#pragma once

#include <thread>
#include <queue>
#include <deque>
#include <variant>
#include <stack>
#include <unordered_map>

#include "GCM.h"
#include "rsx_cache.h"
#include "RSXFIFO.h"
#include "RSXOffload.h"
#include "RSXZCULL.h"
#include "rsx_utils.h"
#include "Common/bitfield.hpp"
#include "Common/profiling_timer.hpp"
#include "Common/texture_cache_types.h"
#include "Program/RSXVertexProgram.h"
#include "Program/RSXFragmentProgram.h"

#include "Utilities/Thread.h"
#include "Utilities/geometry.h"
#include "Capture/rsx_trace.h"
#include "Capture/rsx_replay.h"

#include "Emu/Cell/lv2/sys_rsx.h"
#include "Emu/IdManager.h"
#include "Emu/system_config.h"

extern atomic_t<bool> g_user_asked_for_frame_capture;
extern atomic_t<bool> g_disable_frame_limit; 
extern rsx::frame_trace_data frame_debug;
extern rsx::frame_capture_data frame_capture;

namespace rsx
{
	namespace overlays
	{
		class display_manager;
	}

	struct rsx_iomap_table
	{
		std::array<atomic_t<u32>, 4096> ea;
		std::array<atomic_t<u32>, 4096> io;
		std::array<shared_mutex, 4096> rs{};

		rsx_iomap_table() noexcept
			: ea(fill_array(-1))
			, io(fill_array(-1))
		{
		}

		// Try to get the real address given a mapped address
		// Returns -1 on failure
		u32 get_addr(u32 offs) const noexcept
		{
			return this->ea[offs >> 20] | (offs & 0xFFFFF);
		}

		template<bool IsFullLock>
		bool lock(u32 addr, u32 len, cpu_thread* self = nullptr) noexcept
		{
			if (len <= 1) return false;
			const u32 end = addr + len - 1;

			for (u32 block = (addr >> 20); block <= (end >> 20); ++block)
			{
				auto& mutex_ = rs[block];

				if constexpr (IsFullLock)
				{
					if (self) [[ likely ]]
					{
						while (!mutex_.try_lock())
						{
							self->cpu_wait({});
						}
					}
					else
					{
						mutex_.lock();
					}
				}
				else
				{
					if (!self) [[ likely ]]
					{
						mutex_.lock_shared();
					}
					else
					{
						while (!mutex_.try_lock_shared())
						{
							self->cpu_wait({});
						}
					}
				}
			}

			return true;
		}

		template<bool IsFullLock>
		void unlock(u32 addr, u32 len) noexcept
		{
			ensure(len >= 1);
			const u32 end = addr + len - 1;

			for (u32 block = (addr >> 20); block <= (end >> 20); ++block)
			{
				if constexpr (IsFullLock)
				{
					rs[block].unlock();
				}
				else
				{
					rs[block].unlock_shared();
				}
			}
		}
	};

	enum framebuffer_creation_context : u8
	{
		context_draw = 0,
		context_clear_color = 1,
		context_clear_depth = 2,
		context_clear_all = context_clear_color | context_clear_depth
	};

	enum pipeline_state : u32
	{
		fragment_program_ucode_dirty = 0x1,   // Fragment program ucode changed
		vertex_program_ucode_dirty = 0x2,     // Vertex program ucode changed
		fragment_program_state_dirty = 0x4,   // Fragment program state changed
		vertex_program_state_dirty = 0x8,     // Vertex program state changed
		fragment_state_dirty = 0x10,          // Fragment state changed (alpha test, etc)
		vertex_state_dirty = 0x20,            // Vertex state changed (scale_offset, clip planes, etc)
		transform_constants_dirty = 0x40,     // Transform constants changed
		fragment_constants_dirty = 0x80,      // Fragment constants changed
		framebuffer_reads_dirty = 0x100,      // Framebuffer contents changed
		fragment_texture_state_dirty = 0x200, // Fragment texture parameters changed
		vertex_texture_state_dirty = 0x400,   // Fragment texture parameters changed
		scissor_config_state_dirty = 0x800,   // Scissor region changed
		zclip_config_state_dirty = 0x1000,    // Viewport Z clip changed

		scissor_setup_invalid = 0x2000,       // Scissor configuration is broken
		scissor_setup_clipped = 0x4000,       // Scissor region is cropped by viewport constraint

		polygon_stipple_pattern_dirty = 0x8000,  // Rasterizer stippling pattern changed
		line_stipple_pattern_dirty = 0x10000,    // Line stippling pattern changed

		push_buffer_arrays_dirty = 0x20000,   // Push buffers have data written to them (immediate mode vertex buffers)

		fragment_program_dirty = fragment_program_ucode_dirty | fragment_program_state_dirty,
		vertex_program_dirty = vertex_program_ucode_dirty | vertex_program_state_dirty,
		invalidate_pipeline_bits = fragment_program_dirty | vertex_program_dirty,
		invalidate_zclip_bits = vertex_state_dirty | zclip_config_state_dirty,
		memory_barrier_bits = framebuffer_reads_dirty,

		all_dirty = ~0u
	};

	enum eng_interrupt_reason : u32
	{
		backend_interrupt       = 0x0001,        // Backend-related interrupt
		memory_config_interrupt = 0x0002,        // Memory configuration changed
		display_interrupt       = 0x0004,        // Display handling
		pipe_flush_interrupt    = 0x0008,        // Flush pipelines

		all_interrupt_bits = memory_config_interrupt | backend_interrupt | display_interrupt | pipe_flush_interrupt
	};

	enum FIFO_state : u8
	{
		running = 0,
		empty = 1,    // PUT == GET
		spinning = 2, // Puller continuously jumps to self addr (synchronization technique)
		nop = 3,      // Puller is processing a NOP command
		lock_wait = 4 // Puller is processing a lock acquire
	};

	enum FIFO_hint : u8
	{
		hint_conditional_render_eval = 1,
		hint_zcull_sync = 2
	};

	enum result_flags: u8
	{
		result_none = 0,
		result_error = 1,
		result_zcull_intr = 2
	};

	enum ROP_control : u32
	{
		alpha_test_enable       = (1u << 0),
		framebuffer_srgb_enable = (1u << 1),
		csaa_enable             = (1u << 4),
		msaa_mask_enable        = (1u << 5),
		msaa_config_mask        = (3u << 6),
		polygon_stipple_enable  = (1u << 9),
		alpha_func_mask         = (7u << 16)
	};

	u32 get_vertex_type_size_on_host(vertex_base_type type, u32 size);

	u32 get_address(u32 offset, u32 location, u32 size_to_check = 0,
		u32 line = __builtin_LINE(),
		u32 col = __builtin_COLUMN(),
		const char* file = __builtin_FILE(),
		const char* func = __builtin_FUNCTION());

	struct tiled_region
	{
		u32 address;
		u32 base;
		GcmTileInfo *tile;
		u8 *ptr;

		void write(const void *src, u32 width, u32 height, u32 pitch);
		void read(void *dst, u32 width, u32 height, u32 pitch);
	};

	struct vertex_array_buffer
	{
		rsx::vertex_base_type type;
		u8 attribute_size;
		u8 stride;
		std::span<const std::byte> data;
		u8 index;
		bool is_be;
	};

	struct vertex_array_register
	{
		rsx::vertex_base_type type;
		u8 attribute_size;
		std::array<u32, 4> data;
		u8 index;
	};

	struct empty_vertex_array
	{
		u8 index;
	};

	struct draw_array_command
	{
		u32 __dummy;
	};

	struct draw_indexed_array_command
	{
		std::span<const std::byte> raw_index_buffer;
	};

	struct draw_inlined_array
	{
		u32 __dummy;
		u32 __dummy2;
	};

	struct interleaved_attribute_t
	{
		u8 index;
		bool modulo;
		u16 frequency;
	};

	struct interleaved_range_info
	{
		bool interleaved = false;
		bool single_vertex = false;
		u32  base_offset = 0;
		u32  real_offset_address = 0;
		u8   memory_location = 0;
		u8   attribute_stride = 0;

		rsx::simple_array<interleaved_attribute_t> locations;

		// Check if we need to upload a full unoptimized range, i.e [0-max_index]
		std::pair<u32, u32> calculate_required_range(u32 first, u32 count) const;
	};

	enum attribute_buffer_placement : u8
	{
		none = 0,
		persistent = 1,
		transient = 2
	};

	struct vertex_input_layout
	{
		std::vector<interleaved_range_info> interleaved_blocks{};  // Interleaved blocks to be uploaded as-is
		std::vector<std::pair<u8, u32>> volatile_blocks{};         // Volatile data blocks (immediate draw vertex data for example)
		rsx::simple_array<u8> referenced_registers{};              // Volatile register data

		std::array<attribute_buffer_placement, 16> attribute_placement = fill_array(attribute_buffer_placement::none);

		vertex_input_layout() = default;

		void clear()
		{
			interleaved_blocks.clear();
			volatile_blocks.clear();
			referenced_registers.clear();
		}

		bool validate() const
		{
			// Criteria: At least one array stream has to be defined to feed vertex positions
			// This stream cannot be a const register as the vertices cannot create a zero-area primitive

			if (!interleaved_blocks.empty() && interleaved_blocks.front().attribute_stride != 0)
				return true;

			if (!volatile_blocks.empty())
				return true;

			for (u8 index = 0; index < limits::vertex_count; ++index)
			{
				switch (attribute_placement[index])
				{
				case attribute_buffer_placement::transient:
				{
					// Ignore register reference
					if (std::find(referenced_registers.begin(), referenced_registers.end(), index) != referenced_registers.end())
						continue;

					// The source is inline array or immediate draw push buffer
					return true;
				}
				case attribute_buffer_placement::persistent:
				{
					return true;
				}
				case attribute_buffer_placement::none:
				{
					continue;
				}
				default:
				{
					fmt::throw_exception("Unreachable");
				}
				}
			}

			return false;
		}

		u32 calculate_interleaved_memory_requirements(u32 first_vertex, u32 vertex_count) const
		{
			u32 mem = 0;
			for (auto &block : interleaved_blocks)
			{
				const auto range = block.calculate_required_range(first_vertex, vertex_count);
				mem += range.second * block.attribute_stride;
			}

			return mem;
		}
	};

	struct framebuffer_layout
	{
		u16 width;
		u16 height;
		std::array<u32, 4> color_addresses;
		std::array<u32, 4> color_pitch;
		std::array<u32, 4> actual_color_pitch;
		std::array<bool, 4> color_write_enabled;
		u32 zeta_address;
		u32 zeta_pitch;
		u32 actual_zeta_pitch;
		bool zeta_write_enabled;
		rsx::surface_target target;
		rsx::surface_color_format color_format;
		rsx::surface_depth_format2 depth_format;
		rsx::surface_antialiasing aa_mode;
		rsx::surface_raster_type raster_type;
		u32 aa_factors[2];
		bool ignore_change;
	};

	struct frame_statistics_t
	{
		u32 draw_calls;
		u32 submit_count;

		s64 setup_time;
		s64 vertex_upload_time;
		s64 textures_upload_time;
		s64 draw_exec_time;
		s64 flip_time;
	};

	struct display_flip_info_t
	{
		std::deque<u32> buffer_queue;
		u32 buffer;
		bool skip_frame;
		bool emu_flip;
		bool in_progress;
		frame_statistics_t stats;

		inline void push(u32 _buffer)
		{
			buffer_queue.push_back(_buffer);
		}

		inline bool pop(u32 _buffer)
		{
			if (buffer_queue.empty())
			{
				return false;
			}

			do
			{
				const auto index = buffer_queue.front();
				buffer_queue.pop_front();

				if (index == _buffer)
				{
					buffer = _buffer;
					return true;
				}
			}
			while (!buffer_queue.empty());

			// Need to observe this happening in the wild
			rsx_log.error("Display queue was discarded while not empty!");
			return false;
		}
	};

	struct backend_configuration
	{
		bool supports_multidraw;             // Draw call batching
		bool supports_hw_a2c;                // Alpha to coverage
		bool supports_hw_renormalization;    // Should be true on NV hardware which matches PS3 texture renormalization behaviour
		bool supports_hw_msaa;               // MSAA support
		bool supports_hw_a2one;              // Alpha to one
		bool supports_hw_conditional_render; // Conditional render
		bool supports_passthrough_dma;       // DMA passthrough
		bool supports_asynchronous_compute;  // Async compute
		bool supports_host_gpu_labels;       // Advanced host synchronization
	};

	struct sampled_image_descriptor_base;

	class thread : public cpu_thread
	{
		u64 timestamp_ctrl = 0;
		u64 timestamp_subvalue = 0;
		u64 m_cycles_counter = 0;

		display_flip_info_t m_queued_flip{};

		void cpu_task() override;
	protected:
		atomic_t<bool> m_rsx_thread_exiting{ true };

		std::array<push_buffer_vertex_info, 16> vertex_push_buffers;
		std::vector<u32> element_push_buffer;

		s32 m_skip_frame_ctr = 0;
		bool skip_current_frame = false;

		backend_configuration backend_config{};

		// FIFO
	public:
		std::unique_ptr<FIFO::FIFO_control> fifo_ctrl;
		std::vector<std::pair<u32, u32>> dump_callstack_list() const override;

	protected:
		FIFO::flattening_helper m_flattener;
		u32 fifo_ret_addr = RSX_CALL_STACK_EMPTY;
		u32 saved_fifo_ret = RSX_CALL_STACK_EMPTY;

		// Occlusion query
		bool zcull_surface_active = false;
		std::unique_ptr<reports::ZCULL_control> zcull_ctrl;

		// Framebuffer setup
		rsx::gcm_framebuffer_info m_surface_info[rsx::limits::color_buffers_count];
		rsx::gcm_framebuffer_info m_depth_surface_info;
		framebuffer_layout m_framebuffer_layout{};
		bool framebuffer_status_valid = false;

		// Overlays
		rsx::overlays::display_manager* m_overlay_manager = nullptr;

		// Invalidated memory range
		address_range m_invalidated_memory_range;

		// Profiler
		rsx::profiling_timer m_profiler;
		frame_statistics_t m_frame_stats;

	public:
		RsxDmaControl* ctrl = nullptr;
		u32 dma_address{0};
		rsx_iomap_table iomap_table;
		u32 restore_point = 0;
		u32 dbg_step_pc = 0;
		u32 last_known_code_start = 0;
		atomic_t<u32> external_interrupt_lock{ 0 };
		atomic_t<bool> external_interrupt_ack{ false };
		atomic_t<bool> is_inited{ false };
		bool is_fifo_idle() const;
		void flush_fifo();

		// Returns [count of found commands, PC of their start]
		std::pair<u32, u32> try_get_pc_of_x_cmds_backwards(u32 count, u32 get) const;

		void recover_fifo(u32 line = __builtin_LINE(),
			u32 col = __builtin_COLUMN(),
			const char* file = __builtin_FILE(),
			const char* func = __builtin_FUNCTION());

		static void fifo_wake_delay(u64 div = 1);
		u32 get_fifo_cmd() const;

		std::string dump_regs() const override;
		void cpu_wait(bs_t<cpu_flag> old) override;

		static constexpr u32 id_base = 0x5555'5555; // See get_current_cpu_thread()

		// Performance approximation counters
		struct
		{
			atomic_t<u64> idle_time{ 0 };  // Time spent idling in microseconds
			u64 last_update_timestamp = 0; // Timestamp of last load update
			u64 FIFO_idle_timestamp = 0;   // Timestamp of when FIFO queue becomes idle
			FIFO_state state = FIFO_state::running;
			u32 approximate_load = 0;
			u32 sampled_frames = 0;
		}
		performance_counters;

		enum class flip_request : u32
		{
			emu_requested = 1,
			native_ui = 2,

			any = emu_requested | native_ui
		};

		atomic_bitmask_t<flip_request> async_flip_requested{};
		u8 async_flip_buffer{ 0 };

		GcmTileInfo tiles[limits::tiles_count];
		GcmZcullInfo zculls[limits::zculls_count];

		void capture_frame(const std::string &name);
		const backend_configuration& get_backend_config() const { return backend_config; }

	public:
		std::shared_ptr<named_thread<class ppu_thread>> intr_thread;

		// I hate this flag, but until hle is closer to lle, its needed
		bool isHLE{ false };

		u32 flip_status;
		int debug_level;

		atomic_t<bool> requested_vsync{false};
		atomic_t<bool> enable_second_vhandler{false};

		RsxDisplayInfo display_buffers[8];
		u32 display_buffers_count{0};
		u32 current_display_buffer{0};

		shared_mutex sys_rsx_mtx;
		u32 device_addr{0};
		u32 label_addr{0};
		u32 main_mem_size{0};
		u32 local_mem_size{0};
		u32 rsx_event_port{0};
		u32 driver_info{0};

		void send_event(u64, u64, u64) const;

		bool m_rtts_dirty = true;
		std::array<bool, 16> m_textures_dirty;
		std::array<bool, 4> m_vertex_textures_dirty;
		bool m_framebuffer_state_contested = false;
		rsx::framebuffer_creation_context m_current_framebuffer_context = rsx::framebuffer_creation_context::context_draw;

		rsx::atomic_bitmask_t<rsx::eng_interrupt_reason> m_eng_interrupt_mask;
		u32 m_graphics_state = 0;
		u64 ROP_sync_timestamp = 0;

		program_hash_util::fragment_program_utils::fragment_program_metadata current_fp_metadata = {};
		program_hash_util::vertex_program_utils::vertex_program_metadata current_vp_metadata = {};

	protected:
		std::array<u32, 4> get_color_surface_addresses() const;
		u32 get_zeta_surface_address() const;

		void get_framebuffer_layout(rsx::framebuffer_creation_context context, framebuffer_layout &layout);
		bool get_scissor(areau& region, bool clip_viewport);

		/**
		 * Analyze vertex inputs and group all interleaved blocks
		 */
		void analyse_inputs_interleaved(vertex_input_layout&);

		RSXVertexProgram current_vertex_program = {};
		RSXFragmentProgram current_fragment_program = {};

		vertex_program_texture_state current_vp_texture_state = {};
		fragment_program_texture_state current_fp_texture_state = {};

		// Runs shader prefetch and resolves pipeline status flags
		void analyse_current_rsx_pipeline();

		// Prefetch and analyze the currently active fragment program ucode
		void prefetch_fragment_program();

		// Prefetch and analyze the currently active vertex program ucode
		void prefetch_vertex_program();

		void get_current_vertex_program(const std::array<std::unique_ptr<rsx::sampled_image_descriptor_base>, rsx::limits::vertex_textures_count>& sampler_descriptors);

		/**
		 * Gets current fragment program and associated fragment state
		 */
		void get_current_fragment_program(const std::array<std::unique_ptr<rsx::sampled_image_descriptor_base>, rsx::limits::fragment_textures_count>& sampler_descriptors);

	public:
		bool invalidate_fragment_program(u32 dst_dma, u32 dst_offset, u32 size);
		void on_framebuffer_options_changed(u32 opt);

	public:
		u64 target_rsx_flip_time = 0;
		u64 int_flip_index = 0;
		u64 last_guest_flip_timestamp = 0;
		u64 last_host_flip_timestamp = 0;

		vm::ptr<void(u32)> flip_handler = vm::null;
		vm::ptr<void(u32)> user_handler = vm::null;
		vm::ptr<void(u32)> vblank_handler = vm::null;
		atomic_t<u64> vblank_count{0};
		bool capture_current_frame = false;

		bool wait_for_flip_sema = false;
		u32 flip_sema_wait_val = 0;

	public:
		atomic_t<bool> sync_point_request = false;
		bool in_begin_end = false;

		struct desync_fifo_cmd_info
		{
			u32 cmd;
			u64 timestamp;
		};

		std::queue<desync_fifo_cmd_info> recovered_fifo_cmds_history;

		atomic_t<s32> async_tasks_pending{ 0 };

		bool zcull_stats_enabled = false;
		bool zcull_rendering_enabled = false;
		bool zcull_pixel_cnt_enabled = false;

		reports::conditional_render_eval cond_render_ctrl;

		virtual u64 get_cycles() = 0;
		virtual ~thread();

		static constexpr auto thread_name = "rsx::thread"sv;

	protected:
		thread();
		virtual void on_task();
		virtual void on_exit();

		/**
		 * Execute a backend local task queue
		 */
		virtual void do_local_task(FIFO_state state);

		virtual void emit_geometry(u32) {}

		void run_FIFO();

	public:
		thread(const thread&) = delete;
		thread& operator=(const thread&) = delete;

		virtual void clear_surface(u32 /*arg*/) {}
		virtual void begin();
		virtual void end();
		virtual void execute_nop_draw();

		virtual void on_init_thread() = 0;
		virtual void on_frame_end(u32 buffer, bool forced = false);
		virtual void flip(const display_flip_info_t& info) = 0;
		virtual u64 timestamp();
		virtual bool on_access_violation(u32 /*address*/, bool /*is_writing*/) { return false; }
		virtual void on_invalidate_memory_range(const address_range & /*range*/, rsx::invalidation_cause) {}
		virtual void notify_tile_unbound(u32 /*tile*/) {}

		// control
		virtual void renderctl(u32 /*request_code*/, void* /*args*/) {}

		// zcull
		void notify_zcull_info_changed();
		void clear_zcull_stats(u32 type);
		void check_zcull_status(bool framebuffer_swap);
		void get_zcull_stats(u32 type, vm::addr_t sink);
		u32  copy_zcull_stats(u32 memory_range_start, u32 memory_range, u32 destination);

		void enable_conditional_rendering(vm::addr_t ref);
		void disable_conditional_rendering();
		virtual void begin_conditional_rendering(const std::vector<reports::occlusion_query_info*>& sources);
		virtual void end_conditional_rendering();

		// sync
		void sync();
		flags32_t read_barrier(u32 memory_address, u32 memory_range, bool unconditional);
		virtual void sync_hint(FIFO_hint hint, reports::sync_hint_payload_t payload);
		virtual bool release_GCM_label(u32 /*address*/, u32 /*value*/) { return false; }

		std::span<const std::byte> get_raw_index_array(const draw_clause& draw_indexed_clause) const;

		std::variant<draw_array_command, draw_indexed_array_command, draw_inlined_array>
		get_draw_command(const rsx::rsx_state& state) const;

		/**
		* Immediate mode rendering requires a temp push buffer to hold attrib values
		* Appends a value to the push buffer (currently only supports 32-wide types)
		*/
		void append_to_push_buffer(u32 attribute, u32 size, u32 subreg_index, vertex_base_type type, u32 value);
		u32 get_push_buffer_vertex_count() const;

		void append_array_element(u32 index);
		u32 get_push_buffer_index_count() const;

	protected:

		/**
		 * Computes VRAM requirements needed to upload raw vertex streams
		 * result.first contains persistent memory requirements
		 * result.second contains volatile memory requirements
		 */
		std::pair<u32, u32> calculate_memory_requirements(const vertex_input_layout& layout, u32 first_vertex, u32 vertex_count);

		/**
		 * Generates vertex input descriptors as an array of 16x4 s32s
		 */
		void fill_vertex_layout_state(const vertex_input_layout& layout, u32 first_vertex, u32 vertex_count, s32* buffer, u32 persistent_offset = 0, u32 volatile_offset = 0);

		/**
		 * Uploads vertex data described in the layout descriptor
		 * Copies from local memory to the write-only output buffers provided in a sequential manner
		 */
		void write_vertex_data_to_memory(const vertex_input_layout& layout, u32 first_vertex, u32 vertex_count, void *persistent_data, void *volatile_data);

	private:
		shared_mutex m_mtx_task;

		void handle_emu_flip(u32 buffer);
		void handle_invalidated_memory_range();

	public:
		/**
		 * Fill buffer with 4x4 scale offset matrix.
		 * Vertex shader's position is to be multiplied by this matrix.
		 * if flip_y is set, the matrix is modified to use d3d convention.
		 */
		void fill_scale_offset_data(void *buffer, bool flip_y) const;

		/**
		 * Fill buffer with user clip information
		 */
		void fill_user_clip_data(void *buffer) const;

		/**
		* Fill buffer with vertex program constants.
		* Relocation table allows to do a partial fill with only selected registers.
		*/
		void fill_vertex_program_constants_data(void* buffer, const std::vector<u16>& reloc_table);

		/**
		 * Fill buffer with fragment rasterization state.
		 * Fills current fog values, alpha test parameters and texture scaling parameters
		 */
		void fill_fragment_state_buffer(void* buffer, const RSXFragmentProgram& fragment_program);

		/**
		 * Notify that a section of memory has been mapped
		 * If there is a notify_memory_unmapped request on this range yet to be handled,
		 * handles it immediately.
		 */
		void on_notify_memory_mapped(u32 address_base, u32 size);

		/**
		 * Notify that a section of memory has been unmapped
		 * Any data held in the defined range is discarded
		 */
		void on_notify_memory_unmapped(u32 address_base, u32 size);

		/**
		 * Notify to check internal state during semaphore wait
		 */
		virtual void on_semaphore_acquire_wait() {}

		virtual std::pair<std::string, std::string> get_programs() const { return std::make_pair("", ""); }

		virtual bool scaled_image_from_memory(blit_src_info& /*src_info*/, blit_dst_info& /*dst_info*/, bool /*interpolate*/) { return false; }

	public:
		void reset();
		void init(u32 ctrlAddress);

		// Emu App/Game flip, only immediately flips when called from rsxthread
		void request_emu_flip(u32 buffer);

		void pause();
		void unpause();
		void wait_pause();

		// Get RSX approximate load in %
		u32 get_load();

		// Get stats object
		frame_statistics_t& get_stats() { return m_frame_stats; }

		// Returns true if the current thread is the active RSX thread
		inline bool is_current_thread() const
		{
			return !!cpu_thread::get_current<rsx::thread>();
		}
	};

	inline thread* get_current_renderer()
	{
		return g_fxo->try_get<rsx::thread>();
	}

	template<bool IsFullLock = false>
	class reservation_lock
	{
		u32 addr = 0, length = 0;
		bool locked = false;

		inline void lock_range(u32 addr, u32 length)
		{
			this->addr = addr;
			this->length = length;

			auto renderer = get_current_renderer();
			cpu_thread* lock_owner = renderer->is_current_thread() ? renderer : nullptr;
			this->locked = renderer->iomap_table.lock<IsFullLock>(addr, length, lock_owner);
		}

	public:
		reservation_lock(u32 addr, u32 length)
		{
			if (g_cfg.core.rsx_accurate_res_access &&
				addr < constants::local_mem_base)
			{
				lock_range(addr, length);
			}
		}

		// Multi-range lock. If ranges overlap, the combined range will be acquired.
		// If ranges do not overlap, the first range that is in main memory will be acquired.
		reservation_lock(u32 dst_addr, u32 dst_length, u32 src_addr, u32 src_length)
		{
			if (g_cfg.core.rsx_accurate_res_access)
			{
				const auto range1 = utils::address_range::start_length(dst_addr, dst_length);
				const auto range2 = utils::address_range::start_length(src_addr, src_length);
				utils::address_range target_range;

				if (!range1.overlaps(range2)) [[likely]]
				{
					target_range = (dst_addr < constants::local_mem_base) ? range1 : range2;
				}
				else
				{
					// Very unlikely
					target_range = range1.get_min_max(range2);
				}

				if (target_range.start < constants::local_mem_base)
				{
					lock_range(target_range.start, target_range.length());
				}
			}
		}

		~reservation_lock()
		{
			if (locked)
			{
				get_current_renderer()->iomap_table.unlock<IsFullLock>(addr, length);
			}
		}
	};

	class eng_lock
	{
		rsx::thread* pthr;
	public:
		eng_lock(rsx::thread* target)
			:pthr(target)
		{
			if (pthr->is_current_thread())
			{
				pthr = nullptr;
			}
			else
			{
				pthr->pause();
			}
		}

		~eng_lock()
		{
			if (pthr) pthr->unpause();
		}
	};
}