rsx: fbo fixes 2.5

- Implement flush-always behaviour to partially fix readback from a currently bound fbo
  - Without this, only the first read is correct, as more draws are added the results become 'wrong'
  - Fixes WCB and cpublit behviour
- Synchronize blit_dst surfaces to avoid data loss when gpu texture scaling is used
  - Its still faster in such cases to disable gpu texture scaling but some types cannot be disabled without force cpu blit (e.g framebuffer transfers)
- Memory management tuning
  - rsx: on-demand texture cache rescanning for unprotected sections
  - rsx: Only framebuffer resources are upscaled
  - Do not resize regular blit engine resources
  - Lazy initialize readback buffer when using opengl
  -- These measures should help minimize vram usage

rpcs3/Emu/RSX/Common/FragmentProgramDecompiler.cpp

@@ -806,9 +806,17 @@ std::string FragmentProgramDecompiler::Decompile()
 				if (m_loop_count) AddFlowOp("break");
 				else LOG_ERROR(RSX, "BRK opcode found outside of a loop");
 				break;
-			case RSX_FP_OPCODE_CAL: LOG_ERROR(RSX, "Unimplemented SIP instruction: CAL"); break;
-			case RSX_FP_OPCODE_FENCT: forced_unit = FORCE_SCT; break;
-			case RSX_FP_OPCODE_FENCB: forced_unit = FORCE_SCB; break;
+			case RSX_FP_OPCODE_CAL:
+				LOG_ERROR(RSX, "Unimplemented SIP instruction: CAL");
+				break;
+			case RSX_FP_OPCODE_FENCT:
+				AddCode("//FENCT");
+				forced_unit = FORCE_SCT;
+				break;
+			case RSX_FP_OPCODE_FENCB:
+				AddCode("//FENCB");
+				forced_unit = FORCE_SCB;
+				break;
 			case RSX_FP_OPCODE_IFE:
 				AddCode("if($cond)");
 				if (src2.end_offset != src1.else_offset)
@@ -849,7 +857,9 @@ std::string FragmentProgramDecompiler::Decompile()
 					m_code_level++;
 				}
 				break;
-			case RSX_FP_OPCODE_RET: AddFlowOp("return"); break;
+			case RSX_FP_OPCODE_RET:
+				AddFlowOp("return");
+				break;
 
 			default:
 				return false;

rpcs3/Emu/RSX/Common/texture_cache.h

@@ -23,6 +23,12 @@ namespace rsx
 		status_ready = 1
 	};
 
+	enum memory_read_flags
+	{
+		flush_always = 0,
+		flush_once = 1
+	};
+
 	struct cached_texture_section : public rsx::buffered_section
 	{
 		u16 width;
@@ -37,6 +43,7 @@ namespace rsx
 
 		u64 cache_tag = 0;
 
+		memory_read_flags readback_behaviour = memory_read_flags::flush_once;
 		rsx::texture_create_flags view_flags = rsx::texture_create_flags::default_component_order;
 		rsx::texture_upload_context context = rsx::texture_upload_context::shader_read;
 		rsx::texture_dimension_extended image_type = rsx::texture_dimension_extended::texture_dimension_2d;
@@ -97,6 +104,11 @@ namespace rsx
 			gcm_format = format;
 		}
 
+		void set_memory_read_flags(memory_read_flags flags)
+		{
+			readback_behaviour = flags;
+		}
+
 		u16 get_width() const
 		{
 			return width;
@@ -127,6 +139,11 @@ namespace rsx
 			return gcm_format;
 		}
 
+		memory_read_flags get_memory_read_flags() const
+		{
+			return readback_behaviour;
+		}
+
 		rsx::texture_sampler_status get_sampler_status() const
 		{
 			return sampler_status;
@@ -159,6 +176,12 @@ namespace rsx
 				valid_count++;
 			}
 
+			void notify()
+			{
+				verify(HERE), valid_count >= 0;
+				valid_count++;
+			}
+
 			void add(section_storage_type& section, u32 addr, u32 data_size)
 			{
 				data.push_back(std::move(section));
@@ -277,6 +300,10 @@ namespace rsx
 		//Set when a shader read-only texture data suddenly becomes contested, usually by fbo memory
 		bool read_only_tex_invalidate = false;
 
+		//Store of all objects in a flush_always state. A lazy readback is attempted every draw call
+		std::unordered_map<u32, u32> m_flush_always_cache;
+		u64 m_flush_always_update_timestamp = 0;
+
 		//Memory usage
 		const s32 m_max_zombie_objects = 64; //Limit on how many texture objects to keep around for reuse after they are invalidated
 		std::atomic<s32> m_unreleased_texture_objects = { 0 }; //Number of invalidated objects not yet freed from memory
@@ -697,8 +724,13 @@ namespace rsx
 						{
 							if (!confirm_dimensions || tex.matches(rsx_address, width, height, depth, mipmaps))
 							{
-								if (!tex.is_locked() && tex.get_context() == texture_upload_context::framebuffer_storage)
-									range_data.notify(rsx_address, rsx_size);
+								if (!tex.is_locked())
+								{
+									//Data is valid from cache pov but region has been unlocked and flushed
+									if (tex.get_context() == texture_upload_context::framebuffer_storage ||
+										tex.get_context() == texture_upload_context::blit_engine_dst)
+										range_data.notify();
+								}
 
 								return tex;
 							}
@@ -803,6 +835,27 @@ namespace rsx
 			region.set_sampler_status(rsx::texture_sampler_status::status_uninitialized);
 			region.set_image_type(rsx::texture_dimension_extended::texture_dimension_2d);
 			update_cache_tag();
+
+			region.set_memory_read_flags(memory_read_flags::flush_always);
+			m_flush_always_cache[memory_address] = memory_size;
+		}
+
+		void set_memory_read_flags(u32 memory_address, u32 memory_size, memory_read_flags flags)
+		{
+			writer_lock lock(m_cache_mutex);
+
+			if (flags != memory_read_flags::flush_always)
+				m_flush_always_cache.erase(memory_address);
+
+			section_storage_type& region = find_cached_texture(memory_address, memory_size, false);
+
+			if (!region.exists() || region.get_context() != texture_upload_context::framebuffer_storage)
+				return;
+
+			if (flags == memory_read_flags::flush_always)
+				m_flush_always_cache[memory_address] = memory_size;
+
+			region.set_memory_read_flags(flags);
 		}
 
 		template <typename ...Args>
@@ -1816,6 +1869,12 @@ namespace rsx
 
 			if (cached_dest)
 			{
+				if (!cached_dest->is_locked())
+				{
+					cached_dest->reprotect(utils::protection::no);
+					m_cache[get_block_address(cached_dest->get_section_base())].notify();
+				}
+
 				//Prep surface
 				auto channel_order = src_is_render_target ? rsx::texture_create_flags::native_component_order :
 					dst_is_argb8 ? rsx::texture_create_flags::default_component_order :
@@ -1869,6 +1928,30 @@ namespace rsx
 			return true;
 		}
 
+		void do_update()
+		{
+			if (m_flush_always_cache.size())
+			{
+				if (m_cache_update_tag.load(std::memory_order_consume) != m_flush_always_update_timestamp)
+				{
+					writer_lock lock(m_cache_mutex);
+
+					for (const auto &It : m_flush_always_cache)
+					{
+						auto& section = find_cached_texture(It.first, It.second);
+						if (section.get_protection() != utils::protection::no)
+						{
+							auto &range = m_cache[get_block_address(It.first)];
+							section.reprotect(utils::protection::no);
+							range.notify();
+						}
+					}
+
+					m_flush_always_update_timestamp = m_cache_update_tag.load(std::memory_order_consume);
+				}
+			}
+		}
+
 		virtual const u32 get_unreleased_textures_count() const
 		{
 			return m_unreleased_texture_objects;

rpcs3/Emu/RSX/GL/GLGSRender.cpp

@@ -1451,10 +1451,10 @@ void GLGSRender::do_local_task(bool /*idle*/)
 {
 	m_frame->clear_wm_events();
 
-	std::lock_guard<std::mutex> lock(queue_guard);
-
 	if (!work_queue.empty())
 	{
+		std::lock_guard<std::mutex> lock(queue_guard);
+
 		work_queue.remove_if([](work_item &q) { return q.received; });
 
 		for (work_item& q : work_queue)
@@ -1470,6 +1470,12 @@ void GLGSRender::do_local_task(bool /*idle*/)
 			q.cv.notify_one();
 		}
 	}
+	else if (!in_begin_end)
+	{
+		//This will re-engage locks and break the texture cache if another thread is waiting in access violation handler!
+		//Only call when there are no waiters
+		m_gl_texture_cache.do_update();
+	}
 
 	if (m_overlay_cleanup_requests.size())
 	{

rpcs3/Emu/RSX/GL/GLHelpers.cpp

@@ -35,8 +35,10 @@ namespace gl
 		switch (type)
 		{
 		case GL_DEBUG_TYPE_ERROR:
+		{
 			LOG_ERROR(RSX, "%s", message);
 			return;
+		}
 		default:
 			LOG_WARNING(RSX, "%s", message);
 			return;
@@ -372,9 +374,9 @@ namespace gl
 
 		glTexParameteri((GLenum)m_parent->get_target(), GL_TEXTURE_MAX_LEVEL, m_max_level);
 
-		if (m_pixels)
+		if (m_pixels && m_generate_mipmap)
 		{
-			glTexParameteri((GLenum)m_parent->get_target(), GL_GENERATE_MIPMAP, m_generate_mipmap ? GL_TRUE : GL_FALSE);
+			glGenerateMipmap((GLenum)m_parent->get_target());
 		}
 
 		glTexParameteri((GLenum)m_parent->get_target(), GL_TEXTURE_WRAP_S, (GLint)m_wrap_s);

rpcs3/Emu/RSX/GL/GLHelpers.h

@@ -210,6 +210,7 @@ namespace gl
 	{
 		GLsync m_value = nullptr;
 		GLenum flags = GL_SYNC_FLUSH_COMMANDS_BIT;
+		bool signaled = false;
 
 	public:
 
@@ -245,11 +246,16 @@ namespace gl
 		{
 			verify(HERE), m_value != nullptr;
 
+			if (signaled)
+				return true;
+
 			if (flags)
 			{
 				GLenum err = glClientWaitSync(m_value, flags, 0);
 				flags = 0;
-				return (err == GL_ALREADY_SIGNALED || err == GL_CONDITION_SATISFIED);
+
+				if (!(err == GL_ALREADY_SIGNALED || err == GL_CONDITION_SATISFIED))
+					return false;
 			}
 			else
 			{
@@ -257,8 +263,12 @@ namespace gl
 				GLint tmp;
 
 				glGetSynciv(m_value, GL_SYNC_STATUS, 4, &tmp, &status);
-				return (status == GL_SIGNALED);
+
+				if (status != GL_SIGNALED)
+					return false;
 			}
+
+			signaled = true;
 			return true;
 		}
 
@@ -266,44 +276,49 @@ namespace gl
 		{
 			verify(HERE), m_value != nullptr;
 
-			GLenum err = GL_WAIT_FAILED;
-			bool done = false;
-
-			while (!done)
+			if (signaled == GL_FALSE)
 			{
-				if (flags)
-				{
-					err = glClientWaitSync(m_value, flags, 0);
-					flags = 0;
+				GLenum err = GL_WAIT_FAILED;
+				bool done = false;
 
-					switch (err)
+				while (!done)
+				{
+					if (flags)
 					{
-					default:
-						LOG_ERROR(RSX, "gl::fence sync returned unknown error 0x%X", err);
-					case GL_ALREADY_SIGNALED:
-					case GL_CONDITION_SATISFIED:
-						done = true;
-						break;
-					case GL_TIMEOUT_EXPIRED:
-						continue;
+						err = glClientWaitSync(m_value, flags, 0);
+						flags = 0;
+
+						switch (err)
+						{
+						default:
+							LOG_ERROR(RSX, "gl::fence sync returned unknown error 0x%X", err);
+						case GL_ALREADY_SIGNALED:
+						case GL_CONDITION_SATISFIED:
+							done = true;
+							break;
+						case GL_TIMEOUT_EXPIRED:
+							continue;
+						}
+					}
+					else
+					{
+						GLint status = GL_UNSIGNALED;
+						GLint tmp;
+
+						glGetSynciv(m_value, GL_SYNC_STATUS, 4, &tmp, &status);
+
+						if (status == GL_SIGNALED)
+							break;
 					}
 				}
-				else
-				{
-					GLint status = GL_UNSIGNALED;
-					GLint tmp;
 
-					glGetSynciv(m_value, GL_SYNC_STATUS, 4, &tmp, &status);
-
-					if (status == GL_SIGNALED)
-						break;
-				}
+				signaled = (err == GL_ALREADY_SIGNALED || err == GL_CONDITION_SATISFIED);
 			}
 
 			glDeleteSync(m_value);
 			m_value = nullptr;
 
-			return (err == GL_ALREADY_SIGNALED || err == GL_CONDITION_SATISFIED);
+			return signaled;
 		}
 	};
 

rpcs3/Emu/RSX/GL/GLRenderTargets.cpp

@@ -337,6 +337,7 @@ void GLGSRender::init_buffers(rsx::framebuffer_creation_context context, bool sk
 				old_format_found = true;
 			}
 
+			m_gl_texture_cache.set_memory_read_flags(m_surface_info[i].address, m_surface_info[i].pitch * m_surface_info[i].height, rsx::memory_read_flags::flush_once);
 			m_gl_texture_cache.flush_if_cache_miss_likely(old_format, m_surface_info[i].address, m_surface_info[i].pitch * m_surface_info[i].height);
 		}
 
@@ -361,6 +362,15 @@ void GLGSRender::init_buffers(rsx::framebuffer_creation_context context, bool sk
 
 	if (std::get<0>(m_rtts.m_bound_depth_stencil))
 	{
+		if (m_depth_surface_info.pitch && g_cfg.video.write_depth_buffer)
+		{
+			auto bpp = m_depth_surface_info.pitch / m_depth_surface_info.width;
+			auto old_format = (bpp == 2) ? gl::texture::format::depth : gl::texture::format::depth_stencil;
+
+			m_gl_texture_cache.set_memory_read_flags(m_depth_surface_info.address, m_depth_surface_info.pitch * m_depth_surface_info.height, rsx::memory_read_flags::flush_once);
+			m_gl_texture_cache.flush_if_cache_miss_likely(old_format, m_depth_surface_info.address, m_depth_surface_info.pitch * m_depth_surface_info.height);
+		}
+
 		auto ds = std::get<1>(m_rtts.m_bound_depth_stencil);
 		u8 texel_size = 2;
 

rpcs3/Emu/RSX/GL/GLTextureCache.h

@@ -90,7 +90,7 @@ namespace gl
 	{
 	private:
 		fence m_fence;
-		u32 pbo_id = 0;
+		//u32 pbo_id = 0;
 		u32 pbo_size = 0;
 
 		u32 vram_texture = 0;
@@ -173,8 +173,15 @@ namespace gl
 
 		void init_buffer()
 		{
+			const f32 resolution_scale = (context == rsx::texture_upload_context::framebuffer_storage? rsx::get_resolution_scale() : 1.f);
+			const u32 real_buffer_size = (resolution_scale <= 1.f) ? cpu_address_range : (u32)(resolution_scale * resolution_scale * cpu_address_range);
+			const u32 buffer_size = align(real_buffer_size, 4096);
+
 			if (pbo_id)
 			{
+				if (pbo_size >= buffer_size)
+					return;
+
 				glDeleteBuffers(1, &pbo_id);
 				pbo_id = 0;
 				pbo_size = 0;
@@ -182,10 +189,6 @@ namespace gl
 
 			glGenBuffers(1, &pbo_id);
 
-			const f32 resolution_scale = rsx::get_resolution_scale();
-			const u32 real_buffer_size = (resolution_scale < 1.f)? cpu_address_range: (u32)(resolution_scale * resolution_scale * cpu_address_range);
-			const u32 buffer_size = align(real_buffer_size, 4096);
-
 			glBindBuffer(GL_PIXEL_PACK_BUFFER, pbo_id);
 			glBufferStorage(GL_PIXEL_PACK_BUFFER, buffer_size, nullptr, GL_MAP_READ_BIT);
 
@@ -193,15 +196,12 @@ namespace gl
 		}
 
 	public:
-
+		u32 pbo_id = 0;
 		void reset(u32 base, u32 size, bool flushable=false)
 		{
 			rsx::protection_policy policy = g_cfg.video.strict_rendering_mode ? rsx::protection_policy::protect_policy_full_range : rsx::protection_policy::protect_policy_conservative;
 			rsx::buffered_section::reset(base, size, policy);
 
-			if (flushable)
-				init_buffer();
-
 			flushed = false;
 			copied = false;
 			is_depth = false;
@@ -252,6 +252,12 @@ namespace gl
 			real_pitch = 0;
 		}
 
+		void make_flushable()
+		{
+			//verify(HERE), pbo_id == 0;
+			init_buffer();
+		}
+
 		void set_dimensions(u32 width, u32 height, u32 /*depth*/, u32 pitch)
 		{
 			this->width = width;
@@ -300,8 +306,14 @@ namespace gl
 				return;
 			}
 
+			if (!pbo_id)
+			{
+				init_buffer();
+			}
+
 			u32 target_texture = vram_texture;
-			if (real_pitch != rsx_pitch || rsx::get_resolution_scale_percent() != 100)
+			if ((rsx::get_resolution_scale_percent() != 100 && context == rsx::texture_upload_context::framebuffer_storage) ||
+				(real_pitch != rsx_pitch))
 			{
 				u32 real_width = width;
 				u32 real_height = height;
@@ -392,7 +404,7 @@ namespace gl
 					}
 				}
 
-				if (!recovered && rsx::get_resolution_scale_percent() != 100)
+				if (!recovered && rsx::get_resolution_scale_percent() != 100 && context == rsx::texture_upload_context::framebuffer_storage)
 				{
 					LOG_ERROR(RSX, "Texture readback failed. Disable resolution scaling to get the 'Write Color Buffers' option to work properly");
 				}
@@ -449,6 +461,13 @@ namespace gl
 			//throw if map failed since we'll segfault anyway
 			verify(HERE), data != nullptr;
 
+			bool require_manual_shuffle = false;
+			if (pack_unpack_swap_bytes)
+			{
+				if (type == gl::texture::type::sbyte || type == gl::texture::type::ubyte)
+					require_manual_shuffle = true;
+			}
+
 			if (real_pitch >= rsx_pitch || scaled_texture != 0)
 			{
 				memcpy(dst, data, cpu_address_range);
@@ -461,12 +480,25 @@ namespace gl
 				rsx::scale_image_nearest(dst, const_cast<const void*>(data), width, height, rsx_pitch, real_pitch, pixel_size, samples_u, samples_v);
 			}
 
+			if (require_manual_shuffle)
+			{
+				//byte swapping does not work on byte types, use uint_8_8_8_8 for rgba8 instead to avoid penalty
+				rsx::shuffle_texel_data_wzyx<u8>(dst, rsx_pitch, width, height);
+			}
+
 			glUnmapBuffer(GL_PIXEL_PACK_BUFFER);
 			glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
 
 			return true;
 		}
 
+		void reprotect(utils::protection prot)
+		{
+			flushed = false;
+			copied = false;
+			protect(prot);
+		}
+
 		void destroy()
 		{
 			if (!locked && pbo_id == 0 && vram_texture == 0 && m_fence.is_empty())
@@ -777,13 +809,31 @@ namespace gl
 			cached.set_sampler_status(rsx::texture_sampler_status::status_uninitialized);
 			cached.set_image_type(type);
 
-			//Its not necessary to lock blit dst textures as they are just reused as necessary
-			if (context != rsx::texture_upload_context::blit_engine_dst || g_cfg.video.strict_rendering_mode)
+			if (context != rsx::texture_upload_context::blit_engine_dst)
 			{
 				cached.protect(utils::protection::ro);
-				update_cache_tag();
+			}
+			else
+			{
+				//TODO: More tests on byte order
+				//ARGB8+native+unswizzled is confirmed with Dark Souls II character preview
+				if (gcm_format == CELL_GCM_TEXTURE_A8R8G8B8)
+				{
+					bool bgra = (flags == rsx::texture_create_flags::native_component_order);
+					cached.set_format(bgra? gl::texture::format::bgra : gl::texture::format::rgba, gl::texture::type::uint_8_8_8_8, false);
+				}
+				else
+				{
+					cached.set_format(gl::texture::format::rgb, gl::texture::type::ushort_5_6_5, true);
+				}
+
+				cached.make_flushable();
+				cached.set_dimensions(width, height, depth, (rsx_size / height));
+				cached.protect(utils::protection::no);
+				no_access_range = cached.get_min_max(no_access_range);
 			}
 
+			update_cache_tag();
 			return &cached;
 		}
 

rpcs3/Emu/RSX/VK/VKGSRender.cpp

@@ -2032,6 +2032,12 @@ void VKGSRender::do_local_task(bool /*idle*/)
 		m_flush_requests.clear_pending_flag();
 		m_flush_requests.consumer_wait();
 	}
+	else if (!in_begin_end)
+	{
+		//This will re-engage locks and break the texture cache if another thread is waiting in access violation handler!
+		//Only call when there are no waiters
+		m_texture_cache.do_update();
+	}
 
 	if (m_last_flushable_cb > -1)
 	{
@@ -2633,6 +2639,7 @@ void VKGSRender::prepare_rtts(rsx::framebuffer_creation_context context)
 			if (old_format == VK_FORMAT_UNDEFINED)
 				old_format = vk::get_compatible_surface_format(m_surface_info[i].color_format).first;
 
+			m_texture_cache.set_memory_read_flags(m_surface_info[i].address, m_surface_info[i].pitch * m_surface_info[i].height, rsx::memory_read_flags::flush_once);
 			m_texture_cache.flush_if_cache_miss_likely(old_format, m_surface_info[i].address, m_surface_info[i].pitch * m_surface_info[i].height,
 				*m_current_command_buffer, m_memory_type_mapping, m_swap_chain->get_present_queue());
 		}
@@ -2643,10 +2650,21 @@ void VKGSRender::prepare_rtts(rsx::framebuffer_creation_context context)
 		m_surface_info[i].color_format = color_fmt;
 	}
 
-	m_depth_surface_info.address = m_depth_surface_info.pitch = 0;
-	m_depth_surface_info.width = clip_width;
-	m_depth_surface_info.height = clip_height;
-	m_depth_surface_info.depth_format = depth_fmt;
+	//Process depth surface as well
+	{
+		if (m_depth_surface_info.pitch && g_cfg.video.write_depth_buffer)
+		{
+			auto old_format = vk::get_compatible_depth_surface_format(m_optimal_tiling_supported_formats, m_depth_surface_info.depth_format);
+			m_texture_cache.set_memory_read_flags(m_depth_surface_info.address, m_depth_surface_info.pitch * m_depth_surface_info.height, rsx::memory_read_flags::flush_once);
+			m_texture_cache.flush_if_cache_miss_likely(old_format, m_depth_surface_info.address, m_depth_surface_info.pitch * m_depth_surface_info.height,
+				*m_current_command_buffer, m_memory_type_mapping, m_swap_chain->get_present_queue());
+		}
+
+		m_depth_surface_info.address = m_depth_surface_info.pitch = 0;
+		m_depth_surface_info.width = clip_width;
+		m_depth_surface_info.height = clip_height;
+		m_depth_surface_info.depth_format = depth_fmt;
+	}
 
 	//Bind created rtts as current fbo...
 	std::vector<u8> draw_buffers = rsx::utility::get_rtt_indexes(target);

rpcs3/Emu/RSX/VK/VKTextureCache.h

@@ -168,8 +168,8 @@ namespace vk
 				cmd.begin();
 			}
 
-			const u16 internal_width = std::min(width, rsx::apply_resolution_scale(width, true));
-			const u16 internal_height = std::min(height, rsx::apply_resolution_scale(height, true));
+			const u16 internal_width = (context != rsx::texture_upload_context::framebuffer_storage? width : std::min(width, rsx::apply_resolution_scale(width, true)));
+			const u16 internal_height = (context != rsx::texture_upload_context::framebuffer_storage? height : std::min(height, rsx::apply_resolution_scale(height, true)));
 
 			VkImageAspectFlags aspect_flag = VK_IMAGE_ASPECT_COLOR_BIT;
 			switch (vram_texture->info.format)
@@ -332,6 +332,21 @@ namespace vk
 			return result;
 		}
 
+		void set_unpack_swap_bytes(bool swap_bytes)
+		{
+			pack_unpack_swap_bytes = swap_bytes;
+		}
+
+		void reprotect(utils::protection prot)
+		{
+			//Reset properties and protect again
+			flushed = false;
+			synchronized = false;
+			sync_timestamp = 0ull;
+
+			protect(prot);
+		}
+
 		bool is_synchronized() const
 		{
 			return synchronized;
@@ -730,13 +745,20 @@ namespace vk
 			region.set_image_type(type);
 
 			//Its not necessary to lock blit dst textures as they are just reused as necessary
-			if (context != rsx::texture_upload_context::blit_engine_dst || g_cfg.video.strict_rendering_mode)
+			if (context != rsx::texture_upload_context::blit_engine_dst)
 			{
 				region.protect(utils::protection::ro);
-				update_cache_tag();
+				read_only_range = region.get_min_max(read_only_range);
+			}
+			else
+			{
+				//TODO: Confirm byte swap patterns
+				region.protect(utils::protection::no);
+				region.set_unpack_swap_bytes(true);
+				no_access_range = region.get_min_max(no_access_range);
 			}
 
-			read_only_range = region.get_min_max(read_only_range);
+			update_cache_tag();
 			return &region;
 		}
 

rpcs3/Emu/RSX/rsx_cache.h

@@ -115,6 +115,7 @@ namespace rsx
 			else
 				locked_address_range = align(base + length, 4096) - locked_address_base;
 
+			verify(HERE), locked_address_range > 0;
 			protection = utils::protection::rw;
 			guard_policy = protect_policy;
 			locked = false;
@@ -124,6 +125,7 @@ namespace rsx
 		{
 			if (prot == protection) return;
 
+			verify(HERE), locked_address_range > 0;
 			utils::memory_protect(vm::base(locked_address_base), locked_address_range, prot);
 			protection = prot;
 			locked = prot != utils::protection::rw;

rpcs3/Emu/RSX/rsx_methods.cpp

@@ -694,7 +694,8 @@ namespace rsx
 
 			if (convert_w == 0 || convert_h == 0)
 			{
-				LOG_ERROR(RSX, "NV3089_IMAGE_IN: Invalid dimensions or scaling factor. Request ignored");
+				LOG_ERROR(RSX, "NV3089_IMAGE_IN: Invalid dimensions or scaling factor. Request ignored (ds_dx=%d, dt_dy=%d)",
+					method_registers.blit_engine_ds_dx(), method_registers.blit_engine_dt_dy());
 				return;
 			}