GS: Improve TrianglesAreQuads check to make sure 2 edges match a previous triangle

pcsx2/GS/GSState.cpp

@@ -2899,17 +2899,24 @@ bool GSState::TrianglesAreQuads(bool shuffle_check)
 		if (idx > 0)
 		{
 			const u16* const prev_tri= m_index.buff + (idx - 3);
-			GIFRegXYZ vert = v[i[0]].XYZ;
-			GIFRegXYZ last_vert = v[i[2]].XYZ;
+			GIFRegXYZ new_verts[3] = {v[i[0]].XYZ, v[i[1]].XYZ, v[i[2]].XYZ};
 
 			if (shuffle_check)
 			{
-				vert.X -= 8 << 4;
-				last_vert.X -= 8 << 4;
+				new_verts[0].X -= 8 << 4;
+				new_verts[1].X -= 8 << 4;
+				new_verts[2].X -= 8 << 4;
 			}
+			u32 match_vert_count = 0;
 
-			if (vert != m_vertex.buff[prev_tri[0]].XYZ && vert != m_vertex.buff[prev_tri[1]].XYZ && vert != m_vertex.buff[prev_tri[2]].XYZ &&
-				last_vert != m_vertex.buff[prev_tri[0]].XYZ && last_vert != m_vertex.buff[prev_tri[1]].XYZ && last_vert != m_vertex.buff[prev_tri[2]].XYZ)
+			if (!(new_verts[0] != m_vertex.buff[prev_tri[0]].XYZ && new_verts[0] != m_vertex.buff[prev_tri[1]].XYZ && new_verts[0] != m_vertex.buff[prev_tri[2]].XYZ))
+				match_vert_count++;
+			if (!(new_verts[1] != m_vertex.buff[prev_tri[0]].XYZ && new_verts[1] != m_vertex.buff[prev_tri[1]].XYZ && new_verts[1] != m_vertex.buff[prev_tri[2]].XYZ))
+				match_vert_count++;
+			if (!(new_verts[2] != m_vertex.buff[prev_tri[0]].XYZ && new_verts[2] != m_vertex.buff[prev_tri[1]].XYZ && new_verts[2] != m_vertex.buff[prev_tri[2]].XYZ))
+				match_vert_count++;
+
+			if (match_vert_count != 2)
 				return false;
 		}
 		// Degenerate triangles should've been culled already, so we can check indices.

pcsx2/GS/Renderers/HW/GSRendererHW.cpp

@@ -2482,7 +2482,7 @@ void GSRendererHW::Draw()
 	//                                |       0.5,2.25 |        1-1 |    1 |
 	//                                |        0.5,2.5 |        1-2 |    2 |
 	//                                --------------------------------------
-	m_r = GSVector4i(m_vt.m_min.p.upld(m_vt.m_max.p) + GSVector4::cxpr(0.5f));
+	m_r = GSVector4i((m_vt.m_min.p.upld(m_vt.m_max.p) + GSVector4::cxpr(0.4f)).round<Round_NearestInt>());
 	m_r = m_r.blend8(m_r + GSVector4i::cxpr(0, 0, 1, 1), (m_r.xyxy() == m_r.zwzw()));
 	m_r_no_scissor = m_r;
 	m_r = m_r.rintersect(context->scissor.in);
@@ -6492,8 +6492,9 @@ __ri void GSRendererHW::HandleTextureHazards(const GSTextureCache::Target* rt, c
 		// 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())
 		{
-			const GSVector4 dst_rect = GSVector4(GSVector4i::loadh(src_unscaled_size));
-			g_gs_device->StretchRect(src_target->m_texture, GSVector4::cxpr(0.0f, 0.0f, 1.0f, 1.0f), src_copy.get(), dst_rect,
+			GSVector4 src_rect = GSVector4(tmm.coverage) / GSVector4(GSVector4i::loadh(src_unscaled_size).zwzw());
+			const GSVector4 dst_rect = GSVector4(tmm.coverage);
+			g_gs_device->StretchRect(src_target->m_texture, src_rect, src_copy.get(), dst_rect,
 				src_target->m_texture->IsDepthStencil() ? ShaderConvert::DEPTH_COPY : ShaderConvert::COPY, false);
 		}
 		else

pcsx2/GS/Renderers/HW/GSTextureCache.cpp

@@ -1530,7 +1530,8 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 						GL_CACHE("TC: Attempt to repopulate RGB for target [%x] on source lookup", t->m_TEX0.TBP0);
 						for (Target* dst_match : m_dst[DepthStencil])
 						{
-							if (dst_match->m_TEX0.TBP0 != t->m_TEX0.TBP0 || !dst_match->m_valid_rgb)
+							// Be careful of dirty overlap on the targets, we don't really want dirty data.
+							if (dst_match->m_TEX0.TBP0 != t->m_TEX0.TBP0 || !dst_match->m_valid_rgb ||(!dst_match->m_dirty.empty() && !dst_match->m_dirty.GetTotalRect(dst_match->m_TEX0, dst_match->m_unscaled_size).rintersect(block_boundary_rect).rempty()))
 								continue;
 
 							if (!CopyRGBFromDepthToColor(t, dst_match))
@@ -2147,7 +2148,23 @@ GSTextureCache::Target* GSTextureCache::LookupTarget(GIFRegTEX0 TEX0, const GSVe
 				// 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
-				if (can_use && ((!is_shuffle && t->m_dirty.size() >= 1) || (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)
+				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;
 				}