improvement/fixes: blur quality, blur range, Mac support, performance

- blur heuristics improved for reduced smearing
- velocity blur range fixed a bug
- fixed MacOs bug regarding doubles in shader
- lowered buffer accuracy and changed doubles to floats, improved performance

addons/MyJumpFloodIteration/jfp_backtracking_experimental.glsl

@@ -3,8 +3,6 @@
 
 #define FLT_MAX 3.402823466e+38
 #define FLT_MIN 1.175494351e-38
-#define DBL_MAX 1.7976931348623158e+308
-#define DBL_MIN 2.2250738585072014e-308
 
 layout(set = 0, binding = 0) uniform sampler2D depth_sampler;
 layout(set = 0, binding = 1) uniform sampler2D velocity_sampler;
@@ -15,7 +13,7 @@ layout(push_constant, std430) uniform Params
 {
 	int iteration_index;
 	int last_iteration_index;
-	int nan1;
+	int backtracking_sample_count;
 	int nan2;	
 	float perpen_error_thresh;
 	float sample_step_multiplier;
@@ -81,9 +79,9 @@ void set_value(bool a, ivec2 uvi, vec4 value, ivec2 render_size)
 float get_motion_difference(vec2 V, vec2 V2, float parallel_sensitivity, float perpendicular_sensitivity)
 {
 	vec2 VO = V - V2;
-	double parallel = abs(dot(VO, V) / max(DBL_MIN, dot(V, V)));
+	float parallel = abs(dot(VO, V) / max(FLT_MIN, dot(V, V)));
 	vec2 perpen_V = vec2(V.y, -V.x);
-	double perpendicular = abs(dot(VO, perpen_V) / max(DBL_MIN, dot(V, V)));
+	float perpendicular = abs(dot(VO, perpen_V) / max(FLT_MIN, dot(V, V)));
 	float difference = float(parallel) * parallel_sensitivity + float(perpendicular) * perpendicular_sensitivity;
 	return clamp(difference, 0, 1);
 }
@@ -103,19 +101,19 @@ vec4 sample_fitness(vec2 uv_offset, vec4 uv_sample)
 //		uv_offset = normalize(sample_velocity) * FLT_MIN;
 //	}
 
-	double velocity_space_distance = dot(sample_velocity, uv_offset) / max(FLT_MIN, dot(sample_velocity, sample_velocity));
+	float velocity_space_distance = dot(sample_velocity, uv_offset) / max(FLT_MIN, dot(sample_velocity, sample_velocity));
 
-	double mid_point = params.motion_blur_intensity / 2;
+	float mid_point = params.motion_blur_intensity / 2;
 
-	double absolute_velocity_space_distance = abs(velocity_space_distance - mid_point);
+	float absolute_velocity_space_distance = abs(velocity_space_distance - mid_point);
 
-	double within_velocity_range = step(absolute_velocity_space_distance, mid_point);
+	float within_velocity_range = step(absolute_velocity_space_distance, mid_point);
 	
 	vec2 perpen_offset = vec2(uv_offset.y, -uv_offset.x);
 
-	double side_offset = abs(dot(perpen_offset, sample_velocity)) / max(FLT_MIN, dot(sample_velocity, sample_velocity));
+	float side_offset = abs(dot(perpen_offset, sample_velocity)) / max(FLT_MIN, dot(sample_velocity, sample_velocity));
 	
-	double within_perpen_error_range = step(side_offset, params.perpen_error_thresh * params.motion_blur_intensity);
+	float within_perpen_error_range = step(side_offset, params.perpen_error_thresh * params.motion_blur_intensity);
 
 	return vec4(absolute_velocity_space_distance, velocity_space_distance, uv_sample.z, within_velocity_range * within_perpen_error_range);
 }
@@ -135,8 +133,6 @@ bool is_sample_better(vec4 a, vec4 b)
 vec4 get_backtracked_sample(vec2 uvn, vec2 chosen_uv, vec2 chosen_velocity, vec4 best_sample_fitness, vec2 render_size)
 {
 	//return vec4(chosen_uv, best_sample_fitness.x, 0);// comment this to enable backtracking
-	
-	int step_count = 16;
 
 	float smallest_step = 1 / max(render_size.x, render_size.y);
 
@@ -152,9 +148,9 @@ vec4 get_backtracked_sample(vec2 uvn, vec2 chosen_uv, vec2 chosen_velocity, vec4
 
 	int steps_to_compare = initial_steps_to_compare;
 
-	for(int i = -step_count; i < step_count + 1; i++)
+	for(int i = -params.backtracking_sample_count; i < params.backtracking_sample_count + 1; i++)
 	{
-		float velocity_multiplier = general_velocity_multiplier * (1 + float(i) /  float(step_count));
+		float velocity_multiplier = general_velocity_multiplier * (1 + float(i) /  float(params.backtracking_sample_count));
 
 		if(velocity_multiplier > params.motion_blur_intensity + 0.2 || velocity_multiplier < FLT_MIN)
 		{

addons/MyJumpFloodIteration/jump_flood_blur.gd

@@ -31,9 +31,15 @@ class_name MotionBlurSphynxJumpFlood
 @export var perpen_error_threshold : float = 0.3
 
 ## an initial step size that can increase the dilation radius proportionally, at the 
-## sacrifice of some quality in the final resolution of the dilation
+## sacrifice of some quality in the final resolution of the dilation.[br][br]
+## the formula for the maximum radius of the dilation (in pixels) is: pow(2, JFA_pass_count) * sample_step_multiplier
 @export var sample_step_multiplier : float = 8
 
+## how many steps along a range of 2 velocities from the 
+## dilation target velocity space do we go along to find a better fitting velocity sample
+## higher samples meaning higher detail getting captured and blurred
+@export var backtracking_sample_count : int = 8
+
 ## how sensitive the backtracking for velocities be
 @export var backtracking_velocity_match_threshold : float = 0.9
 
@@ -50,7 +56,8 @@ class_name MotionBlurSphynxJumpFlood
 @export var backtracbing_depth_match_threshold : float = 0.001
 
 ## the number of passes performed by the jump flood algorithm based dilation, 
-## each pass added doubles the maximum radius of dilation available
+## each pass added doubles the maximum radius of dilation available.[br][br]
+## the formula for the maximum radius of the dilation (in pixels) is: pow(2, JFA_pass_count) * sample_step_multiplier
 @export var JFA_pass_count : int = 3
 
 ## wether this motion blur stays the same intensity below
@@ -180,8 +187,8 @@ func _render_callback(p_effect_callback_type, p_render_data):
 				return
 			
 			ensure_texture(texture, render_scene_buffers)
-			ensure_texture(buffer_a, render_scene_buffers, true)#, Vector2(0.25, 0.25))
-			ensure_texture(buffer_b, render_scene_buffers, true)#, Vector2(0.25, 0.25))
+			ensure_texture(buffer_a, render_scene_buffers)
+			ensure_texture(buffer_b, render_scene_buffers)
 			ensure_texture(past_color, render_scene_buffers)
 
 			rd.draw_command_begin_label("Motion Blur", Color(1.0, 1.0, 1.0, 1.0))
@@ -229,7 +236,7 @@ func _render_callback(p_effect_callback_type, p_render_data):
 					var jf_push_constants : PackedInt32Array = [
 						i,
 						last_iteration_index,
-						16,
+						backtracking_sample_count,
 						16
 					]
 					

addons/MyJumpFloodIteration/jump_flood_blur.glsl

@@ -118,10 +118,15 @@ float get_velocity_curl(vec2 uv, vec2 render_size)
 
 // McGuire's functions https://docs.google.com/document/d/1IIlAKTj-O01hcXEdGxTErQbCHO9iBmRx6oFUy_Jm0fI/edit
 // ----------------------------------------------------------
+// This function would return 1 if depth_x is bigger than depth_y by sze amount. meaning, it would return 1 if x is closer to camera
 float soft_depth_compare(float depth_X, float depth_Y)
 {
 	return clamp(1 - (depth_X - depth_Y) / sze, 0, 1);
 }
+float soft_depth_compare_custom(float depth_X, float depth_Y, float csze)
+{
+	return clamp(1 - (depth_X - depth_Y) / csze, 0, 1);
+}
 
 float cone(vec2 X, vec2 Y, vec2 v)
 {
@@ -173,9 +178,9 @@ void main()
 
 	vec3 naive_velocity = -textureLod(vector_sampler, uvn, 0.0).xyz;
 
-	float max_dialtion_radius = pow(2, params.last_iteration_index) * params.sample_step_multiplier / max(render_size.x, render_size.y);
+	float max_dialtion_radius = pow(2, params.last_iteration_index) * params.sample_step_multiplier * 2 / max(render_size.x, render_size.y);
 
-	if ((dot(velocity, velocity) == 0 || params.motion_blur_intensity == 0) && params.debug == 0) //(uvn.y > 0.5)//
+	if ((dot(velocity, velocity) == 0 || params.motion_blur_intensity == 0) && params.debug == 0)
 	{
 		imageStore(output_image, uvi, base);
 		imageStore(past_color_image, uvi, base);
@@ -193,19 +198,21 @@ void main()
 	//float d = 1.0 - min(1.0, 2.0 * distance(uvn, vec2(0.5)));
 	//sample_step *= 1.0 - d * params.fade_padding.x;
 
-	float total_weight = 1;// max(0.0001, length(naive_velocity));
+	float total_weight = 1;
 	
-	vec2 offset = vec2(sample_step * noise_offset);
+	vec2 offset = vec2(sample_step * noise_offset);//vec2(0);//
 	
 	vec4 col = base * total_weight;
 
 	float depth = max(FLT_MIN, textureLod(depth_sampler, velocity_map_sample.xy, 0.0).x);
 
 	float naive_depth = max(FLT_MIN, textureLod(depth_sampler, uvn, 0.0).x);
-
+	
+	float naive_background = soft_depth_compare_custom(depth, naive_depth, 0.0001);
+	
 	for (int i = 1; i < iteration_count; i++) 
 	{
-		offset += sample_step.xy;// * interleaved_gradient_noise(uvi, int(params.frame) + i);
+		offset += sample_step.xy;
 
 		vec2 uvo = uvn + offset;
 
@@ -220,17 +227,23 @@ void main()
 
 		float current_depth = max(FLT_MIN, textureLod(depth_sampler, velocity_map_sample_step.xy, 0.0).x);
 
-		float sample_weight = 1;
+		float current_naive_depth = max(FLT_MIN, textureLod(depth_sampler, uvo, 0.0).x);
 
 		float motion_difference = get_motion_difference(velocity.xy, current_velocity.xy, 0.1);
 
 		float foreground = soft_depth_compare(npd / current_depth, npd / depth);
 
+		float naive_foreground = soft_depth_compare(npd / current_naive_depth - sze, npd / depth);//soft_depth_compare_custom(depth, current_naive_depth, 0.0001);//
+
+		float sample_weight = 1;
+
 		sample_weight *= 1 - (foreground * motion_difference);
 
 		total_weight += sample_weight;
 
-		col += textureLod(color_sampler, uvo, 0.0) * sample_weight;
+		vec2 sample_uv = mix(uvo, uvn, 1 - max(naive_background, naive_foreground));//uvo;//
+
+		col += textureLod(color_sampler, sample_uv, 0.0) * sample_weight;
 	}
 
 	col /= total_weight;
@@ -244,7 +257,7 @@ void main()
 
 	vec4 tl_col = vec4(abs(textureLod(vector_sampler, uvn, 0.0).xy) * 10, 0, 1);
 
-	vec4 tr_col = vec4(abs(velocity.xy) * 10, 0, 1);
+	vec4 tr_col = vec4(abs(velocity.xy) * 10, 0, 1);//vec4(naive_background);//
 
 	vec4 bl_col = vec4(abs(velocity_map_sample.xyw - vec3(uvn, 0)) * vec3(10, 10, 1), 1);