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.

pcsx2-qt/Translations/pcsx2-qt_en.ts

@@ -11905,7 +11905,7 @@ This action cannot be undone.</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp" line="5052"/>
+        <location filename="../../pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp" line="5056"/>
         <source>Spin GPU During Readbacks is enabled, but calibrated timestamps are unavailable.  This might be really slow.</source>
         <translation type="unfinished"></translation>
     </message>
@@ -11934,12 +11934,12 @@ To use the Vulkan renderer, you should remove this app package.</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/DX11/GSDevice11.cpp" line="117"/>
+        <location filename="../../pcsx2/GS/Renderers/DX11/GSDevice11.cpp" line="126"/>
         <source>Failed to create D3D11 device: 0x{:08X}. A GPU which supports Direct3D Feature Level 10.0 is required.</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/DX11/GSDevice11.cpp" line="516"/>
+        <location filename="../../pcsx2/GS/Renderers/DX11/GSDevice11.cpp" line="525"/>
         <source>The Direct3D 11 renderer is running at feature level 10.0. This is an UNSUPPORTED configuration.
 Do not request support, please upgrade your hardware/drivers first.</source>
         <translation type="unfinished"></translation>

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

@@ -1314,6 +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();
@@ -1321,6 +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);
 
@@ -1381,11 +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);
 
@@ -1540,6 +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;
@@ -2181,6 +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);
@@ -2188,6 +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);
@@ -2401,6 +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);
 }
 
@@ -2437,7 +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])
 			{
@@ -2449,14 +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])
 			{
@@ -2468,7 +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());
 		}
 	}
 }
@@ -2477,9 +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;
@@ -2623,7 +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);
@@ -2652,7 +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);
 		}
 	}
 
@@ -2723,20 +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);
 
@@ -2837,7 +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);

pcsx2/GS/Renderers/DX12/GSDevice12.cpp

@@ -1748,6 +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);
@@ -1885,6 +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;
@@ -3906,6 +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());
@@ -3938,6 +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
@@ -4084,7 +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);
@@ -4286,7 +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();
 
@@ -4364,7 +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/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;
@@ -7092,8 +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())
 		{
@@ -7121,9 +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;
@@ -9230,11 +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

@@ -2834,7 +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.
@@ -2865,7 +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;
@@ -2938,7 +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.
@@ -2962,7 +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);
@@ -3045,7 +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;
@@ -3326,7 +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);
 					}
 				}
 
@@ -4022,7 +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;
 			}
@@ -4048,7 +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;
 			}
@@ -4108,7 +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;
@@ -4242,7 +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.
@@ -4253,7 +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);
@@ -5291,7 +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);
 		}
@@ -5302,7 +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);
 			}
@@ -5330,7 +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)
@@ -5348,13 +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
@@ -6296,7 +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);
@@ -6348,8 +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)
 				{
@@ -6799,7 +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);
@@ -7090,7 +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);
 			
@@ -7238,7 +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);
 		}
@@ -7907,7 +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);
 			}
 		}
 	}
@@ -8057,14 +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

@@ -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);