extends Control

signal render_initial_ready(key)  # A small chunk at the start has been rendered and is ready to play
signal render_complete(key)       # The full track has been rendered and is ready to pop-in

enum RENDER_MODE {BATCH, DYNAMIC, JAOT, TEST_READBACK}
var render_mode = RENDER_MODE.JAOT

const INPUT_TEX_WIDTH := 4096
const INPUT_FORMAT := Image.FORMAT_RGBA8  # Image.FORMAT_LA8
const INPUT_BYTES_PER_TEXEL := 4         # 2
const OUTPUT_BYTES_PER_TEXEL := 4
const QUAD_COLOR := PoolColorArray([Color.white, Color.white, Color.white, Color.white])
var OUTPUT_FRAMEBUFFER_SIZE: Vector2
var OUTPUT_WIDTH: int
var OUTPUT_HEIGHT: int
onready var viewport: Viewport = self.get_parent()
onready var render_queue: Array = []  # of [key, remaining_samples]
onready var cached_midis: Dictionary = {}  # key: [target_samples, ImageTexture]
onready var cached_renders: Dictionary = {}  # key: [remaining_samples, PoolByteArray]
onready var current_textures: Array = []  # of ImageTextures - Needed to prevent GC before draw
onready var waiting_for_viewport: Array = []
onready var done_first_draw: bool = false
var initialized_instrument_texture := false


func _ready() -> void:
	self.material = ShaderMaterial.new()
	self.material.shader = preload('res://shaders/audio_renderer.gdshader')
	match self.render_mode:
		RENDER_MODE.BATCH, RENDER_MODE.DYNAMIC:
			self._update_viewport(4096, 4096)
		RENDER_MODE.JAOT:
			self._update_viewport(256, 2048)

func _process(_delta) -> void:
	update()

func _draw() -> void:
	# Seems like the first one always fails
	if not self.done_first_draw:
		self.done_first_draw = true
		return

	if self.waiting_for_viewport:
		# Another node later in the draw sequence can call this within the same frame,
		# otherwise, this picks it up the following frame
		self.get_result()

	# self.waiting_for_viewport is cleared at the end of get_result()
	if self.render_queue:
		match self.render_mode:
			RENDER_MODE.BATCH:
				self._render_in_batch()
			RENDER_MODE.DYNAMIC:
				self._render_dynamic_one_at_a_time()
			RENDER_MODE.JAOT:
				self._render_just_ahead_of_time()


func _update_viewport(width: int, height: int) -> void:
	self.OUTPUT_WIDTH = width
	self.OUTPUT_HEIGHT = height
	self.OUTPUT_FRAMEBUFFER_SIZE = Vector2(width, height)
	self.viewport.size = OUTPUT_FRAMEBUFFER_SIZE
	self.material.set_shader_param('OUTPUT_FRAMEBUFFER_SIZE', OUTPUT_FRAMEBUFFER_SIZE)
	self.material.set_shader_param('INT_OUTPUT_WIDTH', OUTPUT_WIDTH)


func initialize_instrument_texture(reinitialize := false) -> void:
	if not reinitialize and self.initialized_instrument_texture:
		return
	SoundLoader.samples_to_texture()
	self.material.set_shader_param('instrument_samples', SoundLoader.samples_tex)
	self.material.set_shader_param('instrument_samples_size', SoundLoader.samples_tex.get_size())
	self.initialized_instrument_texture = true


func _get_cached_midi(key: String) -> Array:  # [target_samples: int, tex: ImageTexture, tempo_scale_thousandths: int]
	if not('-' in key):
		return self.cached_midis[key] + [1000]
	var split := key.split('-')
	var tempo_scale_thousandths := int(split[1])
	var target_time_and_tex = self.cached_midis[split[0]]
	var new_target_samples: int = (target_time_and_tex[0]*32000*1000)/tempo_scale_thousandths
	return [new_target_samples, target_time_and_tex[1], tempo_scale_thousandths]


func _render_midi(key: String, output_rows_drawn_including_this: int, rows_to_draw: int) -> void:
	var target_samples_and_tex_and_tempo = self._get_cached_midi(key)
	var target_samples: int = target_samples_and_tex_and_tempo[0]
	var tex: ImageTexture = target_samples_and_tex_and_tempo[1]
	var tempo_scale_thousandths: int = target_samples_and_tex_and_tempo[2]
	self.material.set_shader_param('tempo_scale_thousandths', tempo_scale_thousandths)
	if rows_to_draw <= OUTPUT_HEIGHT:  # Full draw
		var y_top: int = OUTPUT_HEIGHT - output_rows_drawn_including_this
		var y_bot: int = y_top + rows_to_draw
		var uv_inv_v: float = 1 - (rows_to_draw / OUTPUT_FRAMEBUFFER_SIZE.y)
		var uvs := PoolVector2Array([Vector2(0, uv_inv_v), Vector2(1, uv_inv_v), Vector2(1, 1), Vector2(0, 1)])
		var points := PoolVector2Array([Vector2(0, y_top), Vector2(OUTPUT_WIDTH, y_top), Vector2(OUTPUT_WIDTH, y_bot), Vector2(0, y_bot)])
		draw_primitive(points, QUAD_COLOR, uvs, tex)
		self.waiting_for_viewport.append([rows_to_draw, key])  # Grab the result next draw
	else:
		print('Could not fit %s into %dx%d output framebuffer, it needs %d rows'%[key, OUTPUT_WIDTH, OUTPUT_HEIGHT, rows_to_draw])


func _render_in_batch() -> void:
	self.waiting_for_viewport = []
	var rows_drawn := 0
	while self.render_queue:
		var target_samples: int = self.render_queue[0][1]
		var rows_to_draw := int(ceil(target_samples/float(OUTPUT_WIDTH)))
		rows_drawn += rows_to_draw
		if rows_drawn > OUTPUT_HEIGHT:
			if self.waiting_for_viewport.empty():
				print('Could not fit %s into %dx%d output framebuffer, it needs %d rows'%[self.render_queue[0][0], OUTPUT_WIDTH, OUTPUT_HEIGHT, rows_to_draw])
				self.render_queue.pop_front()
			break
		# Draw the next ImageTexture
		self._render_midi(self.render_queue.pop_front()[0], rows_drawn, rows_to_draw)


func _render_dynamic_one_at_a_time() -> void:  # Non power-of-two dimensioned textures should be restricted to GLES3
	self.waiting_for_viewport = []
	var entry = self.render_queue.pop_front()
	var key: String = entry[0]
	var target_samples: int = entry[1]
	var rows_to_draw := int(ceil(target_samples/float(OUTPUT_WIDTH)))
	self._update_viewport(4096, rows_to_draw)
	# Draw the next ImageTexture
	self._render_midi(key, rows_to_draw, rows_to_draw)


func _render_just_ahead_of_time() -> void:  # Optimized for latency
	var entry = self.render_queue[0]
	var key: String = entry[0]
	var remaining_samples: int = entry[1]

	var target_samples_and_tex_and_tempo = self._get_cached_midi(key)
	var total_target_samples: int = target_samples_and_tex_and_tempo[0]
	var tex: ImageTexture = target_samples_and_tex_and_tempo[1]
	var tempo_scale_thousandths: int = target_samples_and_tex_and_tempo[2]

	var total_rows := int(ceil(total_target_samples/float(OUTPUT_WIDTH)))
	var remaining_rows := int(ceil(remaining_samples/float(OUTPUT_WIDTH)))
	var rows_to_draw := remaining_rows
	# Draw the next ImageTexture
	self.material.set_shader_param('tempo_scale_thousandths', tempo_scale_thousandths)
	var y_top: int = 0
	if remaining_rows <= OUTPUT_HEIGHT:
		y_top = OUTPUT_HEIGHT - rows_to_draw
		self.render_queue.pop_front()
	else:
		rows_to_draw = OUTPUT_HEIGHT
		self.render_queue[0] = [key, remaining_samples - (rows_to_draw*OUTPUT_WIDTH)]
	# Draw the next ImageTexture
	var starting_row := total_rows - remaining_rows  # The start of the unrendered rows
	var ending_row := starting_row + rows_to_draw  # The end of this chunk of unrendered rows
	var uv_top: float = 1 - (ending_row / OUTPUT_FRAMEBUFFER_SIZE.y)
	var uv_bot: float = 1 - (starting_row / OUTPUT_FRAMEBUFFER_SIZE.y)
	var uvs := PoolVector2Array([Vector2(0, uv_top), Vector2(1, uv_top), Vector2(1, uv_bot), Vector2(0, uv_bot)])
	var points := PoolVector2Array([Vector2(0, y_top), Vector2(OUTPUT_WIDTH, y_top), Vector2(OUTPUT_WIDTH, OUTPUT_HEIGHT), Vector2(0, OUTPUT_HEIGHT)])
	draw_primitive(points, QUAD_COLOR, uvs, tex)
	self.waiting_for_viewport.append([rows_to_draw, key])  # Grab the result next draw


func push_image(image: Image, target_time: int, key: String, enqueue: bool = true) -> void:
	if not self.initialized_instrument_texture:
		self.initialize_instrument_texture()

	var tex := ImageTexture.new()
	tex.create_from_image(image, 0)
	self.cached_midis[key] = [target_time, tex]
	self.material.set_shader_param('midi_events_size', tex.get_size())  # Should all be the same size for now, revisit if we need mixed sizes.
	if enqueue:
		self.render_queue.append([key, target_time])

func push_bytes(data: PoolByteArray, target_time: int, key: String, enqueue: bool = true) -> void:
	var rows = int(pow(2, ceil(log((len(data)/INPUT_BYTES_PER_TEXEL) / INPUT_TEX_WIDTH)/log(2))))
	var target_length = rows * INPUT_BYTES_PER_TEXEL * INPUT_TEX_WIDTH
	while len(data) < target_length:  # This is inefficient, but this function should be called with pre-padded data anyway
		data.append(0)
	var image := Image.new()
	image.create_from_data(INPUT_TEX_WIDTH, rows, false, INPUT_FORMAT, data)
	self.push_image(image, target_time, key, enqueue)

func queue_cached_bgm(key: String) -> void:
	var new_target_time_etc := self._get_cached_midi(key)
	self.render_queue.append([key, new_target_time_etc[0]])


func get_result() -> void:
	var result_texture := self.viewport.get_texture()
	var result_image := result_texture.get_data()
	var result_bytes := result_image.get_data()
	match self.render_mode:
		RENDER_MODE.BATCH:  # Multiple complete songs in a fixed-size framebuffer
			var retrieved_rows := 0
			for rows_and_key in self.waiting_for_viewport:
				var entry_rows: int = rows_and_key[0]
				var key: String = rows_and_key[1]
				var bytes_start := retrieved_rows * OUTPUT_WIDTH * OUTPUT_BYTES_PER_TEXEL
				var bytes_end := (retrieved_rows + entry_rows) * OUTPUT_WIDTH * OUTPUT_BYTES_PER_TEXEL
				var entry_bytes := result_bytes.subarray(bytes_start, bytes_end-1)
				self.cached_renders[key] = [0, entry_bytes]
				emit_signal('render_initial_ready', key)
				emit_signal('render_complete', key)
				retrieved_rows += entry_rows
		RENDER_MODE.DYNAMIC:  # One complete song in a framebuffer sized to fit it
			var rows_and_key = self.waiting_for_viewport[0]
			var key: String = rows_and_key[1]
			self.cached_renders[key] = [0, result_bytes]
			emit_signal('render_initial_ready', key)
			emit_signal('render_complete', key)
		RENDER_MODE.JAOT:  # One partial song in a small framebuffer
			var rows_and_key = self.waiting_for_viewport[0]
			var key: String = rows_and_key[1]
			var samples_just_rendered: int = len(result_bytes) / OUTPUT_BYTES_PER_TEXEL
			if not (key in self.cached_renders):
				var samples_and_tex = _get_cached_midi(key)
				var target_samples: int = samples_and_tex[0]
				if samples_just_rendered >= target_samples:
					result_bytes.resize(target_samples * OUTPUT_BYTES_PER_TEXEL)
					self.cached_renders[key] = [0, result_bytes]
					emit_signal('render_initial_ready', key)
					emit_signal('render_complete', key)
				else:
					var remaining_samples: int = target_samples - samples_just_rendered
					self.cached_renders[key] = [remaining_samples, result_bytes]
					emit_signal('render_initial_ready', key)
			else:
				var old_remaining_samples_and_data = self.cached_renders[key]
				var target_samples: int = old_remaining_samples_and_data[0]
				var existing_data: PoolByteArray = old_remaining_samples_and_data[1]
				if samples_just_rendered >= target_samples:
					result_bytes.resize(target_samples * OUTPUT_BYTES_PER_TEXEL)
					self.cached_renders[key] = [0, existing_data + result_bytes]
					emit_signal('render_complete', key)
				else:
					self.cached_renders[key] = [target_samples - samples_just_rendered, existing_data + result_bytes]
		RENDER_MODE.TEST_READBACK:
			pass  # This has bitrotted
			# # Debugging: compare a sequence of all the possible 16bit integers
			# print_debug('result_image format is %d and has size'%result_image.get_format(), result_image.get_size(), result_bytes.subarray(0, 11))
			# test_readback(result_bytes)
	self.waiting_for_viewport = []


# This has bitrotted
# func test_readback(result_bytes: PoolByteArray):
# 	# Debugging: compare a sequence of all the possible 16bit integers
# 	var buff := StreamPeerBuffer.new()
# 	buff.set_data_array(result_bytes)
# 	var tex_readback = 0
# 	var uv_readback = 0
# 	for i in 0x1000:
# 		tex_readback = buff.get_u16()
# 		uv_readback = buff.get_u16()
# 		if tex_readback != i:
# 			print('tex readback %d (0x%04x) was instead %d (0x%04x)'%[i, i, tex_readback, tex_readback])
# 		if uv_readback != i:
# 			print('uv readback %d (0x%04x) was instead %d (0x%04x)'%[i, i, uv_readback, uv_readback])