GS/DX11: Add console logs for supported features.

If not then use DXGI_FORMAT_R16G16B16A16_FLOAT,
it will be less precise but will still work on feature level 10.0 and some 10.1 gpus and allow for hardware blending.

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/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/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()
@@ -1721,6 +1739,7 @@ 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);
@@ -2621,7 +2640,7 @@ 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.");
@@ -2847,6 +2866,7 @@ 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 +2875,28 @@ 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)
@@ -3048,6 +3053,7 @@ 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);

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

@@ -5910,6 +5910,7 @@ 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;

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;
 						}
@@ -3088,67 +3127,28 @@ 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 +3156,68 @@ 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 +3234,7 @@ 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 +3249,7 @@ 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)
@@ -4377,6 +4418,7 @@ 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 +4429,22 @@ 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;

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

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/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;