rsx: Fix surface cache hit tests

- Avoid silly broken tests due to queue_tag being called before pitch is initialized.
- Return actual memory range covered and exclude trailing padding.
- Coordinates in src are to be calculated with src_pitch, not required_pitch.

rpcs3/Emu/RSX/Common/surface_store.h

@@ -753,36 +753,35 @@ namespace rsx
 		{
 			std::vector<surface_overlap_info> result;
 			std::vector<std::pair<u32, bool>> dirty;
-			const u32 limit = texaddr + (required_pitch * required_height);
+
+			const auto surface_internal_pitch = (required_width * required_bpp);
+
+			verify(HERE), surface_internal_pitch <= required_pitch;
+			const auto test_range = utils::address_range::start_length(texaddr, (required_pitch * required_height) - (required_pitch - surface_internal_pitch));
 
 			auto process_list_function = [&](std::unordered_map<u32, surface_storage_type>& data, bool is_depth)
 			{
 				for (auto &tex_info : data)
 				{
-					const auto this_address = tex_info.first;
+					const auto range = tex_info.second->get_memory_range();
-					if (this_address >= limit)
+					if (!range.overlaps(test_range))
 						continue;
 
 					auto surface = tex_info.second.get();
-					const auto pitch = surface->get_rsx_pitch();
 					if (!rsx::pitch_compatible(surface, required_pitch, required_height))
 						continue;
 
-					const auto texture_size = pitch * surface->get_surface_height(rsx::surface_metrics::samples);
-					if ((this_address + texture_size) <= texaddr)
-						continue;
-
 					surface_overlap_info info;
 					info.surface = surface;
-					info.base_address = this_address;
+					info.base_address = range.start;
 					info.is_depth = is_depth;
 
 					const auto normalized_surface_width = surface->get_surface_width(rsx::surface_metrics::bytes) / required_bpp;
 					const auto normalized_surface_height = surface->get_surface_height(rsx::surface_metrics::samples);
 
-					if (LIKELY(this_address >= texaddr))
+					if (LIKELY(range.start >= texaddr))
 					{
-						const auto offset = this_address - texaddr;
+						const auto offset = range.start - texaddr;
 						info.dst_y = (offset / required_pitch);
 						info.dst_x = (offset % required_pitch) / required_bpp;
 
@@ -799,9 +798,10 @@ namespace rsx
 					}
 					else
 					{
-						const auto offset = texaddr - this_address;
+						const auto pitch = surface->get_rsx_pitch();
-						info.src_y = (offset / required_pitch);
+						const auto offset = texaddr - range.start;
-						info.src_x = (offset % required_pitch) / required_bpp;
+						info.src_y = (offset / pitch);
+						info.src_x = (offset % pitch) / required_bpp;
 
 						if (UNLIKELY(info.src_x >= normalized_surface_width || info.src_y >= normalized_surface_height))
 						{
@@ -819,7 +819,7 @@ namespace rsx
 					// Delay this as much as possible to avoid side-effects of spamming barrier
 					if (surface->memory_barrier(cmd, access); !surface->test())
 					{
-						dirty.emplace_back(this_address, is_depth);
+						dirty.emplace_back(range.start, is_depth);
 						continue;
 					}
 
@@ -838,14 +838,12 @@ namespace rsx
 
 			// Range test helper to quickly discard blocks
 			// Fortunately, render targets tend to be clustered anyway
-			rsx::address_range test = rsx::address_range::start_end(texaddr, limit-1);
+			if (test_range.overlaps(m_render_targets_memory_range))
-
-			if (test.overlaps(m_render_targets_memory_range))
 			{
 				process_list_function(m_render_targets_storage, false);
 			}
 
-			if (test.overlaps(m_depth_stencil_memory_range))
+			if (test_range.overlaps(m_depth_stencil_memory_range))
 			{
 				process_list_function(m_depth_stencil_storage, true);
 			}

rpcs3/Emu/RSX/Common/surface_utils.h

@@ -357,6 +357,8 @@ namespace rsx
 #else
 		void queue_tag(u32 address)
 		{
+			verify(HERE), native_pitch, rsx_pitch;
+
 			base_addr = address;
 
 			const u32 size_x = (native_pitch > 8)? (native_pitch - 8) : 0u;
@@ -569,7 +571,8 @@ namespace rsx
 		rsx::address_range get_memory_range() const
 		{
 			const u32 internal_height = get_surface_height(rsx::surface_metrics::samples);
-			return rsx::address_range::start_length(base_addr, internal_height * get_rsx_pitch());
+			const u32 excess = (rsx_pitch - native_pitch);
+			return rsx::address_range::start_length(base_addr, internal_height * rsx_pitch - excess);
 		}
 
 		template <typename T>

rpcs3/Emu/RSX/GL/GLRenderTargets.h

@@ -212,14 +212,11 @@ struct gl_render_target_traits
 
 			sink->set_spp(ref->get_spp());
 			sink->set_native_pitch(prev.width * ref->get_bpp() * ref->samples_x);
+			sink->set_rsx_pitch(ref->get_rsx_pitch());
 			sink->set_surface_dimensions(prev.width, prev.height, ref->get_rsx_pitch());
 			sink->set_native_component_layout(ref->get_native_component_layout());
 			sink->queue_tag(address);
 		}
-		else
-		{
-			sink->set_rsx_pitch(ref->get_rsx_pitch());
-		}
 
 		prev.target = sink.get();
 
@@ -236,6 +233,7 @@ struct gl_render_target_traits
 			}
 		}
 
+		sink->set_rsx_pitch(ref->get_rsx_pitch());
 		sink->set_old_contents_region(prev, false);
 		sink->last_use_tag = ref->last_use_tag;
 	}

rpcs3/Emu/RSX/VK/VKRenderTargets.h

@@ -711,6 +711,7 @@ namespace rsx
 				sink->sample_layout = ref->sample_layout;
 				sink->stencil_init_flags = ref->stencil_init_flags;
 				sink->native_pitch = u16(prev.width * ref->get_bpp() * ref->samples_x);
+				sink->rsx_pitch = ref->get_rsx_pitch();
 				sink->surface_width = prev.width;
 				sink->surface_height = prev.height;
 				sink->queue_tag(address);
@@ -724,7 +725,6 @@ namespace rsx
 
 			prev.target = sink.get();
 
-			sink->rsx_pitch = ref->get_rsx_pitch();
 			if (!sink->old_contents.empty())
 			{
 				// Deal with this, likely only needs to clear
@@ -738,6 +738,7 @@ namespace rsx
 				}
 			}
 
+			sink->rsx_pitch = ref->get_rsx_pitch();
 			sink->set_old_contents_region(prev, false);
 			sink->last_use_tag = ref->last_use_tag;
 		}