TEX0.PSM].bpp == 8 && t->Overlaps(bp, TEX0.TBW, TEX0.PSM, min_rect + GSVector4i(0, 0, 0, 32)));
-				if (source_match || (no_target_or_newer && is_aligned_ok && width_match && overlaps && (is_shuffle || ds_offset || is_double_buffer || was_used_last_draw)))
-				{
-					const GSLocalMemory::psm_t& s_psm = GSLocalMemory::m_psm[TEX0.PSM];
-
-					// If it overlaps but the target is huge and the Z isn't offset, we need to split the buffer, so let's shrink this one down.
-					// 896 is just 448 * 2,just gives the buffer chance to be larger than normal, in case they do something like 640x640, or something ridiculous.
-					if (!is_shuffle && (ds && offset == 0 && (t->m_valid.w >= 896) && ((((t->m_end_block + 1) - t->m_TEX0.TBP0) >> 1) + t->m_TEX0.TBP0) <= bp))
-					{
-						const u32 local_offset = (((bp - t->m_TEX0.TBP0) >> 5) / std::max(t->m_TEX0.TBW, 1U)) * s_psm.pgs.y;
-						if ((dst = CreateTarget(TEX0, GSVector2i(t->m_valid.z, t->m_valid.w - local_offset), GSVector2i(t->m_valid.z, t->m_valid.w - local_offset), scale, type, true, fbmask, false, false, preserve_rgb || preserve_alpha, GSVector4i::zero(), src)))
-							dst->m_32_bits_fmt |= (psm_s.bpp != 16);
-
-						break;
-					}
-
-					// I know what you're thinking, and I hate the guy who wrote it too (me). Project Snowblind, Tomb Raider etc decide to offset where they're drawing using a channel shuffle, and this gets messy, so best just to kill the old target.
-					if (is_shuffle && src && src->m_TEX0.PSM == PSMT8 && GSRendererHW::GetInstance()->m_context->FRAME.FBW == 1 && t->m_last_draw != (GSState::s_n - 1) && src->m_from_target && (src->m_from_target->m_TEX0.TBP0 == src->m_TEX0.TBP0 || (((src->m_TEX0.TBP0 - src->m_from_target->m_TEX0.TBP0) >> 5) % std::max(src->m_from_target->m_TEX0.TBW, 1U) == 0)) && widthpage_offset && src->m_from_target != t)
-					{
-						GL_INS("TC: Deleting RT BP 0x%x BW %d PSM %s offset overwrite shuffle", t->m_TEX0.TBP0, t->m_TEX0.TBW, GSUtil::GetPSMName(t->m_TEX0.PSM));
-						InvalidateSourcesFromTarget(t);
-						i = list.erase(i);
-						delete t;
-
-						continue;
-					}
-
-					if (!is_shuffle && (!GSUtil::HasSameSwizzleBits(t->m_TEX0.PSM, TEX0.PSM) ||
-										   ((widthpage_offset % std::max(t->m_TEX0.TBW, 1U)) != 0 && ((widthpage_offset + (min_rect.width() + (s_psm.pgs.x - 1)) / s_psm.pgs.x)) > t->m_TEX0.TBW)))
-					{
-						const int page_offset = TEX0.TBP0 - t->m_TEX0.TBP0;
-						const int number_pages = page_offset / 32;
-						const u32 tbw = std::max(t->m_TEX0.TBW, 1u);
-						const int row_offset = number_pages / tbw;
-						const int page_height = GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y;
-						const int vertical_position = row_offset * page_height;
-
-						if (src && src->m_from_target == t && src->m_target_direct && vertical_position >= t->m_valid.w / 2)
-						{
-							// Valids and drawn since last read doesn't match, keep the target but resize it.
-							src->m_valid_rect.w = std::min(vertical_position, src->m_valid_rect.w);
-							t->m_valid.w = std::min(vertical_position, t->m_valid.w);
-							t->ResizeValidity(t->m_valid);
-							t->ResizeDrawn(t->m_valid);
-							++i;
-						}
-						else
-						{
-							GL_INS("TC: Deleting RT BP 0x%x BW %d PSM %s due to change in target", t->m_TEX0.TBP0, t->m_TEX0.TBW, GSUtil::GetPSMName(t->m_TEX0.PSM));
-							InvalidateSourcesFromTarget(t);
-							i = list.erase(i);
-							delete t;
-						}
-
-						continue;
-					}
-
-					GSVector4i lookup_rect = min_rect;
-
-					if (is_shuffle)
-						lookup_rect = lookup_rect & GSVector4i(~8);
-
-					const GSVector4i translated_rect = GSVector4i(0, 0, 0, 0).max_i32(TranslateAlignedRectByPage(t, TEX0.TBP0, TEX0.PSM, TEX0.TBW, lookup_rect));
-					const GSVector4i dirty_rect = t->m_dirty.empty() ? GSVector4i::zero() : t->m_dirty.GetTotalRect(t->m_TEX0, t->m_unscaled_size);
-					const bool all_dirty = dirty_rect.eq(t->m_valid);
-
-
-					if (!is_shuffle && !dirty_rect.rempty() && (!preserve_alpha && !preserve_rgb) && (GSState::s_n - 3) > t->m_last_draw)
-					{
-						GL_INS("TC: Deleting RT BP 0x%x BW %d PSM %s due to dirty areas not preserved (Likely change in target)", t->m_TEX0.TBP0, t->m_TEX0.TBW, GSUtil::GetPSMName(t->m_TEX0.PSM));
-						InvalidateSourcesFromTarget(t);
-						i = list.erase(i);
-						delete t;
-
-						continue;
-					}
-
-					if (!all_dirty && ((translated_rect.w <= t->m_valid.w) || widthpage_offset == 0 || (GSState::s_n - 3) <= t->m_last_draw))
-					{
-						if (TEX0.TBW == t->m_TEX0.TBW && !is_shuffle && widthpage_offset == 0 && ((min_rect.w + 63) / 64) > 1)
-						{
-							// Beyond Good and Evil does this awful thing where it puts one framebuffer at 0xf00, with the first row of pages blanked out, and the whole thing goes down to 0x2080
-							// which is a problem, because it then puts the Z buffer at 0x1fc0, then offsets THAT by 1 row of pages, so it starts at, you guessed it, 2080.
-							// So let's check the *real* start.
-							u32 real_start_address = GSLocalMemory::GetStartBlockAddress(t->m_TEX0.TBP0, t->m_TEX0.TBW, t->m_TEX0.PSM, t->m_drawn_since_read);
-							u32 new_end_address = GSLocalMemory::GetEndBlockAddress(TEX0.TBP0, TEX0.TBW, TEX0.PSM, min_rect);
-
-							// Not really overlapping.
-							if (real_start_address > new_end_address)
-							{
-								i++;
-								continue;
-							}
-						}
-
-						//DevCon.Warning("Here draw %d wanted %x PSM %x got %x PSM %x offset of %d pages width %d pages draw width %d", GSState::s_n, bp, TEX0.PSM, t->m_TEX0.TBP0, t->m_TEX0.PSM, (bp - t->m_TEX0.TBP0) >> 5, t->m_TEX0.TBW, draw_rect.width());
-						dst = t;
-
-						dst->m_32_bits_fmt |= (psm_s.bpp != 16);
-						//Continue just in case there's a newer target
 						if (used)
-							list.MoveFront(i.Index());
-						if (t->m_TEX0.TBP0 <= bp || GSLocalMemory::GetStartBlockAddress(TEX0.TBP0, TEX0.TBW, TEX0.PSM, min_rect) >= bp)
-							break;
-						else
-							continue;
+							list->MoveFront(i.Index());
+
+						new_dst = t;
+						new_dst->m_32_bits_fmt |= (psm_s.bpp != 16);
+
+						break;
+					}
+					else if (!(src && src->m_from_target == t))
+					{
+						GL_INS("TC: Deleting RT BP 0x%x BW %d PSM %s due to change in target", t->m_TEX0.TBP0, t->m_TEX0.TBW, GSUtil::GetPSMName(t->m_TEX0.PSM));
+						InvalidateSourcesFromTarget(t);
+						i = list->erase(i);
+						delete t;
+
+						continue;
 					}
 				}
-			}
+				// Probably pointing to half way through the target
+				else if (!min_rect.rempty() && GSConfig.UserHacks_TextureInsideRt >= GSTextureInRtMode::InsideTargets)
+				{
+					// Some games misuse the scissor so it ends up valid 1 pixel over, which causes hell for us. So check if it still overlaps without the extra pixel.
+					const GSVector4i adjusted_valid = GSVector4i(t->m_valid.x, t->m_valid.y, std::min(t->m_valid.z, static_cast<int>(t->m_TEX0.TBW) * 64), t->m_valid.w - 1);
+					const u32 adjusted_endblock = GSLocalMemory::GetEndBlockAddress(t->m_TEX0.TBP0, t->m_TEX0.TBW, t->m_TEX0.PSM, adjusted_valid);
+					if (adjusted_endblock <= bp)
+					{
+						i++;
+						continue;
+					}
 
-			i++;
+					const u32 widthpage_offset = (std::abs(static_cast<int>(bp - t->m_TEX0.TBP0)) >> 5) % std::max(t->m_TEX0.TBW, 1U);
+					const bool is_aligned_ok = widthpage_offset == 0 || ((min_rect.width() <= static_cast<int>((t->m_TEX0.TBW - widthpage_offset) * 64) && (t->m_TEX0.TBW == TEX0.TBW || TEX0.TBW == 1)) && bp >= t->m_TEX0.TBP0);
+					const bool no_target_or_newer = (!new_dst || ((GSState::s_n - new_dst->m_last_draw) < (GSState::s_n - t->m_last_draw)));
+					const bool width_match = (t->m_TEX0.TBW == TEX0.TBW || (TEX0.TBW == 1 && draw_rect.w <= GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y));
+					const bool ds_offset = !ds || offset != 0;
+					const bool is_double_buffer = TEX0.TBP0 == ((((t->m_end_block + 1) - t->m_TEX0.TBP0) / 2) + t->m_TEX0.TBP0);
+					const bool source_match = src && src->m_TEX0.TBP0 <= bp && src->m_end_block > bp && src->m_TEX0.TBW == TEX0.TBW && src->m_from_target && src->m_from_target == t && t->Inside(bp, TEX0.TBW, TEX0.PSM, min_rect);
+					const bool was_used_last_draw = t->m_last_draw == (GSState::s_n - 1);
+					// if it's a shuffle, some games tend to offset back by a page, such as Tomb Raider, for no disernable reason, but it then causes problems.
+					// This can also happen horizontally (Catwoman moves everything one page left with shuffles), but this is too messy to deal with right now.
+					const bool overlaps = t->Overlaps(bp, TEX0.TBW, TEX0.PSM, min_rect) || (is_shuffle && src && GSLocalMemory::m_psm[src->m_TEX0.PSM].bpp == 8 && t->Overlaps(bp, TEX0.TBW, TEX0.PSM, min_rect + GSVector4i(0, 0, 0, 32)));
+					if (source_match || (no_target_or_newer && is_aligned_ok && width_match && overlaps && (is_shuffle || ds_offset || is_double_buffer || was_used_last_draw)))
+					{
+						const GSLocalMemory::psm_t& s_psm = GSLocalMemory::m_psm[TEX0.PSM];
+
+						// If it overlaps but the target is huge and the Z isn't offset, we need to split the buffer, so let's shrink this one down.
+						// 896 is just 448 * 2,just gives the buffer chance to be larger than normal, in case they do something like 640x640, or something ridiculous.
+						if (!is_shuffle && (ds && offset == 0 && (t->m_valid.w >= 896) && ((((t->m_end_block + 1) - t->m_TEX0.TBP0) >> 1) + t->m_TEX0.TBP0) <= bp))
+						{
+							const u32 local_offset = (((bp - t->m_TEX0.TBP0) >> 5) / std::max(t->m_TEX0.TBW, 1U)) * s_psm.pgs.y;
+							if ((new_dst = CreateTarget(TEX0, GSVector2i(t->m_valid.z, t->m_valid.w - local_offset), GSVector2i(t->m_valid.z, t->m_valid.w - local_offset), scale, type, true, fbmask, false, false, preserve_rgb || preserve_alpha, GSVector4i::zero(), src)))
+								new_dst->m_32_bits_fmt |= (psm_s.bpp != 16);
+
+							break;
+						}
+
+						// I know what you're thinking, and I hate the guy who wrote it too (me). Project Snowblind, Tomb Raider etc decide to offset where they're drawing using a channel shuffle, and this gets messy, so best just to kill the old target.
+						if (is_shuffle && src && src->m_TEX0.PSM == PSMT8 && GSRendererHW::GetInstance()->m_context->FRAME.FBW == 1 && t->m_last_draw != (GSState::s_n - 1) && src->m_from_target && (src->m_from_target->m_TEX0.TBP0 == src->m_TEX0.TBP0 || (((src->m_TEX0.TBP0 - src->m_from_target->m_TEX0.TBP0) >> 5) % std::max(src->m_from_target->m_TEX0.TBW, 1U) == 0)) && widthpage_offset && src->m_from_target != t)
+						{
+							if (iteration == 0)
+							{
+								GL_INS("TC: Deleting RT BP 0x%x BW %d PSM %s offset overwrite shuffle", t->m_TEX0.TBP0, t->m_TEX0.TBW, GSUtil::GetPSMName(t->m_TEX0.PSM));
+								InvalidateSourcesFromTarget(t);
+								i = list->erase(i);
+								delete t;
+							}
+							else
+								i++;
+							continue;
+						}
+
+						if (!is_shuffle && (!GSUtil::HasSameSwizzleBits(t->m_TEX0.PSM, TEX0.PSM) ||
+											   ((widthpage_offset % std::max(t->m_TEX0.TBW, 1U)) != 0 && ((widthpage_offset + (min_rect.width() + (s_psm.pgs.x - 1)) / s_psm.pgs.x)) > t->m_TEX0.TBW)))
+						{
+							const int page_offset = TEX0.TBP0 - t->m_TEX0.TBP0;
+							const int number_pages = page_offset / 32;
+							const u32 tbw = std::max(t->m_TEX0.TBW, 1u);
+							const int row_offset = number_pages / tbw;
+							const int page_height = GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y;
+							const int vertical_position = row_offset * page_height;
+
+							if (src && src->m_from_target == t && src->m_target_direct && vertical_position >= t->m_valid.w / 2)
+							{
+								// Valids and drawn since last read doesn't match, keep the target but resize it.
+								src->m_valid_rect.w = std::min(vertical_position, src->m_valid_rect.w);
+								t->m_valid.w = std::min(vertical_position, t->m_valid.w);
+								t->ResizeValidity(t->m_valid);
+								t->ResizeDrawn(t->m_valid);
+								++i;
+							}
+							else
+							{
+								if (iteration == 0)
+								{
+									GL_INS("TC: Deleting RT BP 0x%x BW %d PSM %s due to change in target", t->m_TEX0.TBP0, t->m_TEX0.TBW, GSUtil::GetPSMName(t->m_TEX0.PSM));
+									InvalidateSourcesFromTarget(t);
+									i = list->erase(i);
+									delete t;
+								}
+								else
+									i++;
+							}
+
+							continue;
+						}
+
+						GSVector4i lookup_rect = min_rect;
+
+						if (is_shuffle)
+							lookup_rect = lookup_rect & GSVector4i(~8);
+
+						const GSVector4i translated_rect = GSVector4i(0, 0, 0, 0).max_i32(TranslateAlignedRectByPage(t, TEX0.TBP0, TEX0.PSM, TEX0.TBW, lookup_rect));
+						const GSVector4i dirty_rect = t->m_dirty.empty() ? GSVector4i::zero() : t->m_dirty.GetTotalRect(t->m_TEX0, t->m_unscaled_size);
+						const bool all_dirty = dirty_rect.eq(t->m_valid);
+
+
+						if (!is_shuffle && !dirty_rect.rempty() && (!preserve_alpha && !preserve_rgb) && (GSState::s_n - 3) > t->m_last_draw)
+						{
+							GL_INS("TC: Deleting RT BP 0x%x BW %d PSM %s due to dirty areas not preserved (Likely change in target)", t->m_TEX0.TBP0, t->m_TEX0.TBW, GSUtil::GetPSMName(t->m_TEX0.PSM));
+							InvalidateSourcesFromTarget(t);
+							i = list->erase(i);
+							delete t;
+
+							continue;
+						}
+
+						if (!all_dirty && ((translated_rect.w <= t->m_valid.w) || widthpage_offset == 0 || (GSState::s_n - 3) <= t->m_last_draw))
+						{
+							if (TEX0.TBW == t->m_TEX0.TBW && !is_shuffle && widthpage_offset == 0 && ((min_rect.w + 63) / 64) > 1)
+							{
+								// Beyond Good and Evil does this awful thing where it puts one framebuffer at 0xf00, with the first row of pages blanked out, and the whole thing goes down to 0x2080
+								// which is a problem, because it then puts the Z buffer at 0x1fc0, then offsets THAT by 1 row of pages, so it starts at, you guessed it, 2080.
+								// So let's check the *real* start.
+								u32 real_start_address = GSLocalMemory::GetStartBlockAddress(t->m_TEX0.TBP0, t->m_TEX0.TBW, t->m_TEX0.PSM, t->m_drawn_since_read);
+								u32 new_end_address = GSLocalMemory::GetEndBlockAddress(TEX0.TBP0, TEX0.TBW, TEX0.PSM, min_rect);
+
+								// Not really overlapping.
+								if (real_start_address > new_end_address)
+								{
+									i++;
+									continue;
+								}
+							}
+
+							//DevCon.Warning("Here draw %d wanted %x PSM %x got %x PSM %x offset of %d pages width %d pages draw width %d", GSState::s_n, bp, TEX0.PSM, t->m_TEX0.TBP0, t->m_TEX0.PSM, (bp - t->m_TEX0.TBP0) >> 5, t->m_TEX0.TBW, draw_rect.width());
+							new_dst = t;
+							new_dst->m_32_bits_fmt |= (psm_s.bpp != 16);
+
+							//Continue just in case there's a newer target
+							if (used)
+								list->MoveFront(i.Index());
+							if (t->m_TEX0.TBP0 <= bp || GSLocalMemory::GetStartBlockAddress(TEX0.TBP0, TEX0.TBW, TEX0.PSM, min_rect) >= bp)
+								break;
+							else
+								continue;
+						}
+					}
+				}
+
+				i++;
+			}
 		}
 	}
 	else
 	{
 		pxAssert(type == RenderTarget);
 		// Let's try to find a perfect frame that contains valid data
-		for (auto i = list.begin(); i != list.end(); ++i)
+		for (auto i = list->begin(); i != list->end(); ++i)
 		{
 			Target* t = *i;
 
@@ -2673,7 +2712,7 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 						{
 							DevCon.Warning("Wanted %x psm %x bw %x, got %x psm %x bw %x, deleting", TEX0.TBP0, TEX0.PSM, TEX0.TBW, t->m_TEX0.TBP0, t->m_TEX0.PSM, t->m_TEX0.TBW);
 							InvalidateSourcesFromTarget(t);
-							i = list.erase(i);
+							i = list->erase(i);
 							delete t;
 							continue;
 						}
@@ -2691,7 +2730,7 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 		// 2nd try ! Try to find a frame at the requested bp -> bp + size is inside of (or equal to)
 		if (!dst)
 		{
-			for (auto i = list.begin(); i != list.end(); ++i)
+			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));
@@ -2711,7 +2750,7 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 						{
 							DevCon.Warning("2 Wanted %x psm %x bw %x, got %x psm %x bw %x, deleting", TEX0.TBP0, TEX0.PSM, TEX0.TBW, t->m_TEX0.TBP0, t->m_TEX0.PSM, t->m_TEX0.TBW);
 							InvalidateSourcesFromTarget(t);
-							i = list.erase(i);
+							i = list->erase(i);
 							delete t;
 							continue;
 						}
@@ -2731,7 +2770,7 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 		// 3rd try ! Try to find a frame that doesn't contain valid data (honestly I'm not sure we need to do it)
 		if (!dst)
 		{
-			for (auto i = list.begin(); i != list.end(); ++i)
+			for (auto i = list->begin(); i != list->end(); ++i)
 			{
 				Target* t = *i;
 				if (bp == t->m_TEX0.TBP0 && TEX0.TBW == t->m_TEX0.TBW)
@@ -2743,7 +2782,7 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 						{
 							DevCon.Warning("3 Wanted %x psm %x bw %x, got %x psm %x bw %x, deleting", TEX0.TBP0, TEX0.PSM, TEX0.TBW, t->m_TEX0.TBP0, t->m_TEX0.PSM, t->m_TEX0.TBW);
 							InvalidateSourcesFromTarget(t);
-							i = list.erase(i);
+							i = list->erase(i);
 							delete t;
 							continue;
 						}
@@ -2795,7 +2834,6 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 			else
 				g_gs_device->StretchRect(dst->m_texture, sRect, tex, dRect, (type == RenderTarget) ? ShaderConvert::COPY : ShaderConvert::DEPTH_COPY, type == RenderTarget && !preserve_scale);
 
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			m_target_memory_usage = (m_target_memory_usage - dst->m_texture->GetMemUsage()) + tex->GetMemUsage();
 
 			// If we're changing resolution scale, just toss the texture, it's not going to get reused.
@@ -2826,7 +2864,6 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 			if (!tex)
 				return nullptr;
 			g_gs_device->StretchRect(dst->m_texture, sRect, tex, dRect, ShaderConvert::FLOAT32_TO_FLOAT24, false);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			g_gs_device->Recycle(dst->m_texture);
 
 			dst->m_texture = tex;
@@ -2899,7 +2936,6 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 
 						g_gs_device->StretchRect(dst->m_texture, source_rect, tex, dRect, (type == RenderTarget) ? ShaderConvert::COPY : ShaderConvert::DEPTH_COPY, false);
 
-						g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 						m_target_memory_usage = m_target_memory_usage + tex->GetMemUsage();
 
 						// Don't kill the target here as it's being used for the source.
@@ -2923,7 +2959,6 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 						else
 							g_gs_device->StretchRect(dst->m_texture, source_rect, tex, dRect, (type == RenderTarget) ? ShaderConvert::COPY : ShaderConvert::DEPTH_COPY, false);
 
-						g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 						m_target_memory_usage = (m_target_memory_usage - dst->m_texture->GetMemUsage()) + tex->GetMemUsage();
 
 						g_gs_device->Recycle(dst->m_texture);
@@ -3006,7 +3041,6 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 
 						g_gs_device->StretchRect(dst_match->m_texture, GSVector4(0, 0, 1, 1),
 							dst->m_texture, GSVector4(dst->GetUnscaledRect()) * GSVector4(dst->GetScale()), shader, false);
-						g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 
 						dst_match->m_valid_rgb = !used;
 						dst_match->m_was_dst_matched = true;
@@ -3088,67 +3122,28 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 	}
 	else if (!is_frame && !GSConfig.UserHacks_DisableDepthSupport)
 	{
-		const int rev_type = (type == DepthStencil) ? RenderTarget : DepthStencil;
-
-		// Depth stencil/RT can be an older RT/DS but only check recent RT/DS to avoid to pick
-		// some bad data.
-		auto& rev_list = m_dst[rev_type];
-		Target* dst_match = nullptr;
-		for (auto i = rev_list.begin(); i != rev_list.end(); ++i)
+		if (!dst_match)
 		{
-			Target* t = *i;
-			// Don't pull in targets without valid lower 24 bits unless the Z is 32bits and the alpha is valid, it makes no sense to convert them otherwise.
-			// FIXME: Technically the difference in size is fine, but if the target gets reinterpreted, the hw renderer doesn't rearrange the target.
-			// This does cause some extra uploads in some games (like Burnout), but without this, bad data gets displayed in games like Transformers.
-			if (bp != t->m_TEX0.TBP0 || (!t->m_valid_rgb && (!(GSUtil::GetChannelMask(TEX0.PSM) & 0x8) || !(t->m_valid_alpha_low || t->m_valid_alpha_high))) ||
-				(!is_shuffle && t->m_TEX0.TBW != TEX0.TBW && (possible_clear || ((~GSLocalMemory::m_psm[t->m_TEX0.PSM].fmsk | fbmask) == 0xffffffff))))
+			const int rev_type = (type == DepthStencil) ? RenderTarget : DepthStencil;
+			// Depth stencil/RT can be an older RT/DS but only check recent RT/DS to avoid to pick
+			// some bad data.
+			auto& rev_list = m_dst[rev_type];
+			for (auto i = rev_list.begin(); i != rev_list.end(); ++i)
 			{
-				continue;
-			}
-			// If the format is completely different, but it's the same location, it's likely just overwriting it, so get rid.
-			// Make sure it's not currently in use, that could be bad.
-			if (!is_shuffle && (!ds || (ds != t)) &&
-				t->m_TEX0.TBW != TEX0.TBW && TEX0.TBW != 1 && !preserve_rgb && min_rect.w > GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y)
-			{
-				if (src && src->m_target && src->m_from_target == t && src->m_target_direct)
+				Target* t = *i;
+				// Don't pull in targets without valid lower 24 bits unless the Z is 32bits and the alpha is valid, it makes no sense to convert them otherwise.
+				// FIXME: Technically the difference in size is fine, but if the target gets reinterpreted, the hw renderer doesn't rearrange the target.
+				// This does cause some extra uploads in some games (like Burnout), but without this, bad data gets displayed in games like Transformers.
+				if (bp != t->m_TEX0.TBP0 || (!t->m_valid_rgb && (!(GSUtil::GetChannelMask(TEX0.PSM) & 0x8) || !(t->m_valid_alpha_low || t->m_valid_alpha_high))) ||
+					(!is_shuffle && t->m_TEX0.TBW != TEX0.TBW && (possible_clear || ((~GSLocalMemory::m_psm[t->m_TEX0.PSM].fmsk | fbmask) == 0xffffffff))))
 				{
-					src->m_target_direct = false;
-					src->m_shared_texture = false;
-					t->m_texture = nullptr;
+					continue;
 				}
-
-				GL_CACHE("TC: Deleting Z draw %d", GSState::s_n);
-				InvalidateSourcesFromTarget(t);
-				i = rev_list.erase(i);
-				delete t;
-				continue;
-			}
-			const GSLocalMemory::psm_t& t_psm_s = GSLocalMemory::m_psm[t->m_TEX0.PSM];
-			if (t_psm_s.bpp != psm_s.bpp)
-			{
-				bool remove_target = possible_clear || (used && !is_shuffle);
-
-				// If we have a BW change, and it's not a multiple of 2 (for a shuffle), the game's going to get a jigsaw
-				// puzzle of pages and can't be expecting to have legitimate data. Tokimeki Memorial 3 reuses a BW 17
-				// buffer as BW 10, and if we don't discard the BW 17 buffer, the BW 10 buffer ends up twice the size.
-				const u32 shuffle_bw = (psm_s.bpp > t_psm_s.bpp) ? (TEX0.TBW / 2u) : (TEX0.TBW * 2u);
-				if (t->m_TEX0.TBW != TEX0.TBW && (t->m_TEX0.TBW != shuffle_bw && !is_shuffle))
+				// If the format is completely different, but it's the same location, it's likely just overwriting it, so get rid.
+				// Make sure it's not currently in use, that could be bad.
+				if (!is_shuffle && (!ds || (ds != t)) &&
+					t->m_TEX0.TBW != TEX0.TBW && TEX0.TBW != 1 && !preserve_rgb && min_rect.w > GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs.y)
 				{
-					// But we'll make sure the whole existing target's actually being drawn over to be safe.
-					const u32 end_block = GSLocalMemory::GetUnwrappedEndBlockAddress(TEX0.TBP0, TEX0.TBW, TEX0.PSM, draw_rect);
-					if (end_block >= t->UnwrappedEndBlock())
-					{
-						GL_CACHE("TC: Not converting %s at %x TBW %u with end block of %x when we're drawing through %x",
-							to_string(rev_type), t->m_TEX0.TBP0, t->m_TEX0.TBW, t->UnwrappedEndBlock(), end_block);
-						remove_target = true;
-					}
-				}
-
-				// Probably an old target, get rid of it.
-				if (remove_target)
-				{
-					// DT Racer hits this path and causes a crash when RT in RT is disabled,
-					// so let's make sure source and target texture isn't linked/shared before deleting the target.
 					if (src && src->m_target && src->m_from_target == t && src->m_target_direct)
 					{
 						src->m_target_direct = false;
@@ -3156,26 +3151,68 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 						t->m_texture = nullptr;
 					}
 
+					GL_CACHE("TC: Deleting Z draw %d", GSState::s_n);
 					InvalidateSourcesFromTarget(t);
 					i = rev_list.erase(i);
 					delete t;
 					continue;
 				}
-			}
+				const GSLocalMemory::psm_t& t_psm_s = GSLocalMemory::m_psm[t->m_TEX0.PSM];
+				if (t_psm_s.bpp != psm_s.bpp)
+				{
+					bool remove_target = possible_clear || (used && !is_shuffle);
 
-			if (t->m_age == 0)
-			{
-				dst_match = t;
-				break;
-			}
-			else if (t->m_age == 1 && (preserve_rgb || (preserve_alpha && (t->m_valid_alpha_low || t->m_valid_alpha_high))))
-			{
-				dst_match = t;
+					// If we have a BW change, and it's not a multiple of 2 (for a shuffle), the game's going to get a jigsaw
+					// puzzle of pages and can't be expecting to have legitimate data. Tokimeki Memorial 3 reuses a BW 17
+					// buffer as BW 10, and if we don't discard the BW 17 buffer, the BW 10 buffer ends up twice the size.
+					const u32 shuffle_bw = (psm_s.bpp > t_psm_s.bpp) ? (TEX0.TBW / 2u) : (TEX0.TBW * 2u);
+					if (t->m_TEX0.TBW != TEX0.TBW && (t->m_TEX0.TBW != shuffle_bw && !is_shuffle))
+					{
+						// But we'll make sure the whole existing target's actually being drawn over to be safe.
+						const u32 end_block = GSLocalMemory::GetUnwrappedEndBlockAddress(TEX0.TBP0, TEX0.TBW, TEX0.PSM, draw_rect);
+						if (end_block >= t->UnwrappedEndBlock())
+						{
+							GL_CACHE("TC: Not converting %s at %x TBW %u with end block of %x when we're drawing through %x",
+								to_string(rev_type), t->m_TEX0.TBP0, t->m_TEX0.TBW, t->UnwrappedEndBlock(), end_block);
+							remove_target = true;
+						}
+					}
+
+					// Probably an old target, get rid of it.
+					if (remove_target)
+					{
+						// DT Racer hits this path and causes a crash when RT in RT is disabled,
+						// so let's make sure source and target texture isn't linked/shared before deleting the target.
+						if (src && src->m_target && src->m_from_target == t && src->m_target_direct)
+						{
+							src->m_target_direct = false;
+							src->m_shared_texture = false;
+							t->m_texture = nullptr;
+						}
+
+						InvalidateSourcesFromTarget(t);
+						i = rev_list.erase(i);
+						delete t;
+						continue;
+					}
+				}
+
+				if (t->m_age == 0)
+				{
+					dst_match = t;
+					break;
+				}
+				else if (t->m_age == 1 && (preserve_rgb || (preserve_alpha && (t->m_valid_alpha_low || t->m_valid_alpha_high))))
+				{
+					dst_match = t;
+				}
 			}
 		}
-		// We only want to use a matched target if it's actually being used.
+
 		if (dst_match)
 		{
+			// dst_match, we only want to use a matched target if it's actually being used.
+
 			calcRescale(dst_match);
 
 			// If we don't need A, and the existing target doesn't have valid alpha, don't bother converting it.
@@ -3192,7 +3229,7 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 			// Clear instead of invalidating if there is anything which isn't touched.
 			clear |= (!preserve_target && fbmask != 0);
 			GIFRegTEX0 new_TEX0;
-			new_TEX0.TBP0 = TEX0.TBP0;
+			new_TEX0.TBP0 = GSConfig.UserHacks_TextureInsideRt >= GSTextureInRtMode::InsideTargets ? dst_match->m_TEX0.TBP0 : TEX0.TBP0;
 			new_TEX0.TBW = (!half_width) ? dst_match->m_TEX0.TBW : TEX0.TBW;
 			new_TEX0.PSM = is_shuffle ? dst_match->m_TEX0.PSM : TEX0.PSM;
 
@@ -3207,8 +3244,7 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 			dst->m_valid_alpha_low = dst_match->m_valid_alpha_low; //&& psm_s.trbpp != 24;
 			dst->m_valid_alpha_high = dst_match->m_valid_alpha_high; //&& psm_s.trbpp != 24;
 			dst->m_valid_rgb = dst_match->m_valid_rgb && (dst->m_TEX0.TBW == TEX0.TBW || min_rect.w <= GSLocalMemory::m_psm[dst_match->m_TEX0.PSM].pgs.y);
-			dst->m_was_dst_matched = true;
-			dst_match->m_was_dst_matched = true;
+			dst->m_was_dst_matched = !dst_match->m_was_dst_matched;
 			dst_match->m_valid_rgb = preserve_rgb;
 
 			if (GSLocalMemory::m_psm[dst->m_TEX0.PSM].bpp == 16 && GSLocalMemory::m_psm[dst_match->m_TEX0.PSM].bpp > 16)
@@ -3285,7 +3321,6 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 					{
 						dst_match->UnscaleRTAlpha();
 						g_gs_device->StretchRect(dst_match->m_texture, sRect, dst->m_texture, dRect, shader, false);
-						g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 					}
 				}
 
@@ -3981,7 +4016,6 @@ void GSTextureCache::Target::ScaleRTAlpha()
 				const GSVector4 dRect(m_texture->GetRect().rintersect(valid_rect));
 				const GSVector4 sRect = dRect / GSVector4(rtsize.x, rtsize.y).xyxy();
 				g_gs_device->StretchRect(m_texture, sRect, temp_rt, dRect, ShaderConvert::RTA_CORRECTION, false);
-				g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 				g_gs_device->Recycle(m_texture);
 				m_texture = temp_rt;
 			}
@@ -4007,7 +4041,6 @@ void GSTextureCache::Target::UnscaleRTAlpha()
 				const GSVector4 dRect(m_texture->GetRect().rintersect(valid_rect));
 				const GSVector4 sRect = dRect / GSVector4(rtsize.x, rtsize.y).xyxy();
 				g_gs_device->StretchRect(m_texture, sRect, temp_rt, dRect, ShaderConvert::RTA_DECORRECTION, false);
-				g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 				g_gs_device->Recycle(m_texture);
 				m_texture = temp_rt;
 			}
@@ -4067,7 +4100,6 @@ void GSTextureCache::ScaleTargetForDisplay(Target* t, const GIFRegTEX0& dispfb,
 
 	// Fill the new texture with the old data, and discard the old texture.
 	g_gs_device->StretchRect(old_texture, new_texture, GSVector4(old_texture->GetSize()).zwxy(), ShaderConvert::COPY, false);
-	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 	m_target_memory_usage = (m_target_memory_usage - old_texture->GetMemUsage()) + new_texture->GetMemUsage();
 	g_gs_device->Recycle(old_texture);
 	t->m_texture = new_texture;
@@ -4201,7 +4233,6 @@ bool GSTextureCache::CopyRGBFromDepthToColor(Target* dst, Target* depth_src)
 		const GSVector4 convert_rect = GSVector4(depth_src->GetUnscaledRect().rintersect(GSVector4i::loadh(new_size)));
 		g_gs_device->StretchRect(depth_src->m_texture, convert_rect / GSVector4(depth_src->GetUnscaledSize()).xyxy(),
 			tex, convert_rect * GSVector4(dst->GetScale()), shader, false);
-		g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 	}
 
 	// Copy in alpha if we're a new texture.
@@ -4212,7 +4243,6 @@ bool GSTextureCache::CopyRGBFromDepthToColor(Target* dst, Target* depth_src)
 			const GSVector4 copy_rect = GSVector4(tex->GetRect().rintersect(dst->m_texture->GetRect()));
 			g_gs_device->StretchRect(dst->m_texture, copy_rect / GSVector4(GSVector4i(dst->m_texture->GetSize()).xyxy()), tex,
 				copy_rect, false, false, false, true);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 		}
 
 		g_gs_device->Recycle(dst->m_texture);
@@ -4377,6 +4407,7 @@ 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 &= (fb_mask & 0x00FFFFFF) != 0;
+			t->m_was_dst_matched = false;
 
 			// Don't keep partial depth buffers around.
 			if ((!t->m_valid_alpha_low && !t->m_valid_alpha_high && !t->m_valid_rgb) || type == DepthStencil)
@@ -4387,7 +4418,22 @@ void GSTextureCache::InvalidateContainedTargets(u32 start_bp, u32 end_bp, u32 wr
 					Target* const rev_t = *j;
 					if (rev_t->m_TEX0.TBP0 == t->m_TEX0.TBP0 && GSLocalMemory::m_psm[rev_t->m_TEX0.PSM].bpp == GSLocalMemory::m_psm[t->m_TEX0.PSM].bpp)
 					{
-						rev_t->m_was_dst_matched = false;
+						if (t->m_last_draw > rev_t->m_last_draw && GSUtil::GetChannelMask(t->m_TEX0.PSM) == GSUtil::GetChannelMask(rev_t->m_TEX0.PSM))
+						{
+							if (type == DepthStencil && GSUtil::GetChannelMask(t->m_TEX0.PSM) == 0x7)
+							{
+								rev_t->m_valid_rgb = false;
+								rev_t->m_was_dst_matched = false;
+							}
+							else
+							{
+								GL_CACHE("TC: InvalidateContainedTargets: Remove Target %s[%x, %s]", to_string(1 - type), rev_t->m_TEX0.TBP0, GSUtil::GetPSMName(rev_t->m_TEX0.PSM));
+								rev_list.erase(j);
+								delete rev_t;
+							}
+						}
+						else
+							rev_t->m_was_dst_matched = false;
 						break;
 					}
 					++j;
@@ -5234,7 +5280,6 @@ bool GSTextureCache::Move(u32 SBP, u32 SBW, u32 SPSM, int sx, int sy, u32 DBP, u
 			const GSVector4 src_rect = GSVector4(scaled_sx, scaled_sy, scaled_sx + scaled_w, scaled_sy + scaled_h);
 			const GSVector4 tmp_rect = src_rect / (GSVector4(tmp_texture->GetSize()).xyxy());
 			const GSVector4 dst_rect = GSVector4(scaled_dx, scaled_dy, (scaled_dx + scaled_w), (scaled_dy + scaled_h));
-			g_perfmon.Put(GSPerfMon::TextureCopies, 2);
 			g_gs_device->StretchRect(src->m_texture, tmp_rect, tmp_texture, src_rect, ShaderConvert::DEPTH_COPY, false);
 			g_gs_device->StretchRect(tmp_texture, tmp_rect, dst->m_texture, dst_rect, ShaderConvert::DEPTH_COPY, false);
 		}
@@ -5245,7 +5290,6 @@ bool GSTextureCache::Move(u32 SBP, u32 SBW, u32 SPSM, int sx, int sy, u32 DBP, u
 				const GSVector4 src_rect = GSVector4(scaled_sx, scaled_sy, scaled_sx + scaled_w, scaled_sy + scaled_h);
 				const GSVector4 tmp_rect = src_rect / (GSVector4(tmp_texture->GetSize()).xyxy());
 				const GSVector4 dst_rect = GSVector4(scaled_dx, scaled_dy, (scaled_dx + scaled_w), (scaled_dy + scaled_h));
-				g_perfmon.Put(GSPerfMon::TextureCopies, 2);
 				g_gs_device->StretchRect(src->m_texture, tmp_rect, tmp_texture, src_rect, false, false, false, true);
 				g_gs_device->StretchRect(tmp_texture, tmp_rect, dst->m_texture, dst_rect, false, false, false, true);
 			}
@@ -5273,7 +5317,6 @@ bool GSTextureCache::Move(u32 SBP, u32 SBW, u32 SPSM, int sx, int sy, u32 DBP, u
 		{
 			const ShaderConvert shader = ShaderConvert::COPY;
 			const GSVector4 dst_rect = GSVector4(scaled_dx, scaled_dy, (scaled_dx + scaled_w), (scaled_dy + scaled_h));
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			g_gs_device->StretchRect(src->m_texture, src_rect, dst->m_texture, dst_rect, false, false, false, true, shader);
 		}
 		else if (src->m_type != dst->m_type)
@@ -5291,13 +5334,11 @@ bool GSTextureCache::Move(u32 SBP, u32 SBW, u32 SPSM, int sx, int sy, u32 DBP, u
 				default:
 					break;
 			}
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			g_gs_device->StretchRect(src->m_texture, src_rect, dst->m_texture, scaled_src_rect, shader, false);
 		}
 		else if (src->m_texture->IsDepthStencil())
 		{
 			const GSVector4 dst_rect = GSVector4(scaled_dx, scaled_dy, (scaled_dx + scaled_w), (scaled_dy + scaled_h));
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			g_gs_device->StretchRect(src->m_texture, src_rect, dst->m_texture, dst_rect, ShaderConvert::DEPTH_COPY, false);
 		}
 		else
@@ -6239,7 +6280,6 @@ GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, con
 			{
 				if (dst->m_rt_alpha_scale)
 				{
-					g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 					const GSVector4 sRectF = GSVector4(sRect) / GSVector4(1, 1, sTex->GetWidth(), sTex->GetHeight());
 					g_gs_device->StretchRect(
 						sTex, sRectF, dTex, GSVector4(destX, destY, sRect.width(), sRect.height()), ShaderConvert::RTA_DECORRECTION, false);
@@ -6291,8 +6331,6 @@ GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, con
 						sTex, sRectF, dTex, GSVector4(destX, destY, new_size.x, new_size.y), shader, false);
 				}
 
-				g_perfmon.Put(GSPerfMon::TextureCopies, 1);
-
 #ifdef PCSX2_DEVBUILD
 				if (GSConfig.UseDebugDevice)
 				{
@@ -6742,7 +6780,6 @@ GSTextureCache::Source* GSTextureCache::CreateMergedSource(GIFRegTEX0 TEX0, GIFR
 	// Sort rect list by the texture, we want to batch as many as possible together.
 	g_gs_device->SortMultiStretchRects(copy_queue, copy_count);
 	g_gs_device->DrawMultiStretchRects(copy_queue, copy_count, dtex, ShaderConvert::COPY);
-	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 
 	if (lmtex)
 		g_gs_device->Recycle(lmtex);
@@ -7033,7 +7070,6 @@ GSTexture* GSTextureCache::LookupPaletteSource(u32 CBP, u32 CPSM, u32 CBW, GSVec
 					return nullptr;
 
 				g_gs_device->StretchRect(t->m_texture, GSVector4(0, 0, 1, 1), tex, GSVector4(GSVector4i::loadh(t->m_unscaled_size)), (t->m_type == RenderTarget) ? ShaderConvert::COPY : ShaderConvert::DEPTH_COPY, false);
-				g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 				m_target_memory_usage = (m_target_memory_usage - t->m_texture->GetMemUsage()) + tex->GetMemUsage();
 				g_gs_device->Recycle(t->m_texture);
 			
@@ -7181,7 +7217,6 @@ void GSTextureCache::Read(Target* t, const GSVector4i& r)
 		if (tmp)
 		{
 			g_gs_device->StretchRect(t->m_texture, src, tmp, GSVector4(drc), ps_shader, false);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			dltex->get()->CopyFromTexture(drc, tmp, drc, 0, true);
 			g_gs_device->Recycle(tmp);
 		}
@@ -7850,7 +7885,6 @@ void GSTextureCache::Target::Update(bool cannot_scale)
 				//GL_CACHE("TC: RT in RT Updating Z copy on draw %d z_offset %d", s_n, z_address_info.offset);
 				const GSVector4i dRect = GSVector4i(total_rect.x * m_scale, (z_address_info.offset + total_rect.y) * m_scale, (total_rect.z + (1.0f / m_scale)) * m_scale, (z_address_info.offset + total_rect.w + (1.0f / m_scale)) * m_scale);
 				g_gs_device->StretchRect(m_texture, GSVector4(total_rect.x / static_cast<float>(m_unscaled_size.x), total_rect.y / static_cast<float>(m_unscaled_size.y), (total_rect.z + (1.0f / m_scale)) / static_cast<float>(m_unscaled_size.x), (total_rect.w + (1.0f / m_scale)) / static_cast<float>(m_unscaled_size.y)), g_texture_cache->GetTemporaryZ(), GSVector4(dRect), ShaderConvert::DEPTH_COPY, false);
-				g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			}
 		}
 	}
@@ -8000,14 +8034,12 @@ bool GSTextureCache::Target::ResizeTexture(int new_unscaled_width, int new_unsca
 		{
 			// Can't do partial copies in DirectX for depth textures, and it's probably not ideal in other
 			// APIs either. So use a fullscreen quad setting depth instead.
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			g_gs_device->StretchRect(m_texture, tex, GSVector4(rc), ShaderConvert::DEPTH_COPY, true);
 		}
 		else
 		{
 			if (require_new_rect)
 			{
-				g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 				g_gs_device->StretchRect(m_texture, tex, GSVector4(rc), ShaderConvert::COPY, false);
 			}
 			else

pcsx2/GS/Renderers/Metal/GSDeviceMTL.mm

@@ -1587,6 +1587,7 @@ void GSDeviceMTL::DrawStretchRect(const GSVector4& sRect, const GSVector4& dRect
 	[m_current_render.encoder drawPrimitives:MTLPrimitiveTypeTriangleStrip
 	                             vertexStart:0
 	                             vertexCount:4];
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 	g_perfmon.Put(GSPerfMon::DrawCalls, 1);
 }
 
@@ -1597,6 +1598,7 @@ void GSDeviceMTL::RenderCopy(GSTexture* sTex, id<MTLRenderPipelineState> pipelin
 	MRESetPipeline(pipeline);
 	MRESetTexture(sTex, GSMTLTextureIndexNonHW);
 	[m_current_render.encoder drawPrimitives:MTLPrimitiveTypeTriangle vertexStart:0 vertexCount:3];
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 	g_perfmon.Put(GSPerfMon::DrawCalls, 1);
 }
 
@@ -2164,7 +2166,6 @@ void GSDeviceMTL::RenderHW(GSHWDrawConfig& config)
 		{
 			BeginRenderPass(@"ColorClip Resolve", config.rt, MTLLoadActionLoad, nullptr, MTLLoadActionDontCare);
 			RenderCopy(colclip_rt, m_colclip_resolve_pipeline, config.colclip_update_area);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 
 			Recycle(colclip_rt);
 			
@@ -2194,7 +2195,6 @@ void GSDeviceMTL::RenderHW(GSHWDrawConfig& config)
 				case GSTexture::State::Dirty:
 					BeginRenderPass(@"ColorClip Init", colclip_rt, MTLLoadActionDontCare, nullptr, MTLLoadActionDontCare);
 					RenderCopy(config.rt, m_colclip_init_pipeline, copy_rect);
-					g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 					break;
 
 				case GSTexture::State::Cleared:
@@ -2306,7 +2306,6 @@ void GSDeviceMTL::RenderHW(GSHWDrawConfig& config)
 		{
 			BeginRenderPass(@"ColorClip Resolve", config.rt, MTLLoadActionLoad, nullptr, MTLLoadActionDontCare);
 			RenderCopy(colclip_rt, m_colclip_resolve_pipeline, config.colclip_update_area);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 
 			Recycle(colclip_rt);
 			

pcsx2/GS/Renderers/OpenGL/GSDeviceOGL.cpp

@@ -1301,11 +1301,11 @@ std::string GSDeviceOGL::GenGlslHeader(const std::string_view entry, GLenum type
 	if (GLAD_GL_ARB_conservative_depth)
 	{
 		header += "#extension GL_ARB_conservative_depth : enable\n";
-		header += "#define HAS_CONSERVATIVE_DEPTH 1\n";
+		header += "#define PS_HAS_CONSERVATIVE_DEPTH 1\n";
 	}
 	else
 	{
-		header += "#define HAS_CONSERVATIVE_DEPTH 0\n";
+		header += "#define PS_HAS_CONSERVATIVE_DEPTH 0\n";
 	}
 
 	// Allow to puts several shader in 1 files
@@ -1423,10 +1423,11 @@ std::string GSDeviceOGL::GetPSSource(const PSSelector& sel)
 // Copy a sub part of texture (same as below but force a conversion)
 void GSDeviceOGL::BlitRect(GSTexture* sTex, const GSVector4i& r, const GSVector2i& dsize, bool at_origin, bool linear)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	CommitClear(sTex, true);
 
 	GL_PUSH(fmt::format("CopyRectConv from {}", static_cast<GSTextureOGL*>(sTex)->GetID()).c_str());
-	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 
 	// NOTE: This previously used glCopyTextureSubImage2D(), but this appears to leak memory in
 	// the loading screens of Evolution Snowboarding in Intel/NVIDIA drivers.
@@ -1657,6 +1658,8 @@ void GSDeviceOGL::FilteredDownsampleTexture(GSTexture* sTex, GSTexture* dTex, u3
 
 void GSDeviceOGL::DrawStretchRect(const GSVector4& sRect, const GSVector4& dRect, const GSVector2i& ds)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	// Original code from DX
 	const float left = dRect.x * 2 / ds.x - 1.0f;
 	const float right = dRect.z * 2 / ds.x - 1.0f;
@@ -1727,6 +1730,8 @@ void GSDeviceOGL::DrawMultiStretchRects(
 
 void GSDeviceOGL::DoMultiStretchRects(const MultiStretchRect* rects, u32 num_rects, const GSVector2& ds)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	const u32 vertex_reserve_size = num_rects * 4 * sizeof(GSVertexPT1);
 	const u32 index_reserve_size = num_rects * 6 * sizeof(u16);
 	auto vertex_map = m_vertex_stream_buffer->Map(sizeof(GSVertexPT1), vertex_reserve_size);
@@ -1922,6 +1927,8 @@ void GSDeviceOGL::DoShadeBoost(GSTexture* sTex, GSTexture* dTex, const float par
 
 void GSDeviceOGL::SetupDATE(GSTexture* rt, GSTexture* ds, SetDATM datm, const GSVector4i& bbox)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	GL_PUSH("DATE First Pass");
 
 	// sfex3 (after the capcom logo), vf4 (first menu fading in), ffxii shadows, rumble roses shadows, persona4 shadows
@@ -2458,7 +2465,6 @@ void GSDeviceOGL::RenderHW(GSHWDrawConfig& config)
 			const GSVector4 dRect(config.colclip_update_area);
 			const GSVector4 sRect = dRect / GSVector4(size.x, size.y).xyxy();
 			StretchRect(colclip_rt, sRect, config.rt, dRect, ShaderConvert::COLCLIP_RESOLVE, false);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			Recycle(colclip_rt);
 
 			g_gs_device->SetColorClipTexture(nullptr);
@@ -2493,7 +2499,6 @@ void GSDeviceOGL::RenderHW(GSHWDrawConfig& config)
 			const GSVector4 dRect = GSVector4((config.colclip_mode == GSHWDrawConfig::ColClipMode::ConvertOnly) ? GSVector4i::loadh(rtsize) : config.drawarea);
 			const GSVector4 sRect = dRect / GSVector4(rtsize.x, rtsize.y).xyxy();
 			StretchRect(config.rt, sRect, colclip_rt, dRect, ShaderConvert::COLCLIP_INIT, false);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 		}
 	}
 
@@ -2762,7 +2767,6 @@ void GSDeviceOGL::RenderHW(GSHWDrawConfig& config)
 			const GSVector4 dRect(config.colclip_update_area);
 			const GSVector4 sRect = dRect / GSVector4(size.x, size.y).xyxy();
 			StretchRect(colclip_rt, sRect, config.rt, dRect, ShaderConvert::COLCLIP_RESOLVE, false);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			Recycle(colclip_rt);
 
 			g_gs_device->SetColorClipTexture(nullptr);

pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp

@@ -2929,6 +2929,8 @@ void GSDeviceVK::DrawMultiStretchRects(
 void GSDeviceVK::DoMultiStretchRects(
 	const MultiStretchRect* rects, u32 num_rects, GSTextureVK* dTex, ShaderConvert shader)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	// Set up vertices first.
 	const u32 vertex_reserve_size = num_rects * 4 * sizeof(GSVertexPT1);
 	const u32 index_reserve_size = num_rects * 6 * sizeof(u16);
@@ -3082,6 +3084,8 @@ void GSDeviceVK::DoStretchRect(GSTextureVK* sTex, const GSVector4& sRect, GSText
 
 void GSDeviceVK::DrawStretchRect(const GSVector4& sRect, const GSVector4& dRect, const GSVector2i& ds)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	// ia
 	const float left = dRect.x * 2 / ds.x - 1.0f;
 	const float top = 1.0f - dRect.y * 2 / ds.y;
@@ -5468,6 +5472,8 @@ void GSDeviceVK::SetPSConstantBuffer(const GSHWDrawConfig::PSConstantBuffer& cb)
 
 void GSDeviceVK::SetupDATE(GSTexture* rt, GSTexture* ds, SetDATM datm, const GSVector4i& bbox)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	GL_PUSH("SetupDATE {%d,%d} %dx%d", bbox.left, bbox.top, bbox.width(), bbox.height());
 
 	const GSVector2i size(ds->GetSize());
@@ -5495,6 +5501,8 @@ void GSDeviceVK::SetupDATE(GSTexture* rt, GSTexture* ds, SetDATM datm, const GSV
 
 GSTextureVK* GSDeviceVK::SetupPrimitiveTrackingDATE(GSHWDrawConfig& config)
 {
+	g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+
 	// How this is done:
 	// - can't put a barrier for the image in the middle of the normal render pass, so that's out
 	// - so, instead of just filling the int texture with INT_MAX, we sample the RT and use -1 for failing values
@@ -5676,7 +5684,6 @@ void GSDeviceVK::RenderHW(GSHWDrawConfig& config)
 			SetPipeline(m_colclip_finish_pipelines[pipe.ds][pipe.IsRTFeedbackLoop()]);
 			SetUtilityTexture(colclip_rt, m_point_sampler);
 			DrawStretchRect(sRect, drawareaf, rtsize);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 
 			Recycle(colclip_rt);
 			g_gs_device->SetColorClipTexture(nullptr);
@@ -5893,7 +5900,6 @@ void GSDeviceVK::RenderHW(GSHWDrawConfig& config)
 		const GSVector4 drawareaf = GSVector4((config.colclip_mode == GSHWDrawConfig::ColClipMode::ConvertOnly) ? GSVector4i::loadh(rtsize) : config.drawarea);
 		const GSVector4 sRect(drawareaf / GSVector4(rtsize).xyxy());
 		DrawStretchRect(sRect, drawareaf, rtsize);
-		g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 
 		GL_POP();
 		OMSetRenderTargets(draw_rt, draw_ds, config.scissor, static_cast<FeedbackLoopFlag>(pipe.feedback_loop_flags));
@@ -5995,7 +6001,6 @@ void GSDeviceVK::RenderHW(GSHWDrawConfig& config)
 			SetPipeline(m_colclip_finish_pipelines[pipe.ds][pipe.IsRTFeedbackLoop()]);
 			SetUtilityTexture(colclip_rt, m_point_sampler);
 			DrawStretchRect(sRect, drawareaf, rtsize);
-			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 
 			Recycle(colclip_rt);
 			g_gs_device->SetColorClipTexture(nullptr);

pcsx2/GameDatabase.cpp

@@ -380,7 +380,6 @@ static const char* s_gs_hw_fix_names[] = {
 	"trilinearFiltering",
 	"skipDrawStart",
 	"skipDrawEnd",
-	"halfBottomOverride",
 	"halfPixelOffset",
 	"roundSprite",
 	"nativeScaling",

pcsx2/GameDatabase.h

@@ -65,7 +65,6 @@ namespace GameDatabaseSchema
 		TrilinearFiltering,
 		SkipDrawStart,
 		SkipDrawEnd,
-		HalfBottomOverride,
 		HalfPixelOffset,
 		RoundSprite,
 		NativeScaling,

pcsx2/ImGui/FullscreenUI_Internal.h

@@ -0,0 +1,496 @@
+// SPDX-FileCopyrightText: 2002-2026 PCSX2 Dev Team
+// SPDX-License-Identifier: GPL-3.0+
+
+#pragma once
+
+#include "FullscreenUI.h"
+#include "ImGuiFullscreen.h"
+
+#include "common/Timer.h"
+#include "Input/InputManager.h"
+
+#define TR_CONTEXT "FullscreenUI"
+
+template <size_t L>
+class IconStackString : public SmallStackString<L>
+{
+public:
+	__fi IconStackString(const char* icon, const char* str)
+	{
+		SmallStackString<L>::format("{} {}", icon, Host::TranslateToStringView(TR_CONTEXT, str));
+	}
+	__fi IconStackString(const char8_t* icon, const char* str)
+	{
+		SmallStackString<L>::format("{} {}", reinterpret_cast<const char*>(icon), Host::TranslateToStringView(TR_CONTEXT, str));
+	}
+	__fi IconStackString(const char* icon, const char* str, const char* suffix)
+	{
+		SmallStackString<L>::format("{} {}##{}", icon, Host::TranslateToStringView(TR_CONTEXT, str), suffix);
+	}
+	__fi IconStackString(const char8_t* icon, const char* str, const char* suffix)
+	{
+		SmallStackString<L>::format("{} {}##{}", reinterpret_cast<const char*>(icon), Host::TranslateToStringView(TR_CONTEXT, str), suffix);
+	}
+};
+
+#define FSUI_ICONSTR(icon, str) IconStackString<256>(icon, str).c_str()
+#define FSUI_ICONSTR_S(icon, str, suffix) IconStackString<256>(icon, str, suffix).c_str()
+#define FSUI_STR(str) Host::TranslateToString(TR_CONTEXT, str)
+#define FSUI_CSTR(str) Host::TranslateToCString(TR_CONTEXT, str)
+#define FSUI_VSTR(str) Host::TranslateToStringView(TR_CONTEXT, str)
+#define FSUI_FSTR(str) fmt::runtime(Host::TranslateToStringView(TR_CONTEXT, str))
+#define FSUI_NSTR(str) str
+
+using ImGuiFullscreen::ActiveButton;
+using ImGuiFullscreen::AddNotification;
+using ImGuiFullscreen::BeginFullscreenColumns;
+using ImGuiFullscreen::BeginFullscreenColumnWindow;
+using ImGuiFullscreen::BeginFullscreenWindow;
+using ImGuiFullscreen::BeginHorizontalMenu;
+using ImGuiFullscreen::BeginMenuButtons;
+using ImGuiFullscreen::BeginNavBar;
+using ImGuiFullscreen::CenterImage;
+using ImGuiFullscreen::CloseChoiceDialog;
+using ImGuiFullscreen::CloseFileSelector;
+using ImGuiFullscreen::EndFullscreenColumns;
+using ImGuiFullscreen::EndFullscreenColumnWindow;
+using ImGuiFullscreen::EndFullscreenWindow;
+using ImGuiFullscreen::EndHorizontalMenu;
+using ImGuiFullscreen::EndMenuButtons;
+using ImGuiFullscreen::EndNavBar;
+using ImGuiFullscreen::EnumChoiceButton;
+using ImGuiFullscreen::FloatingButton;
+using ImGuiFullscreen::FocusResetType;
+using ImGuiFullscreen::ForceKeyNavEnabled;
+using ImGuiFullscreen::g_large_font;
+using ImGuiFullscreen::g_layout_padding_left;
+using ImGuiFullscreen::g_layout_padding_top;
+using ImGuiFullscreen::g_medium_font;
+using ImGuiFullscreen::GetCachedSvgTexture;
+using ImGuiFullscreen::GetCachedSvgTextureAsync;
+using ImGuiFullscreen::GetCachedTexture;
+using ImGuiFullscreen::GetCachedTextureAsync;
+using ImGuiFullscreen::GetPlaceholderTexture;
+using ImGuiFullscreen::GetQueuedFocusResetType;
+using ImGuiFullscreen::HorizontalMenuItem;
+using ImGuiFullscreen::HorizontalMenuSvgItem;
+using ImGuiFullscreen::InputFilterType;
+using ImGuiFullscreen::IsFocusResetQueued;
+using ImGuiFullscreen::IsGamepadInputSource;
+using ImGuiFullscreen::LAYOUT_FOOTER_HEIGHT;
+using ImGuiFullscreen::LAYOUT_LARGE_FONT_SIZE;
+using ImGuiFullscreen::LAYOUT_MEDIUM_FONT_SIZE;
+using ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT;
+using ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT_NO_SUMMARY;
+using ImGuiFullscreen::LAYOUT_MENU_BUTTON_X_PADDING;
+using ImGuiFullscreen::LAYOUT_MENU_BUTTON_Y_PADDING;
+using ImGuiFullscreen::LAYOUT_SCREEN_HEIGHT;
+using ImGuiFullscreen::LAYOUT_SCREEN_WIDTH;
+using ImGuiFullscreen::LayoutScale;
+using ImGuiFullscreen::LoadSvgTexture;
+using ImGuiFullscreen::LoadTexture;
+using ImGuiFullscreen::MenuButton;
+using ImGuiFullscreen::MenuButtonFrame;
+using ImGuiFullscreen::MenuButtonWithoutSummary;
+using ImGuiFullscreen::MenuButtonWithValue;
+using ImGuiFullscreen::MenuHeading;
+using ImGuiFullscreen::MenuHeadingButton;
+using ImGuiFullscreen::MenuImageButton;
+using ImGuiFullscreen::ModAlpha;
+using ImGuiFullscreen::MulAlpha;
+using ImGuiFullscreen::NavButton;
+using ImGuiFullscreen::NavTitle;
+using ImGuiFullscreen::OpenChoiceDialog;
+using ImGuiFullscreen::OpenConfirmMessageDialog;
+using ImGuiFullscreen::OpenFileSelector;
+using ImGuiFullscreen::OpenInfoMessageDialog;
+using ImGuiFullscreen::OpenInputStringDialog;
+using ImGuiFullscreen::PopPrimaryColor;
+using ImGuiFullscreen::PushPrimaryColor;
+using ImGuiFullscreen::QueueResetFocus;
+using ImGuiFullscreen::ResetFocusHere;
+using ImGuiFullscreen::RightAlignNavButtons;
+using ImGuiFullscreen::SetFullscreenFooterText;
+using ImGuiFullscreen::ShowToast;
+using ImGuiFullscreen::SvgScaling;
+using ImGuiFullscreen::ThreeWayToggleButton;
+using ImGuiFullscreen::ToggleButton;
+using ImGuiFullscreen::UIBackgroundColor;
+using ImGuiFullscreen::UIBackgroundHighlightColor;
+using ImGuiFullscreen::UIBackgroundLineColor;
+using ImGuiFullscreen::UIBackgroundTextColor;
+using ImGuiFullscreen::UIDisabledColor;
+using ImGuiFullscreen::UIPopupBackgroundColor;
+using ImGuiFullscreen::UIPrimaryColor;
+using ImGuiFullscreen::UIPrimaryDarkColor;
+using ImGuiFullscreen::UIPrimaryLightColor;
+using ImGuiFullscreen::UIPrimaryLineColor;
+using ImGuiFullscreen::UIPrimaryTextColor;
+using ImGuiFullscreen::UISecondaryColor;
+using ImGuiFullscreen::UISecondaryStrongColor;
+using ImGuiFullscreen::UISecondaryTextColor;
+using ImGuiFullscreen::UISecondaryWeakColor;
+using ImGuiFullscreen::UITextHighlightColor;
+using ImGuiFullscreen::WantsToCloseMenu;
+
+namespace FullscreenUI
+{
+	enum class MainWindowType
+	{
+		None,
+		Landing,
+		StartGame,
+		Exit,
+		GameList,
+		GameListSettings,
+		Settings,
+		PauseMenu,
+		Achievements,
+		Leaderboards,
+	};
+
+	enum class PauseSubMenu
+	{
+		None,
+		Exit,
+		Achievements,
+	};
+
+	enum class SettingsPage
+	{
+		Summary,
+		Interface,
+		BIOS,
+		Emulation,
+		Graphics,
+		Audio,
+		MemoryCard,
+		NetworkHDD,
+		Folders,
+		Achievements,
+		Controller,
+		Hotkey,
+		Advanced,
+		Patches,
+		Cheats,
+		GameFixes,
+		Count
+	};
+
+	enum class GameListView
+	{
+		Grid,
+		List,
+		Count
+	};
+
+	enum class IPAddressType
+	{
+		PS2IP,
+		SubnetMask,
+		Gateway,
+		DNS1,
+		DNS2,
+		Other
+	};
+
+	//////////////////////////////////////////////////////////////////////////
+	// Main
+	//////////////////////////////////////////////////////////////////////////
+	void UpdateGameDetails(std::string path, std::string serial, std::string title, u32 disc_crc, u32 crc);
+	bool AreAnyDialogsOpen();
+	void PauseForMenuOpen(bool set_pause_menu_open);
+	void ClosePauseMenu();
+	void OpenPauseSubMenu(PauseSubMenu submenu);
+	void DrawLandingTemplate(ImVec2* menu_pos, ImVec2* menu_size);
+	void DrawLandingWindow();
+	void DrawStartGameWindow();
+	void DrawExitWindow();
+	void DrawPauseMenu(MainWindowType type);
+	void ExitFullscreenAndOpenURL(const std::string_view url);
+	void CopyTextToClipboard(std::string title, const std::string_view text);
+	void DrawAboutWindow();
+	void OpenAboutWindow();
+	void GetStandardSelectionFooterText(SmallStringBase& dest, bool back_instead_of_cancel);
+	void ApplyLayoutSettings(const SettingsInterface* bsi = nullptr);
+
+	void DrawSvgTexture(GSTexture* padded_texture, ImVec2 unpadded_size);
+	void DrawCachedSvgTexture(const std::string& path, ImVec2 size, SvgScaling mode);
+	void DrawCachedSvgTextureAsync(const std::string& path, ImVec2 size, SvgScaling mode);
+	void DrawListSvgTexture(ImDrawList* drawList, GSTexture* padded_texture, const ImVec2& p_min, const ImVec2& p_unpadded_max);
+
+	inline MainWindowType s_current_main_window = MainWindowType::None;
+	inline PauseSubMenu s_current_pause_submenu = PauseSubMenu::None;
+	inline bool s_initialized = false;
+	inline bool s_tried_to_initialize = false;
+	inline bool s_pause_menu_was_open = false;
+	inline bool s_was_paused_on_quick_menu_open = false;
+	inline bool s_about_window_open = false;
+
+	// achievements login dialog state
+	inline bool s_achievements_login_open = false;
+	inline bool s_achievements_login_logging_in = false;
+	inline char s_achievements_login_username[256] = {};
+	inline char s_achievements_login_password[256] = {};
+	inline Achievements::LoginRequestReason s_achievements_login_reason = Achievements::LoginRequestReason::UserInitiated;
+
+	// local copies of the currently-running game
+	inline std::string s_current_game_title;
+	inline std::string s_current_game_subtitle;
+	inline std::string s_current_disc_serial;
+	inline std::string s_current_disc_path;
+	inline u32 s_current_disc_crc;
+
+	//////////////////////////////////////////////////////////////////////////
+	// Resources
+	//////////////////////////////////////////////////////////////////////////
+	bool LoadResources();
+	bool LoadSvgResources();
+	void DestroyResources();
+
+	inline std::array<std::shared_ptr<GSTexture>, static_cast<u32>(GameDatabaseSchema::Compatibility::Perfect)>
+		s_game_compatibility_textures;
+	inline std::shared_ptr<GSTexture> s_banner_texture;
+	inline std::vector<std::unique_ptr<GSTexture>> s_cleanup_textures;
+
+	//////////////////////////////////////////////////////////////////////////
+	// Landing
+	//////////////////////////////////////////////////////////////////////////
+	void SwitchToLanding();
+	ImGuiFullscreen::FileSelectorFilters GetOpenFileFilters();
+	ImGuiFullscreen::FileSelectorFilters GetDiscImageFilters();
+	ImGuiFullscreen::FileSelectorFilters GetAudioFileFilters();
+	ImGuiFullscreen::FileSelectorFilters GetImageFileFilters();
+	void DoVMInitialize(const VMBootParameters& boot_params, bool switch_to_landing_on_failure);
+	void DoStartPath(
+		const std::string& path, std::optional<s32> state_index = std::nullopt, std::optional<bool> fast_boot = std::nullopt);
+	void DoStartFile();
+	void DoStartBIOS();
+	void DoStartDisc(const std::string& drive);
+	void DoStartDisc();
+	void DoToggleFrameLimit();
+	void DoToggleSoftwareRenderer();
+	void RequestShutdown(bool save_state);
+	void DoShutdown(bool save_state);
+	void RequestReset();
+	void DoReset();
+	void DoChangeDiscFromFile();
+	void RequestChangeDisc();
+	void DoRequestExit();
+	void DoDesktopMode();
+	void DoToggleFullscreen();
+
+	void ConfirmShutdownIfMemcardBusy(std::function<void(bool)> callback);
+
+	bool ShouldDefaultToGameList();
+
+	//////////////////////////////////////////////////////////////////////////
+	// Save State List
+	//////////////////////////////////////////////////////////////////////////
+	struct SaveStateListEntry
+	{
+		std::string title;
+		std::string summary;
+		std::string path;
+		std::unique_ptr<GSTexture> preview_texture;
+		time_t timestamp;
+		s32 slot;
+	};
+
+	void InitializePlaceholderSaveStateListEntry(SaveStateListEntry* li, s32 slot);
+	bool InitializeSaveStateListEntry(
+		SaveStateListEntry* li, const std::string& title, const std::string& serial, u32 crc, s32 slot, bool backup = false);
+	void ClearSaveStateEntryList();
+	u32 PopulateSaveStateListEntries(const std::string& title, const std::string& serial, u32 crc);
+	bool OpenLoadStateSelectorForGame(const std::string& game_path);
+	bool OpenSaveStateSelector(bool is_loading);
+	void CloseSaveStateSelector();
+	void DrawSaveStateSelector(bool is_loading);
+	bool OpenLoadStateSelectorForGameResume(const GameList::Entry* entry);
+	void DrawResumeStateSelector();
+	void DoLoadState(std::string path, std::optional<s32> slot, bool backup);
+	void DoSaveState(s32 slot);
+
+	inline std::vector<SaveStateListEntry> s_save_state_selector_slots;
+	inline std::string s_save_state_selector_game_path;
+	inline s32 s_save_state_selector_submenu_index = -1;
+	inline bool s_save_state_selector_open = false;
+	inline bool s_save_state_selector_loading = true;
+	inline bool s_save_state_selector_resuming = false;
+
+	//////////////////////////////////////////////////////////////////////////
+	// Game List
+	//////////////////////////////////////////////////////////////////////////
+	void DrawGameListWindow();
+	void DrawGameList(const ImVec2& heading_size);
+	void DrawGameGrid(const ImVec2& heading_size);
+	void HandleGameListActivate(const GameList::Entry* entry);
+	void HandleGameListOptions(const GameList::Entry* entry);
+	void DrawGameListSettingsWindow();
+	void SwitchToGameList();
+	void PopulateGameListEntryList();
+	GSTexture* GetTextureForGameListEntryType(GameList::EntryType type, const ImVec2& size, SvgScaling mode = SvgScaling::Stretch);
+	GSTexture* GetGameListCover(const GameList::Entry* entry);
+	void DrawGameCover(const GameList::Entry* entry, const ImVec2& size);
+	void DrawGameCover(const GameList::Entry* entry, ImDrawList* draw_list, const ImVec2& min, const ImVec2& max);
+	// For when we have no GameList entry
+	void DrawFallbackCover(const ImVec2& size);
+	void DrawFallbackCover(ImDrawList* draw_list, const ImVec2& min, const ImVec2& max);
+
+	// Lazily populated cover images.
+	inline std::unordered_map<std::string, std::string> s_cover_image_map;
+	inline std::vector<const GameList::Entry*> s_game_list_sorted_entries;
+	inline GameListView s_game_list_view = GameListView::Grid;
+
+	//////////////////////////////////////////////////////////////////////////
+	// Background
+	//////////////////////////////////////////////////////////////////////////
+	void LoadCustomBackground();
+	void DrawCustomBackground();
+
+	inline std::shared_ptr<GSTexture> s_custom_background_texture;
+	inline std::string s_custom_background_path;
+	inline bool s_custom_background_enabled = false;
+
+	//////////////////////////////////////////////////////////////////////////
+	// Achievements
+	//////////////////////////////////////////////////////////////////////////
+	void SwitchToAchievementsWindow();
+	void SwitchToLeaderboardsWindow();
+	void DrawAchievementsLoginWindow();
+
+
+	//////////////////////////////////////////////////////////////////////////
+	// Settings
+	//////////////////////////////////////////////////////////////////////////
+	static constexpr double INPUT_BINDING_TIMEOUT_SECONDS = 5.0;
+	static constexpr u32 NUM_MEMORY_CARD_PORTS = 2;
+
+	void SwitchToSettings();
+	void SwitchToGameSettings();
+	void SwitchToGameSettings(const std::string& path);
+	void SwitchToGameSettings(const GameList::Entry* entry);
+	void SwitchToGameSettings(const std::string_view serial, u32 crc);
+	void DrawSettingsWindow();
+	void DrawSummarySettingsPage();
+	void DrawInterfaceSettingsPage();
+	void DrawBIOSSettingsPage();
+	void DrawEmulationSettingsPage();
+	void DrawGraphicsSettingsPage(SettingsInterface* bsi, bool show_advanced_settings);
+	void DrawAudioSettingsPage();
+	void DrawMemoryCardSettingsPage();
+	void DrawNetworkHDDSettingsPage();
+	void DrawFoldersSettingsPage();
+	void DrawAchievementsSettingsPage(std::unique_lock<std::mutex>& settings_lock);
+	void DrawControllerSettingsPage();
+	void DrawHotkeySettingsPage();
+	void DrawAdvancedSettingsPage();
+	void DrawPatchesOrCheatsSettingsPage(bool cheats);
+	void DrawGameFixesSettingsPage();
+
+	bool IsEditingGameSettings(SettingsInterface* bsi);
+	SettingsInterface* GetEditingSettingsInterface();
+	SettingsInterface* GetEditingSettingsInterface(bool game_settings);
+	bool ShouldShowAdvancedSettings(SettingsInterface* bsi);
+	void SetSettingsChanged(SettingsInterface* bsi);
+	bool GetEffectiveBoolSetting(SettingsInterface* bsi, const char* section, const char* key, bool default_value);
+	s32 GetEffectiveIntSetting(SettingsInterface* bsi, const char* section, const char* key, s32 default_value);
+	void DoCopyGameSettings();
+	void DoClearGameSettings();
+	void ResetControllerSettings();
+	void DoLoadInputProfile();
+	void DoSaveInputProfile();
+	void DoSaveInputProfile(const std::string& name);
+	void DoResetSettings();
+
+	bool DrawToggleSetting(SettingsInterface* bsi, const char* title, const char* summary, const char* section, const char* key,
+		bool default_value, bool enabled = true, bool allow_tristate = true, float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT,
+		std::pair<ImFont*, float> font = g_large_font, std::pair<ImFont*, float> summary_font = g_medium_font);
+	void DrawIntListSetting(SettingsInterface* bsi, const char* title, const char* summary, const char* section, const char* key,
+		int default_value, const char* const* options, size_t option_count, bool translate_options, int option_offset = 0,
+		bool enabled = true, float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT, std::pair<ImFont*, float> font = g_large_font,
+		std::pair<ImFont*, float> summary_font = g_medium_font);
+	void DrawIntRangeSetting(SettingsInterface* bsi, const char* title, const char* summary, const char* section, const char* key,
+		int default_value, int min_value, int max_value, const char* format = "%d", bool enabled = true,
+		float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT, std::pair<ImFont*, float> font = g_large_font, std::pair<ImFont*, float> summary_font = g_medium_font);
+	void DrawIntSpinBoxSetting(SettingsInterface* bsi, const char* title, const char* summary, const char* section, const char* key,
+		int default_value, int min_value, int max_value, int step_value, const char* format = "%d", bool enabled = true,
+		float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT, std::pair<ImFont*, float> font = g_large_font, std::pair<ImFont*, float> summary_font = g_medium_font);
+	void DrawFloatRangeSetting(SettingsInterface* bsi, const char* title, const char* summary, const char* section, const char* key,
+		float default_value, float min_value, float max_value, const char* format = "%f", float multiplier = 1.0f, bool enabled = true,
+		float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT, std::pair<ImFont*, float> font = g_large_font, std::pair<ImFont*, float> summary_font = g_medium_font);
+	void DrawFloatSpinBoxSetting(SettingsInterface* bsi, const char* title, const char* summary, const char* section,
+		const char* key, float default_value, float min_value, float max_value, float step_value, float multiplier,
+		const char* format = "%f", bool enabled = true, float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT,
+		std::pair<ImFont*, float> font = g_large_font, std::pair<ImFont*, float> summary_font = g_medium_font);
+	void DrawIntRectSetting(SettingsInterface* bsi, const char* title, const char* summary, const char* section,
+		const char* left_key, int default_left, const char* top_key, int default_top, const char* right_key, int default_right,
+		const char* bottom_key, int default_bottom, int min_value, int max_value, int step_value, const char* format = "%d",
+		bool enabled = true, float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT, std::pair<ImFont*, float> font = g_large_font,
+		std::pair<ImFont*, float> summary_font = g_medium_font);
+	void DrawStringListSetting(SettingsInterface* bsi, const char* title, const char* summary, const char* section, const char* key,
+		const char* default_value, const char* const* options, const char* const* option_values, size_t option_count,
+		bool translate_options, bool enabled = true, float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT, std::pair<ImFont*, float> font = g_large_font,
+		std::pair<ImFont*, float> summary_font = g_medium_font, const char* translation_ctx = "FullscreenUI");
+	void DrawStringListSetting(SettingsInterface* bsi, const char* title, const char* summary, const char* section, const char* key,
+		const char* default_value, SettingInfo::GetOptionsCallback options_callback, bool enabled = true,
+		float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT, std::pair<ImFont*, float> font = g_large_font, std::pair<ImFont*, float> summary_font = g_medium_font);
+	void DrawIPAddressSetting(SettingsInterface* bsi, const char* title, const char* summary, const char* section, const char* key,
+		const char* default_value, bool enabled = true, float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT,
+		std::pair<ImFont*, float> font = g_large_font, std::pair<ImFont*, float> summary_font = g_medium_font,
+		IPAddressType ip_type = IPAddressType::Other);
+	void DrawFloatListSetting(SettingsInterface* bsi, const char* title, const char* summary, const char* section, const char* key,
+		float default_value, const char* const* options, const float* option_values, size_t option_count, bool translate_options,
+		bool enabled = true, float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT, std::pair<ImFont*, float> font = g_large_font,
+		std::pair<ImFont*, float> summary_font = g_medium_font);
+	template <typename DataType, typename SizeType>
+	void DrawEnumSetting(SettingsInterface* bsi, const char* title, const char* summary, const char* section,
+		const char* key, DataType default_value,
+		std::optional<DataType> (*from_string_function)(const char* str),
+		const char* (*to_string_function)(DataType value),
+		const char* (*to_display_string_function)(DataType value), SizeType option_count,
+		bool enabled = true, float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT,
+		std::pair<ImFont*, float> font = g_large_font, std::pair<ImFont*, float> summary_font = g_medium_font);
+	void DrawFolderSetting(SettingsInterface* bsi, const char* title, const char* section, const char* key,
+		const std::string& runtime_var, float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT, std::pair<ImFont*, float> font = g_large_font,
+		std::pair<ImFont*, float> summary_font = g_medium_font);
+	void DrawPathSetting(SettingsInterface* bsi, const char* title, const char* section, const char* key, const char* default_value,
+		bool enabled = true, float height = ImGuiFullscreen::LAYOUT_MENU_BUTTON_HEIGHT, std::pair<ImFont*, float> font = g_large_font,
+		std::pair<ImFont*, float> summary_font = g_medium_font);
+	void DrawClampingModeSetting(SettingsInterface* bsi, const char* title, const char* summary, int vunum);
+	void PopulateGraphicsAdapterList();
+	void PopulateGameListDirectoryCache(SettingsInterface* si);
+	void PopulatePatchesAndCheatsList(const std::string_view serial, u32 crc);
+	void BeginInputBinding(SettingsInterface* bsi, InputBindingInfo::Type type, const std::string_view section,
+		const std::string_view key, const std::string_view display_name);
+	void DrawInputBindingWindow();
+	void DrawInputBindingButton(SettingsInterface* bsi, InputBindingInfo::Type type, const char* section, const char* name, const char* display_name, const char* icon_name, bool show_type = true);
+	void ClearInputBindingVariables();
+	void StartAutomaticBinding(u32 port);
+	void DrawSettingInfoSetting(SettingsInterface* bsi, const char* section, const char* key, const SettingInfo& si,
+		const char* translation_ctx);
+	void OpenMemoryCardCreateDialog();
+	void DoCreateMemoryCard(std::string name, MemoryCardType type, MemoryCardFileType file_type, bool use_ntfs_compression = false);
+
+	inline SettingsPage s_settings_page = SettingsPage::Interface;
+	inline std::unique_ptr<INISettingsInterface> s_game_settings_interface;
+	inline std::unique_ptr<GameList::Entry> s_game_settings_entry;
+	inline std::vector<std::pair<std::string, bool>> s_game_list_directories_cache;
+	inline std::vector<GSAdapterInfo> s_graphics_adapter_list_cache;
+	inline std::vector<Patch::PatchInfo> s_game_patch_list;
+	inline std::vector<std::string> s_enabled_game_patch_cache;
+	inline std::vector<Patch::PatchInfo> s_game_cheats_list;
+	inline std::vector<std::string> s_enabled_game_cheat_cache;
+	inline u32 s_game_cheat_unlabelled_count = 0;
+	inline std::vector<const HotkeyInfo*> s_hotkey_list_cache;
+	inline std::atomic_bool s_settings_changed{false};
+	inline std::atomic_bool s_game_settings_changed{false};
+	inline InputBindingInfo::Type s_input_binding_type = InputBindingInfo::Type::Unknown;
+	inline std::string s_input_binding_section;
+	inline std::string s_input_binding_key;
+	inline std::string s_input_binding_display_name;
+	inline std::vector<InputBindingKey> s_input_binding_new_bindings;
+	inline std::vector<std::pair<InputBindingKey, std::pair<float, float>>> s_input_binding_value_ranges;
+	inline Common::Timer s_input_binding_timer;
+
+} // namespace FullscreenUI

pcsx2/SPU2/Dma.cpp

@@ -163,7 +163,7 @@ void V_Core::StartADMAWrite(u16* pMem, u32 sz)
 	if ((AutoDMACtrl & (Index + 1)) == 0)
 	{
 		ActiveTSA = 0x2000 + (Index << 10);
-		DMAICounter = size * 4;
+		DMAICounter = size * 48;
 		LastClock = psxRegs.cycle;
 	}
 	else if (size >= 256)
@@ -191,7 +191,7 @@ void V_Core::StartADMAWrite(u16* pMem, u32 sz)
 		if (SPU2::MsgToConsole())
 			SPU2::ConLog("ADMA%c Error Size of %x too small\n", GetDmaIndexChar(), size);
 		InputDataLeft = 0;
-		DMAICounter = size * 4;
+		DMAICounter = size * 48;
 		LastClock = psxRegs.cycle;
 	}
 }
@@ -248,7 +248,7 @@ void V_Core::FinishDMAwrite()
 		DMA7LogWrite(DMAPtr, ReadSize << 1);
 #endif
 
-	u32 buff1end = ActiveTSA + std::min(ReadSize, (u32)0x100 + std::abs(DMAICounter / 4));
+	u32 buff1end = ActiveTSA + std::min(ReadSize, (u32)0x100 + std::abs(DMAICounter / 48));
 	u32 buff2end = 0;
 	if (buff1end > 0x100000)
 	{
@@ -343,7 +343,7 @@ void V_Core::FinishDMAwrite()
 	DMAPtr += TDA - ActiveTSA;
 	ReadSize -= TDA - ActiveTSA;
 
-	DMAICounter = (DMAICounter - ReadSize) * 4;
+	DMAICounter = (DMAICounter - ReadSize) * 48;
 
 	CounterUpdate(DMAICounter);
 
@@ -354,7 +354,7 @@ void V_Core::FinishDMAwrite()
 
 void V_Core::FinishDMAread()
 {
-	u32 buff1end = ActiveTSA + std::min(ReadSize, (u32)0x100 + std::abs(DMAICounter / 4));
+	u32 buff1end = ActiveTSA + std::min(ReadSize, (u32)0x100 + std::abs(DMAICounter / 48));
 	u32 buff2end = 0;
 
 	if (buff1end > 0x100000)
@@ -426,9 +426,9 @@ void V_Core::FinishDMAread()
 
 	// DMA Reads are done AFTER the delay, so to get the timing right we need to scheule one last DMA to catch IRQ's
 	if (ReadSize)
-		DMAICounter = std::min(ReadSize, (u32)0x100) * 4;
+		DMAICounter = std::min(ReadSize, (u32)0x100) * 48;
 	else
-		DMAICounter = 4;
+		DMAICounter = 48;
 
 	CounterUpdate(DMAICounter);
 
@@ -446,7 +446,7 @@ void V_Core::DoDMAread(u16* pMem, u32 size)
 	ReadSize = size;
 	IsDMARead = true;
 	LastClock = psxRegs.cycle;
-	DMAICounter = std::min(ReadSize, (u32)0x100) * 4;
+	DMAICounter = (std::min(ReadSize, (u32)0x100) * 48);
 	Regs.STATX &= ~0x80;
 	Regs.STATX |= 0x400;
 	//Regs.ATTR |= 0x30;
@@ -470,7 +470,7 @@ void V_Core::DoDMAwrite(u16* pMem, u32 size)
 	{
 		Regs.STATX &= ~0x80;
 		//Regs.ATTR |= 0x30;
-		DMAICounter = 1 * 4;
+		DMAICounter = 1 * 48;
 		LastClock = psxRegs.cycle;
 		return;
 	}

pcsx2/ShaderCacheVersion.h

@@ -3,4 +3,4 @@
 
 /// Version number for GS and other shaders. Increment whenever any of the contents of the
 /// shaders change, to invalidate the cache.
-static constexpr u32 SHADER_CACHE_VERSION = 81;
+static constexpr u32 SHADER_CACHE_VERSION = 82;

pcsx2/pcsx2.vcxproj

@@ -254,6 +254,7 @@
     <ClCompile Include="Host\SDLAudioStream.cpp" />
     <ClCompile Include="Hotkeys.cpp" />
     <ClCompile Include="ImGui\FullscreenUI.cpp" />
+    <ClCompile Include="ImGui\FullscreenUI_Settings.cpp" />
     <ClCompile Include="ImGui\ImGuiFullscreen.cpp" />
     <ClCompile Include="ImGui\ImGuiManager.cpp" />
     <ClCompile Include="ImGui\ImGuiOverlays.cpp" />
@@ -700,6 +701,7 @@
     <ClInclude Include="Host\AudioStream.h" />
     <ClInclude Include="Host\AudioStreamTypes.h" />
     <ClInclude Include="ImGui\FullscreenUI.h" />
+    <ClInclude Include="ImGui\FullscreenUI_Internal.h" />
     <ClInclude Include="ImGui\ImGuiAnimated.h" />
     <ClInclude Include="ImGui\ImGuiFullscreen.h" />
     <ClInclude Include="ImGui\ImGuiManager.h" />

pcsx2/pcsx2.vcxproj.filters

@@ -1352,6 +1352,9 @@
     <ClCompile Include="ImGui\FullscreenUI.cpp">
       <Filter>Misc\ImGui</Filter>
     </ClCompile>
+    <ClCompile Include="ImGui\FullscreenUI_Settings.cpp">
+      <Filter>Misc\ImGui</Filter>
+    </ClCompile>
     <ClCompile Include="ImGui\ImGuiFullscreen.cpp">
       <Filter>Misc\ImGui</Filter>
     </ClCompile>
@@ -2304,6 +2307,9 @@
     <ClInclude Include="ImGui\FullscreenUI.h">
       <Filter>Misc\ImGui</Filter>
     </ClInclude>
+    <ClInclude Include="ImGui\FullscreenUI_Internal.h">
+      <Filter>Misc\ImGui</Filter>
+    </ClInclude>
     <ClInclude Include="ImGui\ImGuiFullscreen.h">
       <Filter>Misc\ImGui</Filter>
     </ClInclude>

tools/generate_fullscreen_ui_translation_strings.py

@@ -5,90 +5,110 @@ import os
 START_IDENT = "// TRANSLATION-STRING-AREA-BEGIN"
 END_IDENT = "// TRANSLATION-STRING-AREA-END"
 
-src_file = os.path.join(os.path.dirname(__file__), "..", "pcsx2", "ImGui", "FullscreenUI.cpp")
+src_files = [
+    os.path.join(os.path.dirname(__file__), "..", "pcsx2", "ImGui", "FullscreenUI.cpp"),
+    os.path.join(os.path.dirname(__file__), "..", "pcsx2", "ImGui", "FullscreenUI_Settings.cpp"),
+]
 
-with open(src_file, "r") as f:
-    full_source = f.read()
+def extract_strings_from_source(source_content):
+    """Extract FSUI translation strings from source content."""
+    strings = []
+    for token in ["FSUI_STR", "FSUI_CSTR", "FSUI_FSTR", "FSUI_NSTR", "FSUI_VSTR", "FSUI_ICONSTR", "FSUI_ICONSTR_S"]:
+        token_len = len(token)
+        last_pos = 0
+        while True:
+            last_pos = source_content.find(token, last_pos)
+            if last_pos < 0:
+                break
 
-strings = []
-for token in ["FSUI_STR", "FSUI_CSTR", "FSUI_FSTR", "FSUI_NSTR", "FSUI_VSTR", "FSUI_ICONSTR", "FSUI_ICONSTR_S"]:
-    token_len = len(token)
-    last_pos = 0
-    while True:
-        last_pos = full_source.find(token, last_pos)
-        if last_pos < 0:
-            break
+            if last_pos >= 8 and source_content[last_pos - 8:last_pos] == "#define ":
+                last_pos += len(token)
+                continue
 
-        if last_pos >= 8 and full_source[last_pos - 8:last_pos] == "#define ":
-            last_pos += len(token)
-            continue
+            if source_content[last_pos + token_len] == '(':
+                start_pos = last_pos + token_len + 1
+                end_pos = source_content.find(")", start_pos)
+                s = source_content[start_pos:end_pos]
 
-        if full_source[last_pos + token_len] == '(':
-            start_pos = last_pos + token_len + 1
-            end_pos = full_source.find(")", start_pos)
-            s = full_source[start_pos:end_pos]
+                # Split into string arguments, removing "
+                string_args = [""]
+                arg = 0;
+                cpos = s.find(',')
+                pos = s.find('"')
+                while pos >= 0 or cpos >= 0:
+                    assert pos == 0 or s[pos - 1] != '\\'
+                    if cpos == -1 or pos < cpos:
+                        epos = pos
+                        while True:
+                            epos = s.find('"', epos + 1)
+                            # found ')' in string, extend s to next ')'
+                            if epos == -1:
+                                end_pos = source_content.find(")", end_pos + 1)
+                                s = source_content[start_pos:end_pos]
+                                epos = pos
+                                continue
 
-            # Split into sting arguments, removing "
-            string_args = [""]
-            arg = 0;
-            cpos = s.find(',')
-            pos = s.find('"')
-            while pos >= 0 or cpos >= 0:
-                assert pos == 0 or s[pos - 1] != '\\'
-                if cpos == -1 or pos < cpos:
-                    epos = pos
-                    while True:
-                        epos = s.find('"', epos + 1)
-                        # found ')' in string, extend s to next ')'
-                        if epos == -1:
-                            end_pos = full_source.find(")", end_pos + 1)
-                            s = full_source[start_pos:end_pos]
-                            epos = pos
-                            continue
+                            if s[epos - 1] == '\\':
+                                continue
+                            else:
+                                break
 
-                        if s[epos - 1] == '\\':
-                            continue
-                        else:
-                            break
+                        assert epos > pos
+                        string_args[arg] += s[pos+1:epos]
+                        cpos = s.find(',', epos + 1)
+                        pos = s.find('"', epos + 1)
+                    else:
+                        arg += 1
+                        string_args.append("")
+                        cpos = s.find(',', cpos + 1)
 
-                    assert epos > pos
-                    string_args[arg] += s[pos+1:epos]
-                    cpos = s.find(',', epos + 1)
-                    pos = s.find('"', epos + 1)
+                print(string_args)
+
+                # FSUI_ICONSTR and FSUI_ICONSTR_S need to translate the only the second argument
+                # other defines take only a single argument
+                if len(string_args) >= 2:
+                    new_s = string_args[1]
                 else:
-                    arg += 1
-                    string_args.append("")
-                    cpos = s.find(',', cpos + 1)
+                    new_s = string_args[0]
 
-            print(string_args)
+                assert len(new_s) > 0
 
-            # FSUI_ICONSTR and FSUI_ICONSTR_S need to translate the only the second argument
-            # other defines take only a single argument
-            if len(string_args) >= 2:
-                new_s = string_args[1]
-            else:
-                new_s = string_args[0]
+                if new_s not in strings:
+                    strings.append(new_s)
+            last_pos += len(token)
+    return strings
 
-            assert len(new_s) > 0
+def process_file(src_file):
+    """Process a single source file extract strings and update its translation area."""
+    print(f"\nProcessing: {src_file}")
+    
+    with open(src_file, "r") as f:
+        source = f.read()
 
-            #assert (end_pos - start_pos) < 300
-            #if (end_pos - start_pos) >= 300:
-            #    print("WARNING: Long string")
-            #    print(new_s)
-            if new_s not in strings:
-                strings.append(new_s)
-        last_pos += len(token)
+    start = source.find(START_IDENT)
+    end = source.find(END_IDENT)
+    
+    if start < 0 or end <= start:
+        print(f"  Warning: No translation string area found in {src_file}")
+        return 0
+    
+    source_without_area = source[:start] + source[end + len(END_IDENT):]
+    strings = extract_strings_from_source(source_without_area)
+    
+    print(f"  Found {len(strings)} unique strings.")
+    
+    new_area = ""
+    for string in strings:
+        new_area += f"TRANSLATE_NOOP(\"FullscreenUI\", \"{string}\");\n"
+    
+    new_source = source[:start + len(START_IDENT) + 1] + new_area + source[end:]
+    with open(src_file, "w") as f:
+        f.write(new_source)
+    
+    return len(strings)
 
-print(f"Found {len(strings)} unique strings.")
+total_strings = 0
+for src_file in src_files:
+    total_strings += process_file(src_file)
 
-start = full_source.find(START_IDENT)
-end = full_source.find(END_IDENT)
-assert start >= 0 and end > start
-
-new_area = ""
-for string in list(strings):
-    new_area += f"TRANSLATE_NOOP(\"FullscreenUI\", \"{string}\");\n"
-
-full_source = full_source[:start+len(START_IDENT)+1] + new_area + full_source[end:]
-with open(src_file, "w") as f:
-    f.write(full_source)
+print(f"\nTotal: {total_strings} unique strings across all files.")