rsx: Fix blit transfers when texel sizes mismatch

- Also refactors some bpp handling code
- Simplify texture intersection test to use a normalized/uniform coordinate space
- Fix broken bounds checking as well

rpcs3/Emu/RSX/Common/surface_store.h

@@ -38,6 +38,16 @@ namespace rsx
 		u16 dst_y = 0;
 		u16 width = 0;
 		u16 height = 0;
+
+		areai get_src_area() const
+		{
+			return coordi{ {src_x, src_y}, {width, height} };
+		}
+
+		areai get_dst_area() const
+		{
+			return coordi{ {dst_x, dst_y}, {width, height} };
+		}
 	};
 
 	struct surface_format_info
@@ -89,6 +99,11 @@ namespace rsx
 		virtual u16 get_native_pitch() const = 0;
 		virtual bool is_depth_surface() const = 0;
 
+		u8 get_bpp() const
+		{
+			return u8(get_native_pitch() / get_surface_width());
+		}
+
 		void save_aa_mode()
 		{
 			read_aa_mode = write_aa_mode;
@@ -890,7 +905,7 @@ namespace rsx
 		}
 
 		template <typename commandbuffer_type>
-		std::vector<surface_overlap_info> get_merged_texture_memory_region(commandbuffer_type& cmd, u32 texaddr, u32 required_width, u32 required_height, u32 required_pitch)
+		std::vector<surface_overlap_info> get_merged_texture_memory_region(commandbuffer_type& cmd, u32 texaddr, u32 required_width, u32 required_height, u32 required_pitch, u8 required_bpp)
 		{
 			std::vector<surface_overlap_info> result;
 			std::vector<std::pair<u32, bool>> dirty;
@@ -930,16 +945,26 @@ namespace rsx
 					surface_format_info surface_info{};
 					Traits::get_surface_info(surface, &surface_info);
 
-					if (this_address < texaddr)
+					const auto normalized_surface_width = (surface_info.surface_width * scale_x * surface_info.bpp) / required_bpp;
+					const auto normalized_surface_height = surface_info.surface_height * scale_y;
+
+					if (LIKELY(this_address >= texaddr))
+					{
+						const auto offset = this_address - texaddr;
+						info.src_x = 0;
+						info.src_y = 0;
+						info.dst_y = (offset / required_pitch);
+						info.dst_x = (offset % required_pitch) / required_bpp;
+						info.width = std::min<u32>(normalized_surface_width, required_width - info.dst_x);
+						info.height = std::min<u32>(normalized_surface_height, required_height - info.dst_y);
+					}
+					else
 					{
-						const auto int_required_width = required_width / scale_x;
-						const auto int_required_height = required_height / scale_y;
-
 						const auto offset = texaddr - this_address;
-						info.src_y = (offset / required_pitch) / scale_y;
+						info.src_y = (offset / required_pitch);
-						info.src_x = (offset % required_pitch) / surface_info.bpp / scale_x;
+						info.src_x = (offset % required_pitch) / required_bpp;
 
-						if (UNLIKELY(info.src_x >= surface_info.surface_width || info.src_y >= surface_info.surface_height))
+						if (UNLIKELY(info.src_x >= normalized_surface_width || info.src_y >= normalized_surface_height))
 						{
 							// Region lies outside the actual texture area, but inside the 'tile'
 							// In this case, a small region lies to the top-left corner, partially occupying the  target
@@ -948,31 +973,26 @@ namespace rsx
 
 						info.dst_x = 0;
 						info.dst_y = 0;
-						info.width = std::min<u32>(int_required_width, surface_info.surface_width - info.src_x);
+						info.width = std::min<u32>(required_width, normalized_surface_width - info.src_x);
-						info.height = std::min<u32>(int_required_height, surface_info.surface_height - info.src_y);
+						info.height = std::min<u32>(required_height, normalized_surface_height - info.src_y);
-						info.is_clipped = (info.width < int_required_width || info.height < int_required_height);
 					}
-					else
+
+					info.is_clipped = (info.width < required_width || info.height < required_height);
+
+					if (UNLIKELY(surface_info.bpp != required_bpp))
 					{
-						const auto int_surface_width = surface_info.surface_width * scale_x;
+						// Width is calculated in the coordinate-space of the requester; normalize
-						const auto int_surface_height = surface_info.surface_height * scale_y;
+						info.src_x = (info.src_x * required_bpp) / surface_info.bpp;
+						info.width = (info.width * required_bpp) / surface_info.bpp;
+					}
 
-						const auto offset = this_address - texaddr;
+					if (UNLIKELY(scale_x > 1))
-						info.dst_y = (offset / required_pitch);
+					{
-						info.dst_x = (offset % required_pitch) / surface_info.bpp;
+						info.src_x /= scale_x;
-
+						info.dst_x /= scale_x;
-						if (UNLIKELY(info.dst_x >= int_surface_width || info.dst_y >= int_surface_height))
-						{
-							// False positive
-							continue;
-						}
-
-						info.src_x = 0;
-						info.src_y = 0;
-						info.width = std::min<u32>(int_surface_width, required_width - info.dst_x);
-						info.height = std::min<u32>(int_surface_height, required_height - info.dst_y);
-						info.is_clipped = (info.width < required_width || info.height < required_height);
 						info.width /= scale_x;
+						info.src_y /= scale_y;
+						info.dst_y /= scale_y;
 						info.height /= scale_y;
 					}
 

rpcs3/Emu/RSX/Common/texture_cache.h

@@ -2124,7 +2124,8 @@ namespace rsx
 					break;
 				}
 
-				const auto overlapping_fbos = m_rtts.get_merged_texture_memory_region(cmd, texaddr, tex_width, required_surface_height, tex_pitch);
+				const auto bpp = get_format_block_size_in_bytes(format);
+				const auto overlapping_fbos = m_rtts.get_merged_texture_memory_region(cmd, texaddr, tex_width, required_surface_height, tex_pitch, bpp);
 
 				if (!overlapping_fbos.empty() || !overlapping_locals.empty())
 				{
@@ -2266,9 +2267,9 @@ namespace rsx
 				src_address += (src.width - src_w) * src_bpp;
 			}
 
-			auto rtt_lookup = [&m_rtts, &cmd](u32 address, u32 width, u32 height, u32 pitch, bool allow_clipped) -> typename surface_store_type::surface_overlap_info
+			auto rtt_lookup = [&m_rtts, &cmd](u32 address, u32 width, u32 height, u32 pitch, u32 bpp, bool allow_clipped) -> typename surface_store_type::surface_overlap_info
 			{
-				const auto list = m_rtts.get_merged_texture_memory_region(cmd, address, width, height, pitch);
+				const auto list = m_rtts.get_merged_texture_memory_region(cmd, address, width, height, pitch, bpp);
 				if (list.empty() || (list.back().is_clipped && !allow_clipped))
 				{
 					return {};
@@ -2278,11 +2279,11 @@ namespace rsx
 			};
 
 			// Check if src/dst are parts of render targets
-			auto dst_subres = rtt_lookup(dst_address, dst_w, dst_h, dst.pitch, false);
+			auto dst_subres = rtt_lookup(dst_address, dst_w, dst_h, dst.pitch, dst_bpp, false);
 			dst_is_render_target = dst_subres.surface != nullptr;
 
 			// TODO: Handle cases where src or dst can be a depth texture while the other is a color texture - requires a render pass to emulate
-			auto src_subres = rtt_lookup(src_address, src_w, src_h, src.pitch, true);
+			auto src_subres = rtt_lookup(src_address, src_w, src_h, src.pitch, src_bpp, true);
 			src_is_render_target = src_subres.surface != nullptr;
 
 			// Always use GPU blit if src or dst is in the surface store
@@ -2319,7 +2320,7 @@ namespace rsx
 				src_subres.surface->read_barrier(cmd);
 
 				const auto surf = src_subres.surface;
-				auto bpp = surf->get_native_pitch() / surf->get_surface_width();
+				const auto bpp = surf->get_bpp();
 				if (bpp != src_bpp)
 				{
 					//Enable type scaling in src
@@ -2327,14 +2328,6 @@ namespace rsx
 					typeless_info.src_is_depth = src_subres.is_depth;
 					typeless_info.src_scaling_hint = (f32)bpp / src_bpp;
 					typeless_info.src_gcm_format = src_is_argb8 ? CELL_GCM_TEXTURE_A8R8G8B8 : CELL_GCM_TEXTURE_R5G6B5;
-
-					src_w = (u16)(src_w / typeless_info.src_scaling_hint);
-					if (!src_subres.is_clipped)
-						src_subres.width = (u16)(src_subres.width / typeless_info.src_scaling_hint);
-					else
-						src_subres = rtt_lookup(src_address, src_w, src_h, src.pitch, true);
-
-					verify(HERE), src_subres.surface != nullptr;
 				}
 			}
 
@@ -2343,7 +2336,7 @@ namespace rsx
 				// Full barrier is required in case of partial transfers
 				dst_subres.surface->read_barrier(cmd);
 
-				auto bpp = dst_subres.surface->get_native_pitch() / dst_subres.surface->get_surface_width();
+				auto bpp = dst_subres.surface->get_bpp();
 				if (bpp != dst_bpp)
 				{
 					//Enable type scaling in dst
@@ -2351,14 +2344,6 @@ namespace rsx
 					typeless_info.dst_is_depth = dst_subres.is_depth;
 					typeless_info.dst_scaling_hint = (f32)bpp / dst_bpp;
 					typeless_info.dst_gcm_format = dst_is_argb8 ? CELL_GCM_TEXTURE_A8R8G8B8 : CELL_GCM_TEXTURE_R5G6B5;
-
-					dst_w = (u16)(dst_w / typeless_info.dst_scaling_hint);
-					if (!dst_subres.is_clipped)
-						dst_subres.width = (u16)(dst_subres.width / typeless_info.dst_scaling_hint);
-					else
-						dst_subres = rtt_lookup(dst_address, dst_w, dst_h, dst.pitch, false);
-
-					verify(HERE), dst_subres.surface != nullptr;
 				}
 			}
 
@@ -2379,7 +2364,7 @@ namespace rsx
 				{
 					// Optimizations table based on common width/height pairings. If we guess wrong, the upload resolver will fix it anyway
 					// TODO: Add more entries based on empirical data
-					if (LIKELY(dst.width == 1280))
+					if (LIKELY(dst_dimensions.width == 1280))
 					{
 						dst_dimensions.height = std::max<s32>(dst.height, 720);
 					}
@@ -2450,18 +2435,7 @@ namespace rsx
 			else
 			{
 				// Destination dimensions are relaxed (true)
-				dst_area.x1 = dst_subres.src_x;
+				dst_area = dst_subres.get_src_area();
-				dst_area.y1 = dst_subres.src_y;
-				dst_area.x2 += dst_subres.src_x;
-				dst_area.y2 += dst_subres.src_y;
-
-				f32 scale_x = get_internal_scaling_x(dst_subres.surface);
-				f32 scale_y = get_internal_scaling_y(dst_subres.surface);
-
-				dst_area.x1 = s32(scale_x * dst_area.x1);
-				dst_area.x2 = s32(scale_x * dst_area.x2);
-				dst_area.y1 = s32(scale_y * dst_area.y1);
-				dst_area.y2 = s32(scale_y * dst_area.y2);
 
 				dest_texture = dst_subres.surface->get_surface();
 				typeless_info.dst_context = texture_upload_context::framebuffer_storage;
@@ -2585,27 +2559,7 @@ namespace rsx
 			}
 			else
 			{
-				if (LIKELY(!dst_is_render_target))
+				src_area = src_subres.get_src_area();
-				{
-					u16 src_subres_w = src_subres.width;
-					u16 src_subres_h = src_subres.height;
-					get_rsx_dimensions(src_subres_w, src_subres_h, src_subres.surface);
-
-					const int dst_width = (int)(src_subres_w * scale_x * typeless_info.src_scaling_hint);
-					const int dst_height = (int)(src_subres_h * scale_y);
-
-					dst_area.x2 = dst_area.x1 + dst_width;
-					dst_area.y2 = dst_area.y1 + dst_height;
-				}
-
-				src_area.x2 = src_subres.width;
-				src_area.y2 = src_subres.height;
-
-				src_area.x1 = src_subres.src_x;
-				src_area.y1 = src_subres.src_y;
-				src_area.x2 += src_subres.src_x;
-				src_area.y2 += src_subres.src_y;
-
 				vram_texture = src_subres.surface->get_surface();
 				typeless_info.src_context = texture_upload_context::framebuffer_storage;
 			}

rpcs3/Emu/RSX/GL/GLGSRender.cpp

@@ -1645,7 +1645,7 @@ void GLGSRender::flip(int buffer)
 			else
 			{
 				gl::command_context cmd = { gl_state };
-				const auto overlap_info = m_rtts.get_merged_texture_memory_region(cmd, absolute_address, buffer_width, buffer_height, buffer_pitch);
+				const auto overlap_info = m_rtts.get_merged_texture_memory_region(cmd, absolute_address, buffer_width, buffer_height, buffer_pitch, render_target_texture->get_bpp());
 
 				if (!overlap_info.empty() && overlap_info.back().surface == render_target_texture)
 				{

rpcs3/Emu/RSX/GL/GLRenderTargets.cpp

@@ -613,8 +613,8 @@ void gl::render_target::memory_barrier(gl::command_context& cmd, bool force_init
 		return;
 	}
 
-	auto src_bpp = src_texture->get_native_pitch() / src_texture->get_surface_width();
+	const auto src_bpp = src_texture->get_bpp();
-	auto dst_bpp = get_native_pitch() / get_surface_width();
+	const auto dst_bpp = get_bpp();
 	rsx::typeless_xfer typeless_info{};
 
 	const bool dst_is_depth = is_depth(get_internal_format());

rpcs3/Emu/RSX/GL/GLRenderTargets.h

@@ -217,7 +217,7 @@ struct gl_render_target_traits
 		info->native_pitch = surface->get_native_pitch();
 		info->surface_width = surface->get_surface_width();
 		info->surface_height = surface->get_surface_height();
-		info->bpp = static_cast<u8>(info->native_pitch / info->surface_width);
+		info->bpp = surface->get_bpp();
 	}
 
 	static void prepare_rtt_for_drawing(void *, gl::render_target *rtt) { rtt->reset_refs(); }

rpcs3/Emu/RSX/VK/VKGSRender.cpp

@@ -3288,7 +3288,7 @@ void VKGSRender::flip(int buffer)
 			}
 			else
 			{
-				const auto overlap_info = m_rtts.get_merged_texture_memory_region(*m_current_command_buffer, absolute_address, buffer_width, buffer_height, buffer_pitch);
+				const auto overlap_info = m_rtts.get_merged_texture_memory_region(*m_current_command_buffer, absolute_address, buffer_width, buffer_height, buffer_pitch, render_target_texture->get_bpp());
 				if (!overlap_info.empty() && overlap_info.back().surface == render_target_texture)
 				{
 					// Confirmed to be the newest data source in that range

rpcs3/Emu/RSX/VK/VKRenderTargets.h

@@ -106,8 +106,8 @@ namespace vk
 				return;
 			}
 
-			auto src_bpp = src_texture->get_native_pitch() / src_texture->get_surface_width();
+			const auto src_bpp = src_texture->get_bpp();
-			auto dst_bpp = get_native_pitch() / get_surface_width();
+			const auto dst_bpp = get_bpp();
 			rsx::typeless_xfer typeless_info{};
 
 			const auto region = rsx::get_transferable_region(this);
@@ -259,7 +259,7 @@ namespace rsx
 			info->native_pitch = surface->native_pitch;
 			info->surface_width = surface->get_surface_width();
 			info->surface_height = surface->get_surface_height();
-			info->bpp = static_cast<u8>(info->native_pitch / info->surface_width);
+			info->bpp = surface->get_bpp();
 		}
 
 		static void prepare_rtt_for_drawing(vk::command_buffer* pcmd, vk::render_target *surface)