Merge pull request #44182 from clayjohn/ASSAO

Replace SAO with ASSAO as Godot's new SSAO

12 files changed, 1714 insertions, 580 deletions

diff --git a/servers/rendering/renderer_rd/effects_rd.cpp b/servers/rendering/renderer_rd/effects_rd.cpp
index cbb000cb8c..b33255b54b 100644
--- a/servers/rendering/renderer_rd/effects_rd.cpp
+++ b/servers/rendering/renderer_rd/effects_rd.cpp
@@ -31,6 +31,7 @@
 #include "effects_rd.h"
 
 #include "core/config/project_settings.h"
+#include "core/math/math_defs.h"
 #include "core/os/os.h"
 
 #include "thirdparty/misc/cubemap_coeffs.h"
@@ -125,6 +126,33 @@ RID EffectsRD::_get_compute_uniform_set_from_texture(RID p_texture, bool p_use_m
 	return uniform_set;
 }
 
+RID EffectsRD::_get_compute_uniform_set_from_texture_and_sampler(RID p_texture, RID p_sampler) {
+	TextureSamplerPair tsp;
+	tsp.texture = p_texture;
+	tsp.sampler = p_sampler;
+
+	if (texture_sampler_to_compute_uniform_set_cache.has(tsp)) {
+		RID uniform_set = texture_sampler_to_compute_uniform_set_cache[tsp];
+		if (RD::get_singleton()->uniform_set_is_valid(uniform_set)) {
+			return uniform_set;
+		}
+	}
+
+	Vector<RD::Uniform> uniforms;
+	RD::Uniform u;
+	u.uniform_type = RD::UNIFORM_TYPE_SAMPLER_WITH_TEXTURE;
+	u.binding = 0;
+	u.ids.push_back(p_sampler);
+	u.ids.push_back(p_texture);
+	uniforms.push_back(u);
+	//any thing with the same configuration (one texture in binding 0 for set 0), is good
+	RID uniform_set = RD::get_singleton()->uniform_set_create(uniforms, ssao.blur_shader.version_get_shader(ssao.blur_shader_version, 0), 0);
+
+	texture_sampler_to_compute_uniform_set_cache[tsp] = uniform_set;
+
+	return uniform_set;
+}
+
 RID EffectsRD::_get_compute_uniform_set_from_texture_pair(RID p_texture1, RID p_texture2, bool p_use_mipmaps) {
 	TexturePair tp;
 	tp.texture1 = p_texture1;
@@ -951,157 +979,345 @@ void EffectsRD::bokeh_dof(RID p_base_texture, RID p_depth_texture, const Size2i
 	RD::get_singleton()->compute_list_end();
 }
 
-void EffectsRD::generate_ssao(RID p_depth_buffer, RID p_normal_buffer, const Size2i &p_depth_buffer_size, RID p_depth_mipmaps_texture, const Vector<RID> &depth_mipmaps, RID p_ao1, bool p_half_size, RID p_ao2, RID p_upscale_buffer, float p_intensity, float p_radius, float p_bias, const CameraMatrix &p_projection, RS::EnvironmentSSAOQuality p_quality, RS::EnvironmentSSAOBlur p_blur, float p_edge_sharpness) {
-	//minify first
-	ssao.minify_push_constant.orthogonal = p_projection.is_orthogonal();
-	ssao.minify_push_constant.z_near = p_projection.get_z_near();
-	ssao.minify_push_constant.z_far = p_projection.get_z_far();
-	ssao.minify_push_constant.pixel_size[0] = 1.0 / p_depth_buffer_size.x;
-	ssao.minify_push_constant.pixel_size[1] = 1.0 / p_depth_buffer_size.y;
-	ssao.minify_push_constant.source_size[0] = p_depth_buffer_size.x;
-	ssao.minify_push_constant.source_size[1] = p_depth_buffer_size.y;
+void EffectsRD::gather_ssao(RD::ComputeListID p_compute_list, const Vector<RID> p_ao_slices, const SSAOSettings &p_settings, bool p_adaptive_base_pass) {
+	RD::get_singleton()->compute_list_bind_uniform_set(p_compute_list, ssao.gather_uniform_set, 0);
+	if ((p_settings.quality == RS::ENV_SSAO_QUALITY_ULTRA) && !p_adaptive_base_pass) {
+		RD::get_singleton()->compute_list_bind_uniform_set(p_compute_list, ssao.importance_map_uniform_set, 1);
+	}
+
+	for (int i = 0; i < 4; i++) {
+		if ((p_settings.quality == RS::ENV_SSAO_QUALITY_VERY_LOW) && ((i == 1) || (i == 2))) {
+			continue;
+		}
+
+		ssao.gather_push_constant.pass_coord_offset[0] = i % 2;
+		ssao.gather_push_constant.pass_coord_offset[1] = i / 2;
+		ssao.gather_push_constant.pass_uv_offset[0] = ((i % 2) - 0.0) / p_settings.screen_size.x;
+		ssao.gather_push_constant.pass_uv_offset[1] = ((i / 2) - 0.0) / p_settings.screen_size.y;
+		ssao.gather_push_constant.pass = i;
+		RD::get_singleton()->compute_list_bind_uniform_set(p_compute_list, _get_uniform_set_from_image(p_ao_slices[i]), 2);
+		RD::get_singleton()->compute_list_set_push_constant(p_compute_list, &ssao.gather_push_constant, sizeof(SSAOGatherPushConstant));
+
+		int x_groups = ((p_settings.screen_size.x >> (p_settings.half_size ? 2 : 1)) - 1) / 8 + 1;
+		int y_groups = ((p_settings.screen_size.y >> (p_settings.half_size ? 2 : 1)) - 1) / 8 + 1;
 
+		RD::get_singleton()->compute_list_dispatch(p_compute_list, x_groups, y_groups, 1);
+	}
+	RD::get_singleton()->compute_list_add_barrier(p_compute_list);
+}
+
+void EffectsRD::generate_ssao(RID p_depth_buffer, RID p_normal_buffer, RID p_depth_mipmaps_texture, const Vector<RID> &depth_mipmaps, RID p_ao, const Vector<RID> p_ao_slices, RID p_ao_pong, const Vector<RID> p_ao_pong_slices, RID p_upscale_buffer, RID p_importance_map, RID p_importance_map_pong, const CameraMatrix &p_projection, const SSAOSettings &p_settings, bool p_invalidate_uniform_sets) {
 	RD::ComputeListID compute_list = RD::get_singleton()->compute_list_begin();
 
 	/* FIRST PASS */
-	// Minify the depth buffer.
-
-	for (int i = 0; i < depth_mipmaps.size(); i++) {
-		if (i == 0) {
-			RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[SSAO_MINIFY_FIRST]);
-			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_depth_buffer), 0);
-		} else {
-			if (i == 1) {
-				RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[SSAO_MINIFY_MIPMAP]);
+	// Downsample and deinterleave the depth buffer.
+	{
+		if (p_invalidate_uniform_sets) {
+			Vector<RD::Uniform> uniforms;
+			{
+				RD::Uniform u;
+				u.uniform_type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 0;
+				u.ids.push_back(depth_mipmaps[1]);
+				uniforms.push_back(u);
 			}
+			{
+				RD::Uniform u;
+				u.uniform_type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 1;
+				u.ids.push_back(depth_mipmaps[2]);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.uniform_type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 2;
+				u.ids.push_back(depth_mipmaps[3]);
+				uniforms.push_back(u);
+			}
+			ssao.downsample_uniform_set = RD::get_singleton()->uniform_set_create(uniforms, ssao.downsample_shader.version_get_shader(ssao.downsample_shader_version, 2), 2);
+		}
+
+		float depth_linearize_mul = -p_projection.matrix[3][2];
+		float depth_linearize_add = p_projection.matrix[2][2];
+		if (depth_linearize_mul * depth_linearize_add < 0) {
+			depth_linearize_add = -depth_linearize_add;
+		}
+
+		ssao.downsample_push_constant.orthogonal = p_projection.is_orthogonal();
+		ssao.downsample_push_constant.z_near = depth_linearize_mul;
+		ssao.downsample_push_constant.z_far = depth_linearize_add;
+		if (ssao.downsample_push_constant.orthogonal) {
+			ssao.downsample_push_constant.z_near = p_projection.get_z_near();
+			ssao.downsample_push_constant.z_far = p_projection.get_z_far();
+		}
+		ssao.downsample_push_constant.pixel_size[0] = 1.0 / p_settings.screen_size.x;
+		ssao.downsample_push_constant.pixel_size[1] = 1.0 / p_settings.screen_size.y;
+		ssao.downsample_push_constant.radius_sq = p_settings.radius * p_settings.radius;
+
+		int downsample_pipeline = SSAO_DOWNSAMPLE;
+		if (p_settings.quality == RS::ENV_SSAO_QUALITY_VERY_LOW) {
+			downsample_pipeline = SSAO_DOWNSAMPLE_HALF;
+		} else if (p_settings.quality > RS::ENV_SSAO_QUALITY_MEDIUM) {
+			downsample_pipeline = SSAO_DOWNSAMPLE_MIPMAP;
+		}
 
-			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(depth_mipmaps[i - 1]), 0);
+		if (p_settings.half_size) {
+			downsample_pipeline++;
 		}
-		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(depth_mipmaps[i]), 1);
 
-		RD::get_singleton()->compute_list_set_push_constant(compute_list, &ssao.minify_push_constant, sizeof(SSAOMinifyPushConstant));
-		// shrink after set
-		ssao.minify_push_constant.source_size[0] = MAX(1, ssao.minify_push_constant.source_size[0] >> 1);
-		ssao.minify_push_constant.source_size[1] = MAX(1, ssao.minify_push_constant.source_size[1] >> 1);
+		RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[downsample_pipeline]);
+		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_depth_buffer), 0);
+		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(depth_mipmaps[0]), 1);
+		if (p_settings.quality > RS::ENV_SSAO_QUALITY_MEDIUM) {
+			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, ssao.downsample_uniform_set, 2);
+		}
+		RD::get_singleton()->compute_list_set_push_constant(compute_list, &ssao.downsample_push_constant, sizeof(SSAODownsamplePushConstant));
 
-		int x_groups = (ssao.minify_push_constant.source_size[0] - 1) / 8 + 1;
-		int y_groups = (ssao.minify_push_constant.source_size[1] - 1) / 8 + 1;
+		int x_groups = (MAX(1, p_settings.screen_size.x >> (p_settings.half_size ? 2 : 1)) - 1) / 8 + 1;
+		int y_groups = (MAX(1, p_settings.screen_size.y >> (p_settings.half_size ? 2 : 1)) - 1) / 8 + 1;
 
 		RD::get_singleton()->compute_list_dispatch(compute_list, x_groups, y_groups, 1);
 		RD::get_singleton()->compute_list_add_barrier(compute_list);
 	}
 
 	/* SECOND PASS */
-	// Gather samples
-
-	RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[(SSAO_GATHER_LOW + p_quality) + (p_half_size ? 4 : 0)]);
-	RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_depth_mipmaps_texture), 0);
-	RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_ao1), 1);
-	if (!p_half_size) {
-		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_depth_buffer), 2);
-	}
-	RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_normal_buffer), 3);
-
-	ssao.gather_push_constant.screen_size[0] = p_depth_buffer_size.x;
-	ssao.gather_push_constant.screen_size[1] = p_depth_buffer_size.y;
-	if (p_half_size) {
-		ssao.gather_push_constant.screen_size[0] >>= 1;
-		ssao.gather_push_constant.screen_size[1] >>= 1;
-	}
-	ssao.gather_push_constant.z_far = p_projection.get_z_far();
-	ssao.gather_push_constant.z_near = p_projection.get_z_near();
-	ssao.gather_push_constant.orthogonal = p_projection.is_orthogonal();
-
-	ssao.gather_push_constant.proj_info[0] = -2.0f / (ssao.gather_push_constant.screen_size[0] * p_projection.matrix[0][0]);
-	ssao.gather_push_constant.proj_info[1] = -2.0f / (ssao.gather_push_constant.screen_size[1] * p_projection.matrix[1][1]);
-	ssao.gather_push_constant.proj_info[2] = (1.0f - p_projection.matrix[0][2]) / p_projection.matrix[0][0];
-	ssao.gather_push_constant.proj_info[3] = (1.0f + p_projection.matrix[1][2]) / p_projection.matrix[1][1];
-	//ssao.gather_push_constant.proj_info[2] = (1.0f - p_projection.matrix[0][2]) / p_projection.matrix[0][0];
-	//ssao.gather_push_constant.proj_info[3] = -(1.0f + p_projection.matrix[1][2]) / p_projection.matrix[1][1];
-
-	ssao.gather_push_constant.radius = p_radius;
-
-	ssao.gather_push_constant.proj_scale = float(p_projection.get_pixels_per_meter(ssao.gather_push_constant.screen_size[0]));
-	ssao.gather_push_constant.bias = p_bias;
-	ssao.gather_push_constant.intensity_div_r6 = p_intensity / pow(p_radius, 6.0f);
-
-	ssao.gather_push_constant.pixel_size[0] = 1.0 / p_depth_buffer_size.x;
-	ssao.gather_push_constant.pixel_size[1] = 1.0 / p_depth_buffer_size.y;
-
-	RD::get_singleton()->compute_list_set_push_constant(compute_list, &ssao.gather_push_constant, sizeof(SSAOGatherPushConstant));
+	// Sample SSAO
+	{
+		ssao.gather_push_constant.screen_size[0] = p_settings.screen_size.x;
+		ssao.gather_push_constant.screen_size[1] = p_settings.screen_size.y;
+
+		ssao.gather_push_constant.half_screen_pixel_size[0] = 1.0 / p_settings.half_screen_size.x;
+		ssao.gather_push_constant.half_screen_pixel_size[1] = 1.0 / p_settings.half_screen_size.y;
+		float tan_half_fov_x = 1.0 / p_projection.matrix[0][0];
+		float tan_half_fov_y = 1.0 / p_projection.matrix[1][1];
+		ssao.gather_push_constant.NDC_to_view_mul[0] = tan_half_fov_x * 2.0;
+		ssao.gather_push_constant.NDC_to_view_mul[1] = tan_half_fov_y * -2.0;
+		ssao.gather_push_constant.NDC_to_view_add[0] = tan_half_fov_x * -1.0;
+		ssao.gather_push_constant.NDC_to_view_add[1] = tan_half_fov_y;
+		ssao.gather_push_constant.is_orthogonal = p_projection.is_orthogonal();
+
+		ssao.gather_push_constant.half_screen_pixel_size_x025[0] = ssao.gather_push_constant.half_screen_pixel_size[0] * 0.25;
+		ssao.gather_push_constant.half_screen_pixel_size_x025[1] = ssao.gather_push_constant.half_screen_pixel_size[1] * 0.25;
+
+		float radius_near_limit = (p_settings.radius * 1.2f);
+		if (p_settings.quality <= RS::ENV_SSAO_QUALITY_LOW) {
+			radius_near_limit *= 1.50f;
+
+			if (p_settings.quality == RS::ENV_SSAO_QUALITY_VERY_LOW) {
+				ssao.gather_push_constant.radius *= 0.8f;
+			}
+			if (p_settings.half_size) {
+				ssao.gather_push_constant.radius *= 0.5f;
+			}
+		}
+		radius_near_limit /= tan_half_fov_y;
+		ssao.gather_push_constant.radius = p_settings.radius;
+		ssao.gather_push_constant.intensity = p_settings.intensity;
+		ssao.gather_push_constant.shadow_power = p_settings.power;
+		ssao.gather_push_constant.shadow_clamp = 0.98;
+		ssao.gather_push_constant.fade_out_mul = -1.0 / (p_settings.fadeout_to - p_settings.fadeout_from);
+		ssao.gather_push_constant.fade_out_add = p_settings.fadeout_from / (p_settings.fadeout_to - p_settings.fadeout_from) + 1.0;
+		ssao.gather_push_constant.horizon_angle_threshold = p_settings.horizon;
+		ssao.gather_push_constant.inv_radius_near_limit = 1.0f / radius_near_limit;
+		ssao.gather_push_constant.neg_inv_radius = -1.0 / ssao.gather_push_constant.radius;
+
+		ssao.gather_push_constant.load_counter_avg_div = 9.0 / float((p_settings.quarter_size.x) * (p_settings.quarter_size.y) * 255);
+		ssao.gather_push_constant.adaptive_sample_limit = p_settings.adaptive_target;
+
+		ssao.gather_push_constant.detail_intensity = p_settings.detail;
+		ssao.gather_push_constant.quality = MAX(0, p_settings.quality - 1);
+		ssao.gather_push_constant.size_multiplier = p_settings.half_size ? 2 : 1;
+
+		if (p_invalidate_uniform_sets) {
+			Vector<RD::Uniform> uniforms;
+			{
+				RD::Uniform u;
+				u.uniform_type = RD::UNIFORM_TYPE_SAMPLER_WITH_TEXTURE;
+				u.binding = 0;
+				u.ids.push_back(ssao.mirror_sampler);
+				u.ids.push_back(p_depth_mipmaps_texture);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.uniform_type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 1;
+				u.ids.push_back(p_normal_buffer);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.uniform_type = RD::UNIFORM_TYPE_UNIFORM_BUFFER;
+				u.binding = 2;
+				u.ids.push_back(ssao.gather_constants_buffer);
+				uniforms.push_back(u);
+			}
+			ssao.gather_uniform_set = RD::get_singleton()->uniform_set_create(uniforms, ssao.gather_shader.version_get_shader(ssao.gather_shader_version, 0), 0);
+		}
 
-	int x_groups = (ssao.gather_push_constant.screen_size[0] - 1) / 8 + 1;
-	int y_groups = (ssao.gather_push_constant.screen_size[1] - 1) / 8 + 1;
+		if (p_invalidate_uniform_sets) {
+			Vector<RD::Uniform> uniforms;
+			{
+				RD::Uniform u;
+				u.uniform_type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 0;
+				u.ids.push_back(p_ao_pong);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.uniform_type = RD::UNIFORM_TYPE_SAMPLER_WITH_TEXTURE;
+				u.binding = 1;
+				u.ids.push_back(default_sampler);
+				u.ids.push_back(p_importance_map);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.uniform_type = RD::UNIFORM_TYPE_STORAGE_BUFFER;
+				u.binding = 2;
+				u.ids.push_back(ssao.importance_map_load_counter);
+				uniforms.push_back(u);
+			}
+			ssao.importance_map_uniform_set = RD::get_singleton()->uniform_set_create(uniforms, ssao.gather_shader.version_get_shader(ssao.gather_shader_version, 2), 1);
+		}
 
-	RD::get_singleton()->compute_list_dispatch(compute_list, x_groups, y_groups, 1);
-	RD::get_singleton()->compute_list_add_barrier(compute_list);
+		if (p_settings.quality == RS::ENV_SSAO_QUALITY_ULTRA) {
+			ssao.importance_map_push_constant.half_screen_pixel_size[0] = 1.0 / p_settings.half_screen_size.x;
+			ssao.importance_map_push_constant.half_screen_pixel_size[1] = 1.0 / p_settings.half_screen_size.y;
+			ssao.importance_map_push_constant.intensity = p_settings.intensity;
+			ssao.importance_map_push_constant.power = p_settings.power;
+			//base pass
+			RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[SSAO_GATHER_BASE]);
+			gather_ssao(compute_list, p_ao_pong_slices, p_settings, true);
+			//generate importance map
+			int x_groups = (p_settings.quarter_size.x - 1) / 8 + 1;
+			int y_groups = (p_settings.quarter_size.y - 1) / 8 + 1;
+
+			RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[SSAO_GENERATE_IMPORTANCE_MAP]);
+			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_ao_pong), 0);
+			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_importance_map), 1);
+			RD::get_singleton()->compute_list_set_push_constant(compute_list, &ssao.importance_map_push_constant, sizeof(SSAOImportanceMapPushConstant));
+			RD::get_singleton()->compute_list_dispatch(compute_list, x_groups, y_groups, 1);
+			RD::get_singleton()->compute_list_add_barrier(compute_list);
+			//process importance map A
+			RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[SSAO_PROCESS_IMPORTANCE_MAPA]);
+			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_importance_map), 0);
+			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_importance_map_pong), 1);
+			RD::get_singleton()->compute_list_set_push_constant(compute_list, &ssao.importance_map_push_constant, sizeof(SSAOImportanceMapPushConstant));
+			RD::get_singleton()->compute_list_dispatch(compute_list, x_groups, y_groups, 1);
+			RD::get_singleton()->compute_list_add_barrier(compute_list);
+			//process Importance Map B
+			RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[SSAO_PROCESS_IMPORTANCE_MAPB]);
+			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_importance_map_pong), 0);
+			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_importance_map), 1);
+			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, ssao.counter_uniform_set, 2);
+			RD::get_singleton()->compute_list_set_push_constant(compute_list, &ssao.importance_map_push_constant, sizeof(SSAOImportanceMapPushConstant));
+			RD::get_singleton()->compute_list_dispatch(compute_list, x_groups, y_groups, 1);
+			RD::get_singleton()->compute_list_add_barrier(compute_list);
 
-	/* THIRD PASS */
-	// Blur horizontal
-
-	ssao.blur_push_constant.edge_sharpness = p_edge_sharpness;
-	ssao.blur_push_constant.filter_scale = p_blur;
-	ssao.blur_push_constant.screen_size[0] = ssao.gather_push_constant.screen_size[0];
-	ssao.blur_push_constant.screen_size[1] = ssao.gather_push_constant.screen_size[1];
-	ssao.blur_push_constant.z_far = p_projection.get_z_far();
-	ssao.blur_push_constant.z_near = p_projection.get_z_near();
-	ssao.blur_push_constant.orthogonal = p_projection.is_orthogonal();
-	ssao.blur_push_constant.axis[0] = 1;
-	ssao.blur_push_constant.axis[1] = 0;
-
-	if (p_blur != RS::ENV_SSAO_BLUR_DISABLED) {
-		RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[p_half_size ? SSAO_BLUR_PASS_HALF : SSAO_BLUR_PASS]);
-		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_ao1), 0);
-		if (p_half_size) {
-			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_depth_mipmaps_texture), 1);
+			RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[SSAO_GATHER_ADAPTIVE]);
 		} else {
-			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_depth_buffer), 1);
+			RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[SSAO_GATHER]);
 		}
-		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_ao2), 3);
-
-		RD::get_singleton()->compute_list_set_push_constant(compute_list, &ssao.blur_push_constant, sizeof(SSAOBlurPushConstant));
 
-		RD::get_singleton()->compute_list_dispatch(compute_list, x_groups, y_groups, 1);
-		RD::get_singleton()->compute_list_add_barrier(compute_list);
-
-		/* THIRD PASS */
-		// Blur vertical
-
-		ssao.blur_push_constant.axis[0] = 0;
-		ssao.blur_push_constant.axis[1] = 1;
+		gather_ssao(compute_list, p_ao_slices, p_settings, false);
+	}
 
-		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_ao2), 0);
-		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_ao1), 3);
+	//	/* THIRD PASS */
+	//	// Blur
+	//
+	{
+		ssao.blur_push_constant.edge_sharpness = 1.0 - p_settings.sharpness;
+		ssao.blur_push_constant.half_screen_pixel_size[0] = 1.0 / p_settings.half_screen_size.x;
+		ssao.blur_push_constant.half_screen_pixel_size[1] = 1.0 / p_settings.half_screen_size.y;
+
+		int blur_passes = p_settings.quality > RS::ENV_SSAO_QUALITY_VERY_LOW ? p_settings.blur_passes : 1;
+
+		for (int pass = 0; pass < blur_passes; pass++) {
+			int blur_pipeline = SSAO_BLUR_PASS;
+			if (p_settings.quality > RS::ENV_SSAO_QUALITY_VERY_LOW) {
+				if (pass < blur_passes - 2) {
+					blur_pipeline = SSAO_BLUR_PASS_WIDE;
+				}
+				blur_pipeline = SSAO_BLUR_PASS_SMART;
+			}
 
-		RD::get_singleton()->compute_list_set_push_constant(compute_list, &ssao.blur_push_constant, sizeof(SSAOBlurPushConstant));
+			for (int i = 0; i < 4; i++) {
+				if ((p_settings.quality == RS::ENV_SSAO_QUALITY_VERY_LOW) && ((i == 1) || (i == 2))) {
+					continue;
+				}
+
+				RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[blur_pipeline]);
+				if (pass % 2 == 0) {
+					if (p_settings.quality == RS::ENV_SSAO_QUALITY_VERY_LOW) {
+						RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_ao_slices[i]), 0);
+					} else {
+						RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture_and_sampler(p_ao_slices[i], ssao.mirror_sampler), 0);
+					}
+
+					RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_ao_pong_slices[i]), 1);
+				} else {
+					if (p_settings.quality == RS::ENV_SSAO_QUALITY_VERY_LOW) {
+						RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_ao_pong_slices[i]), 0);
+					} else {
+						RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture_and_sampler(p_ao_pong_slices[i], ssao.mirror_sampler), 0);
+					}
+					RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_ao_slices[i]), 1);
+				}
+				RD::get_singleton()->compute_list_set_push_constant(compute_list, &ssao.blur_push_constant, sizeof(SSAOBlurPushConstant));
+
+				int x_groups = ((p_settings.screen_size.x >> (p_settings.half_size ? 2 : 1)) - 1) / 8 + 1;
+				int y_groups = ((p_settings.screen_size.y >> (p_settings.half_size ? 2 : 1)) - 1) / 8 + 1;
+
+				RD::get_singleton()->compute_list_dispatch(compute_list, x_groups, y_groups, 1);
+			}
 
-		RD::get_singleton()->compute_list_dispatch(compute_list, x_groups, y_groups, 1);
+			if (p_settings.quality > RS::ENV_SSAO_QUALITY_VERY_LOW) {
+				RD::get_singleton()->compute_list_add_barrier(compute_list);
+			}
+		}
 	}
-	if (p_half_size) { //must upscale
 
-		/* FOURTH PASS */
-		// upscale if half size
-		//back to full size
-		ssao.blur_push_constant.screen_size[0] = p_depth_buffer_size.x;
-		ssao.blur_push_constant.screen_size[1] = p_depth_buffer_size.y;
+	/* FOURTH PASS */
+	// Interleave buffers
+	// back to full size
+	{
+		ssao.interleave_push_constant.inv_sharpness = 1.0 - p_settings.sharpness;
+		ssao.interleave_push_constant.pixel_size[0] = 1.0 / p_settings.screen_size.x;
+		ssao.interleave_push_constant.pixel_size[1] = 1.0 / p_settings.screen_size.y;
+		ssao.interleave_push_constant.size_modifier = uint32_t(p_settings.half_size ? 4 : 2);
+
+		int interleave_pipeline = SSAO_INTERLEAVE_HALF;
+		if (p_settings.quality == RS::ENV_SSAO_QUALITY_LOW) {
+			interleave_pipeline = SSAO_INTERLEAVE;
+		} else if (p_settings.quality >= RS::ENV_SSAO_QUALITY_MEDIUM) {
+			interleave_pipeline = SSAO_INTERLEAVE_SMART;
+		}
 
-		RD::get_singleton()->compute_list_add_barrier(compute_list);
-		RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[SSAO_BLUR_UPSCALE]);
+		RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, ssao.pipelines[interleave_pipeline]);
 
-		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_ao1), 0);
-		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_upscale_buffer), 3);
-		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_depth_buffer), 1);
-		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_depth_mipmaps_texture), 2);
+		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_upscale_buffer), 0);
+		if (p_settings.quality > RS::ENV_SSAO_QUALITY_VERY_LOW && p_settings.blur_passes % 2 == 0) {
+			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_ao), 1);
+		} else {
+			RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_ao_pong), 1);
+		}
 
-		RD::get_singleton()->compute_list_set_push_constant(compute_list, &ssao.blur_push_constant, sizeof(SSAOBlurPushConstant)); //not used but set anyway
+		RD::get_singleton()->compute_list_set_push_constant(compute_list, &ssao.interleave_push_constant, sizeof(SSAOInterleavePushConstant));
 
-		x_groups = (p_depth_buffer_size.x - 1) / 8 + 1;
-		y_groups = (p_depth_buffer_size.y - 1) / 8 + 1;
+		int x_groups = (p_settings.screen_size.x - 1) / 8 + 1;
+		int y_groups = (p_settings.screen_size.y - 1) / 8 + 1;
 
 		RD::get_singleton()->compute_list_dispatch(compute_list, x_groups, y_groups, 1);
+		RD::get_singleton()->compute_list_add_barrier(compute_list);
 	}
 
 	RD::get_singleton()->compute_list_end();
+
+	int zero[1] = { 0 };
+	RD::get_singleton()->buffer_update(ssao.importance_map_load_counter, 0, sizeof(uint32_t), &zero, false);
 }
 
 void EffectsRD::roughness_limit(RID p_source_normal, RID p_roughness, const Size2i &p_size, float p_curve) {
@@ -1486,57 +1702,142 @@ EffectsRD::EffectsRD() {
 
 	{
 		// Initialize ssao
+
+		RD::SamplerState sampler;
+		sampler.mag_filter = RD::SAMPLER_FILTER_NEAREST;
+		sampler.min_filter = RD::SAMPLER_FILTER_NEAREST;
+		sampler.mip_filter = RD::SAMPLER_FILTER_NEAREST;
+		sampler.repeat_u = RD::SAMPLER_REPEAT_MODE_MIRRORED_REPEAT;
+		sampler.repeat_v = RD::SAMPLER_REPEAT_MODE_MIRRORED_REPEAT;
+		sampler.repeat_w = RD::SAMPLER_REPEAT_MODE_MIRRORED_REPEAT;
+		sampler.max_lod = 4;
+
+		ssao.mirror_sampler = RD::get_singleton()->sampler_create(sampler);
+
 		uint32_t pipeline = 0;
 		{
 			Vector<String> ssao_modes;
-			ssao_modes.push_back("\n#define MINIFY_START\n");
 			ssao_modes.push_back("\n");
+			ssao_modes.push_back("\n#define USE_HALF_SIZE\n");
+			ssao_modes.push_back("\n#define GENERATE_MIPS\n");
+			ssao_modes.push_back("\n#define GENERATE_MIPS\n#define USE_HALF_SIZE");
+			ssao_modes.push_back("\n#define USE_HALF_BUFFERS\n");
+			ssao_modes.push_back("\n#define USE_HALF_BUFFERS\n#define USE_HALF_SIZE");
 
-			ssao.minify_shader.initialize(ssao_modes);
+			ssao.downsample_shader.initialize(ssao_modes);
 
-			ssao.minify_shader_version = ssao.minify_shader.version_create();
+			ssao.downsample_shader_version = ssao.downsample_shader.version_create();
 
-			for (int i = 0; i <= SSAO_MINIFY_MIPMAP; i++) {
-				ssao.pipelines[pipeline] = RD::get_singleton()->compute_pipeline_create(ssao.minify_shader.version_get_shader(ssao.minify_shader_version, i));
+			for (int i = 0; i <= SSAO_DOWNSAMPLE_HALF_RES_HALF; i++) {
+				ssao.pipelines[pipeline] = RD::get_singleton()->compute_pipeline_create(ssao.downsample_shader.version_get_shader(ssao.downsample_shader_version, i));
 				pipeline++;
 			}
 		}
 		{
 			Vector<String> ssao_modes;
-			ssao_modes.push_back("\n#define SSAO_QUALITY_LOW\n");
+
 			ssao_modes.push_back("\n");
-			ssao_modes.push_back("\n#define SSAO_QUALITY_HIGH\n");
-			ssao_modes.push_back("\n#define SSAO_QUALITY_ULTRA\n");
-			ssao_modes.push_back("\n#define SSAO_QUALITY_LOW\n#define USE_HALF_SIZE\n");
-			ssao_modes.push_back("\n#define USE_HALF_SIZE\n");
-			ssao_modes.push_back("\n#define SSAO_QUALITY_HIGH\n#define USE_HALF_SIZE\n");
-			ssao_modes.push_back("\n#define SSAO_QUALITY_ULTRA\n#define USE_HALF_SIZE\n");
+			ssao_modes.push_back("\n#define SSAO_BASE\n");
+			ssao_modes.push_back("\n#define ADAPTIVE\n");
 
 			ssao.gather_shader.initialize(ssao_modes);
 
 			ssao.gather_shader_version = ssao.gather_shader.version_create();
 
-			for (int i = SSAO_GATHER_LOW; i <= SSAO_GATHER_ULTRA_HALF; i++) {
-				ssao.pipelines[pipeline] = RD::get_singleton()->compute_pipeline_create(ssao.gather_shader.version_get_shader(ssao.gather_shader_version, i - SSAO_GATHER_LOW));
+			for (int i = SSAO_GATHER; i <= SSAO_GATHER_ADAPTIVE; i++) {
+				ssao.pipelines[pipeline] = RD::get_singleton()->compute_pipeline_create(ssao.gather_shader.version_get_shader(ssao.gather_shader_version, i - SSAO_GATHER));
 				pipeline++;
 			}
+
+			ssao.gather_constants_buffer = RD::get_singleton()->uniform_buffer_create(sizeof(SSAOGatherConstants));
+			SSAOGatherConstants gather_constants;
+
+			const int sub_pass_count = 5;
+			for (int pass = 0; pass < 4; pass++) {
+				for (int subPass = 0; subPass < sub_pass_count; subPass++) {
+					int a = pass;
+					int b = subPass;
+
+					int spmap[5]{ 0, 1, 4, 3, 2 };
+					b = spmap[subPass];
+
+					float ca, sa;
+					float angle0 = (float(a) + float(b) / float(sub_pass_count)) * Math_PI * 0.5f;
+
+					ca = Math::cos(angle0);
+					sa = Math::sin(angle0);
+
+					float scale = 1.0f + (a - 1.5f + (b - (sub_pass_count - 1.0f) * 0.5f) / float(sub_pass_count)) * 0.07f;
+
+					gather_constants.rotation_matrices[pass * 20 + subPass * 4 + 0] = scale * ca;
+					gather_constants.rotation_matrices[pass * 20 + subPass * 4 + 1] = scale * -sa;
+					gather_constants.rotation_matrices[pass * 20 + subPass * 4 + 2] = -scale * sa;
+					gather_constants.rotation_matrices[pass * 20 + subPass * 4 + 3] = -scale * ca;
+				}
+			}
+
+			RD::get_singleton()->buffer_update(ssao.gather_constants_buffer, 0, sizeof(SSAOGatherConstants), &gather_constants, false);
 		}
 		{
 			Vector<String> ssao_modes;
-			ssao_modes.push_back("\n#define MODE_FULL_SIZE\n");
-			ssao_modes.push_back("\n");
-			ssao_modes.push_back("\n#define MODE_UPSCALE\n");
+			ssao_modes.push_back("\n#define GENERATE_MAP\n");
+			ssao_modes.push_back("\n#define PROCESS_MAPA\n");
+			ssao_modes.push_back("\n#define PROCESS_MAPB\n");
+
+			ssao.importance_map_shader.initialize(ssao_modes);
+
+			ssao.importance_map_shader_version = ssao.importance_map_shader.version_create();
+
+			for (int i = SSAO_GENERATE_IMPORTANCE_MAP; i <= SSAO_PROCESS_IMPORTANCE_MAPB; i++) {
+				ssao.pipelines[pipeline] = RD::get_singleton()->compute_pipeline_create(ssao.importance_map_shader.version_get_shader(ssao.importance_map_shader_version, i - SSAO_GENERATE_IMPORTANCE_MAP));
+
+				pipeline++;
+			}
+			ssao.importance_map_load_counter = RD::get_singleton()->storage_buffer_create(sizeof(uint32_t));
+			int zero[1] = { 0 };
+			RD::get_singleton()->buffer_update(ssao.importance_map_load_counter, 0, sizeof(uint32_t), &zero, false);
+
+			Vector<RD::Uniform> uniforms;
+			{
+				RD::Uniform u;
+				u.uniform_type = RD::UNIFORM_TYPE_STORAGE_BUFFER;
+				u.binding = 0;
+				u.ids.push_back(ssao.importance_map_load_counter);
+				uniforms.push_back(u);
+			}
+			ssao.counter_uniform_set = RD::get_singleton()->uniform_set_create(uniforms, ssao.importance_map_shader.version_get_shader(ssao.importance_map_shader_version, 2), 2);
+		}
+		{
+			Vector<String> ssao_modes;
+			ssao_modes.push_back("\n#define MODE_NON_SMART\n");
+			ssao_modes.push_back("\n#define MODE_SMART\n");
+			ssao_modes.push_back("\n#define MODE_WIDE\n");
 
 			ssao.blur_shader.initialize(ssao_modes);
 
 			ssao.blur_shader_version = ssao.blur_shader.version_create();
 
-			for (int i = SSAO_BLUR_PASS; i <= SSAO_BLUR_UPSCALE; i++) {
+			for (int i = SSAO_BLUR_PASS; i <= SSAO_BLUR_PASS_WIDE; i++) {
 				ssao.pipelines[pipeline] = RD::get_singleton()->compute_pipeline_create(ssao.blur_shader.version_get_shader(ssao.blur_shader_version, i - SSAO_BLUR_PASS));
 
 				pipeline++;
 			}
 		}
+		{
+			Vector<String> ssao_modes;
+			ssao_modes.push_back("\n#define MODE_NON_SMART\n");
+			ssao_modes.push_back("\n#define MODE_SMART\n");
+			ssao_modes.push_back("\n#define MODE_HALF\n");
+
+			ssao.interleave_shader.initialize(ssao_modes);
+
+			ssao.interleave_shader_version = ssao.interleave_shader.version_create();
+			for (int i = SSAO_INTERLEAVE; i <= SSAO_INTERLEAVE_HALF; i++) {
+				ssao.pipelines[pipeline] = RD::get_singleton()->compute_pipeline_create(ssao.interleave_shader.version_get_shader(ssao.interleave_shader_version, i - SSAO_INTERLEAVE));
+
+				pipeline++;
+			}
+		}
 
 		ERR_FAIL_COND(pipeline != SSAO_MAX);
 	}
@@ -1777,6 +2078,10 @@ EffectsRD::~EffectsRD() {
 	RD::get_singleton()->free(index_buffer); //array gets freed as dependency
 	RD::get_singleton()->free(filter.coefficient_buffer);
 
+	RD::get_singleton()->free(ssao.mirror_sampler);
+	RD::get_singleton()->free(ssao.gather_constants_buffer);
+	RD::get_singleton()->free(ssao.importance_map_load_counter);
+
 	bokeh.shader.version_free(bokeh.shader_version);
 	copy.shader.version_free(copy.shader_version);
 	copy_to_fb.shader.version_free(copy_to_fb.shader_version);
@@ -1791,7 +2096,9 @@ EffectsRD::~EffectsRD() {
 	specular_merge.shader.version_free(specular_merge.shader_version);
 	ssao.blur_shader.version_free(ssao.blur_shader_version);
 	ssao.gather_shader.version_free(ssao.gather_shader_version);
-	ssao.minify_shader.version_free(ssao.minify_shader_version);
+	ssao.downsample_shader.version_free(ssao.downsample_shader_version);
+	ssao.interleave_shader.version_free(ssao.interleave_shader_version);
+	ssao.importance_map_shader.version_free(ssao.importance_map_shader_version);
 	ssr.shader.version_free(ssr.shader_version);
 	ssr_filter.shader.version_free(ssr_filter.shader_version);
 	ssr_scale.shader.version_free(ssr_scale.shader_version);
diff --git a/servers/rendering/renderer_rd/effects_rd.h b/servers/rendering/renderer_rd/effects_rd.h
index 3afc111b9d..8731466dea 100644
--- a/servers/rendering/renderer_rd/effects_rd.h
+++ b/servers/rendering/renderer_rd/effects_rd.h
@@ -51,7 +51,9 @@
 #include "servers/rendering/renderer_rd/shaders/specular_merge.glsl.gen.h"
 #include "servers/rendering/renderer_rd/shaders/ssao.glsl.gen.h"
 #include "servers/rendering/renderer_rd/shaders/ssao_blur.glsl.gen.h"
-#include "servers/rendering/renderer_rd/shaders/ssao_minify.glsl.gen.h"
+#include "servers/rendering/renderer_rd/shaders/ssao_downsample.glsl.gen.h"
+#include "servers/rendering/renderer_rd/shaders/ssao_importance_map.glsl.gen.h"
+#include "servers/rendering/renderer_rd/shaders/ssao_interleave.glsl.gen.h"
 #include "servers/rendering/renderer_rd/shaders/subsurface_scattering.glsl.gen.h"
 #include "servers/rendering/renderer_rd/shaders/tonemap.glsl.gen.h"
 
@@ -288,71 +290,121 @@ class EffectsRD {
 	} bokeh;
 
 	enum SSAOMode {
-		SSAO_MINIFY_FIRST,
-		SSAO_MINIFY_MIPMAP,
-		SSAO_GATHER_LOW,
-		SSAO_GATHER_MEDIUM,
-		SSAO_GATHER_HIGH,
-		SSAO_GATHER_ULTRA,
-		SSAO_GATHER_LOW_HALF,
-		SSAO_GATHER_MEDIUM_HALF,
-		SSAO_GATHER_HIGH_HALF,
-		SSAO_GATHER_ULTRA_HALF,
+		SSAO_DOWNSAMPLE,
+		SSAO_DOWNSAMPLE_HALF_RES,
+		SSAO_DOWNSAMPLE_MIPMAP,
+		SSAO_DOWNSAMPLE_MIPMAP_HALF_RES,
+		SSAO_DOWNSAMPLE_HALF,
+		SSAO_DOWNSAMPLE_HALF_RES_HALF,
+		SSAO_GATHER,
+		SSAO_GATHER_BASE,
+		SSAO_GATHER_ADAPTIVE,
+		SSAO_GENERATE_IMPORTANCE_MAP,
+		SSAO_PROCESS_IMPORTANCE_MAPA,
+		SSAO_PROCESS_IMPORTANCE_MAPB,
 		SSAO_BLUR_PASS,
-		SSAO_BLUR_PASS_HALF,
-		SSAO_BLUR_UPSCALE,
+		SSAO_BLUR_PASS_SMART,
+		SSAO_BLUR_PASS_WIDE,
+		SSAO_INTERLEAVE,
+		SSAO_INTERLEAVE_SMART,
+		SSAO_INTERLEAVE_HALF,
 		SSAO_MAX
 	};
 
-	struct SSAOMinifyPushConstant {
+	struct SSAODownsamplePushConstant {
 		float pixel_size[2];
 		float z_far;
 		float z_near;
-		int32_t source_size[2];
 		uint32_t orthogonal;
-		uint32_t pad;
+		float radius_sq;
+		uint32_t pad[2];
 	};
 
 	struct SSAOGatherPushConstant {
 		int32_t screen_size[2];
-		float z_far;
-		float z_near;
+		int pass;
+		int quality;
+
+		float half_screen_pixel_size[2];
+		int size_multiplier;
+		float detail_intensity;
+
+		float NDC_to_view_mul[2];
+		float NDC_to_view_add[2];
+
+		float pad[2];
+		float half_screen_pixel_size_x025[2];
 
-		uint32_t orthogonal;
-		float intensity_div_r6;
 		float radius;
-		float bias;
+		float intensity;
+		float shadow_power;
+		float shadow_clamp;
 
-		float proj_info[4];
-		float pixel_size[2];
-		float proj_scale;
-		uint32_t pad;
+		float fade_out_mul;
+		float fade_out_add;
+		float horizon_angle_threshold;
+		float inv_radius_near_limit;
+
+		bool is_orthogonal;
+		float neg_inv_radius;
+		float load_counter_avg_div;
+		float adaptive_sample_limit;
+
+		int32_t pass_coord_offset[2];
+		float pass_uv_offset[2];
+	};
+
+	struct SSAOGatherConstants {
+		float rotation_matrices[80]; //5 vec4s * 4
+	};
+
+	struct SSAOImportanceMapPushConstant {
+		float half_screen_pixel_size[2];
+		float intensity;
+		float power;
 	};
 
 	struct SSAOBlurPushConstant {
 		float edge_sharpness;
-		int32_t filter_scale;
-		float z_far;
-		float z_near;
-		uint32_t orthogonal;
-		uint32_t pad[3];
-		int32_t axis[2];
-		int32_t screen_size[2];
+		float pad;
+		float half_screen_pixel_size[2];
+	};
+
+	struct SSAOInterleavePushConstant {
+		float inv_sharpness;
+		uint32_t size_modifier;
+		float pixel_size[2];
 	};
 
 	struct SSAO {
-		SSAOMinifyPushConstant minify_push_constant;
-		SsaoMinifyShaderRD minify_shader;
-		RID minify_shader_version;
+		SSAODownsamplePushConstant downsample_push_constant;
+		SsaoDownsampleShaderRD downsample_shader;
+		RID downsample_shader_version;
+		RID downsample_uniform_set;
 
 		SSAOGatherPushConstant gather_push_constant;
 		SsaoShaderRD gather_shader;
 		RID gather_shader_version;
+		RID gather_uniform_set;
+		RID gather_constants_buffer;
+		bool gather_initialized = false;
+
+		SSAOImportanceMapPushConstant importance_map_push_constant;
+		SsaoImportanceMapShaderRD importance_map_shader;
+		RID importance_map_shader_version;
+		RID importance_map_load_counter;
+		RID importance_map_uniform_set;
+		RID counter_uniform_set;
 
 		SSAOBlurPushConstant blur_push_constant;
 		SsaoBlurShaderRD blur_shader;
 		RID blur_shader_version;
 
+		SSAOInterleavePushConstant interleave_push_constant;
+		SsaoInterleaveShaderRD interleave_shader;
+		RID interleave_shader_version;
+
+		RID mirror_sampler;
 		RID pipelines[SSAO_MAX];
 	} ssao;
 
@@ -598,13 +650,27 @@ class EffectsRD {
 		}
 	};
 
+	struct TextureSamplerPair {
+		RID texture;
+		RID sampler;
+		_FORCE_INLINE_ bool operator<(const TextureSamplerPair &p_pair) const {
+			if (texture == p_pair.texture) {
+				return sampler < p_pair.sampler;
+			} else {
+				return texture < p_pair.texture;
+			}
+		}
+	};
+
 	Map<RID, RID> texture_to_compute_uniform_set_cache;
 	Map<TexturePair, RID> texture_pair_to_compute_uniform_set_cache;
 	Map<TexturePair, RID> image_pair_to_compute_uniform_set_cache;
+	Map<TextureSamplerPair, RID> texture_sampler_to_compute_uniform_set_cache;
 
 	RID _get_uniform_set_from_image(RID p_texture);
 	RID _get_uniform_set_from_texture(RID p_texture, bool p_use_mipmaps = false);
 	RID _get_compute_uniform_set_from_texture(RID p_texture, bool p_use_mipmaps = false);
+	RID _get_compute_uniform_set_from_texture_and_sampler(RID p_texture, RID p_sampler);
 	RID _get_compute_uniform_set_from_texture_pair(RID p_texture, RID p_texture2, bool p_use_mipmaps = false);
 	RID _get_compute_uniform_set_from_image_pair(RID p_texture, RID p_texture2);
 
@@ -664,9 +730,30 @@ public:
 		Vector2i texture_size;
 	};
 
+	struct SSAOSettings {
+		float radius = 1.0;
+		float intensity = 2.0;
+		float power = 1.5;
+		float detail = 0.5;
+		float horizon = 0.06;
+		float sharpness = 0.98;
+
+		RS::EnvironmentSSAOQuality quality = RS::ENV_SSAO_QUALITY_MEDIUM;
+		bool half_size = false;
+		float adaptive_target = 0.5;
+		int blur_passes = 2;
+		float fadeout_from = 50.0;
+		float fadeout_to = 300.0;
+
+		Size2i screen_size = Size2i();
+		Size2i half_screen_size = Size2i();
+		Size2i quarter_size = Size2i();
+	};
+
 	void tonemapper(RID p_source_color, RID p_dst_framebuffer, const TonemapSettings &p_settings);
 
-	void generate_ssao(RID p_depth_buffer, RID p_normal_buffer, const Size2i &p_depth_buffer_size, RID p_depth_mipmaps_texture, const Vector<RID> &depth_mipmaps, RID p_ao1, bool p_half_size, RID p_ao2, RID p_upscale_buffer, float p_intensity, float p_radius, float p_bias, const CameraMatrix &p_projection, RS::EnvironmentSSAOQuality p_quality, RS::EnvironmentSSAOBlur p_blur, float p_edge_sharpness);
+	void gather_ssao(RD::ComputeListID p_compute_list, const Vector<RID> p_ao_slices, const SSAOSettings &p_settings, bool p_adaptive_base_pass);
+	void generate_ssao(RID p_depth_buffer, RID p_normal_buffer, RID p_depth_mipmaps_texture, const Vector<RID> &depth_mipmaps, RID p_ao, const Vector<RID> p_ao_slices, RID p_ao_pong, const Vector<RID> p_ao_pong_slices, RID p_upscale_buffer, RID p_importance_map, RID p_importance_map_pong, const CameraMatrix &p_projection, const SSAOSettings &p_settings, bool p_invalidate_uniform_sets);
 
 	void roughness_limit(RID p_source_normal, RID p_roughness, const Size2i &p_size, float p_curve);
 	void cubemap_downsample(RID p_source_cubemap, RID p_dest_cubemap, const Size2i &p_size);
diff --git a/servers/rendering/renderer_rd/renderer_scene_render_forward.cpp b/servers/rendering/renderer_rd/renderer_scene_render_forward.cpp
index 93f4b7bc8e..01dd2965c4 100644
--- a/servers/rendering/renderer_rd/renderer_scene_render_forward.cpp
+++ b/servers/rendering/renderer_rd/renderer_scene_render_forward.cpp
@@ -766,7 +766,7 @@ void RendererSceneRenderForward::_allocate_normal_roughness_texture(RenderBuffer
 	tf.format = RD::DATA_FORMAT_R8G8B8A8_UNORM;
 	tf.width = rb->width;
 	tf.height = rb->height;
-	tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT;
+	tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
 
 	if (rb->msaa != RS::VIEWPORT_MSAA_DISABLED) {
 		tf.usage_bits |= RD::TEXTURE_USAGE_CAN_COPY_TO_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
@@ -782,7 +782,7 @@ void RendererSceneRenderForward::_allocate_normal_roughness_texture(RenderBuffer
 		fb.push_back(rb->normal_roughness_buffer);
 		rb->depth_normal_roughness_fb = RD::get_singleton()->framebuffer_create(fb);
 	} else {
-		tf.usage_bits = RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | RD::TEXTURE_USAGE_CAN_COPY_FROM_BIT | RD::TEXTURE_USAGE_SAMPLING_BIT;
+		tf.usage_bits = RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | RD::TEXTURE_USAGE_CAN_COPY_FROM_BIT | RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
 		tf.samples = rb->texture_samples;
 		rb->normal_roughness_buffer_msaa = RD::get_singleton()->texture_create(tf, RD::TextureView());
 
diff --git a/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp b/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp
index 4543ced8dc..1d9983a28c 100644
--- a/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp
+++ b/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp
@@ -3122,7 +3122,7 @@ RS::EnvironmentSSRRoughnessQuality RendererSceneRenderRD::environment_get_ssr_ro
 	return ssr_roughness_quality;
 }
 
-void RendererSceneRenderRD::environment_set_ssao(RID p_env, bool p_enable, float p_radius, float p_intensity, float p_bias, float p_light_affect, float p_ao_channel_affect, RS::EnvironmentSSAOBlur p_blur, float p_bilateral_sharpness) {
+void RendererSceneRenderRD::environment_set_ssao(RID p_env, bool p_enable, float p_radius, float p_intensity, float p_power, float p_detail, float p_horizon, float p_sharpness, float p_light_affect, float p_ao_channel_affect) {
 	Environment *env = environment_owner.getornull(p_env);
 	ERR_FAIL_COND(!env);
 
@@ -3133,15 +3133,21 @@ void RendererSceneRenderRD::environment_set_ssao(RID p_env, bool p_enable, float
 	env->ssao_enabled = p_enable;
 	env->ssao_radius = p_radius;
 	env->ssao_intensity = p_intensity;
-	env->ssao_bias = p_bias;
+	env->ssao_power = p_power;
+	env->ssao_detail = p_detail;
+	env->ssao_horizon = p_horizon;
+	env->ssao_sharpness = p_sharpness;
 	env->ssao_direct_light_affect = p_light_affect;
 	env->ssao_ao_channel_affect = p_ao_channel_affect;
-	env->ssao_blur = p_blur;
 }
 
-void RendererSceneRenderRD::environment_set_ssao_quality(RS::EnvironmentSSAOQuality p_quality, bool p_half_size) {
+void RendererSceneRenderRD::environment_set_ssao_quality(RS::EnvironmentSSAOQuality p_quality, bool p_half_size, float p_adaptive_target, int p_blur_passes, float p_fadeout_from, float p_fadeout_to) {
 	ssao_quality = p_quality;
 	ssao_half_size = p_half_size;
+	ssao_adaptive_target = p_adaptive_target;
+	ssao_blur_passes = p_blur_passes;
+	ssao_fadeout_from = p_fadeout_from;
+	ssao_fadeout_to = p_fadeout_to;
 }
 
 bool RendererSceneRenderRD::environment_is_ssao_enabled(RID p_env) const {
@@ -5081,21 +5087,24 @@ void RendererSceneRenderRD::_free_render_buffer_data(RenderBuffers *rb) {
 		rb->luminance.current = RID();
 	}
 
-	if (rb->ssao.ao[0].is_valid()) {
+	if (rb->ssao.depth.is_valid()) {
 		RD::get_singleton()->free(rb->ssao.depth);
-		RD::get_singleton()->free(rb->ssao.ao[0]);
-		if (rb->ssao.ao[1].is_valid()) {
-			RD::get_singleton()->free(rb->ssao.ao[1]);
-		}
-		if (rb->ssao.ao_full.is_valid()) {
-			RD::get_singleton()->free(rb->ssao.ao_full);
-		}
+		RD::get_singleton()->free(rb->ssao.ao_deinterleaved);
+		RD::get_singleton()->free(rb->ssao.ao_pong);
+		RD::get_singleton()->free(rb->ssao.ao_final);
+
+		RD::get_singleton()->free(rb->ssao.importance_map[0]);
+		RD::get_singleton()->free(rb->ssao.importance_map[1]);
 
 		rb->ssao.depth = RID();
-		rb->ssao.ao[0] = RID();
-		rb->ssao.ao[1] = RID();
-		rb->ssao.ao_full = RID();
+		rb->ssao.ao_deinterleaved = RID();
+		rb->ssao.ao_pong = RID();
+		rb->ssao.ao_final = RID();
+		rb->ssao.importance_map[0] = RID();
+		rb->ssao.importance_map[1] = RID();
 		rb->ssao.depth_slices.clear();
+		rb->ssao.ao_deinterleaved_slices.clear();
+		rb->ssao.ao_pong_slices.clear();
 	}
 
 	if (rb->ssr.blur_radius[0].is_valid()) {
@@ -5194,64 +5203,133 @@ void RendererSceneRenderRD::_process_ssao(RID p_render_buffers, RID p_environmen
 
 	RENDER_TIMESTAMP("Process SSAO");
 
-	if (rb->ssao.ao[0].is_valid() && rb->ssao.ao_full.is_valid() != ssao_half_size) {
+	//TODO clear when settings chenge to or from ultra
+	if (rb->ssao.ao_final.is_valid() && ssao_using_half_size != ssao_half_size) {
 		RD::get_singleton()->free(rb->ssao.depth);
-		RD::get_singleton()->free(rb->ssao.ao[0]);
-		if (rb->ssao.ao[1].is_valid()) {
-			RD::get_singleton()->free(rb->ssao.ao[1]);
-		}
-		if (rb->ssao.ao_full.is_valid()) {
-			RD::get_singleton()->free(rb->ssao.ao_full);
-		}
+		RD::get_singleton()->free(rb->ssao.ao_deinterleaved);
+		RD::get_singleton()->free(rb->ssao.ao_pong);
+		RD::get_singleton()->free(rb->ssao.ao_final);
+
+		RD::get_singleton()->free(rb->ssao.importance_map[0]);
+		RD::get_singleton()->free(rb->ssao.importance_map[1]);
 
 		rb->ssao.depth = RID();
-		rb->ssao.ao[0] = RID();
-		rb->ssao.ao[1] = RID();
-		rb->ssao.ao_full = RID();
+		rb->ssao.ao_deinterleaved = RID();
+		rb->ssao.ao_pong = RID();
+		rb->ssao.ao_final = RID();
+		rb->ssao.importance_map[0] = RID();
+		rb->ssao.importance_map[1] = RID();
 		rb->ssao.depth_slices.clear();
+		rb->ssao.ao_deinterleaved_slices.clear();
+		rb->ssao.ao_pong_slices.clear();
+	}
+
+	int buffer_width;
+	int buffer_height;
+	int half_width;
+	int half_height;
+	if (ssao_half_size) {
+		buffer_width = (rb->width + 3) / 4;
+		buffer_height = (rb->height + 3) / 4;
+		half_width = (rb->width + 7) / 8;
+		half_height = (rb->height + 7) / 8;
+	} else {
+		buffer_width = (rb->width + 1) / 2;
+		buffer_height = (rb->height + 1) / 2;
+		half_width = (rb->width + 3) / 4;
+		half_height = (rb->height + 3) / 4;
 	}
-
-	if (!rb->ssao.ao[0].is_valid()) {
+	bool uniform_sets_are_invalid = false;
+	if (rb->ssao.depth.is_null()) {
 		//allocate depth slices
 
 		{
 			RD::TextureFormat tf;
-			tf.format = RD::DATA_FORMAT_R32_SFLOAT;
-			tf.width = rb->width / 2;
-			tf.height = rb->height / 2;
-			tf.mipmaps = Image::get_image_required_mipmaps(tf.width, tf.height, Image::FORMAT_RF) + 1;
+			tf.format = RD::DATA_FORMAT_R16_SFLOAT;
+			tf.texture_type = RD::TEXTURE_TYPE_2D_ARRAY;
+			tf.width = buffer_width;
+			tf.height = buffer_height;
+			tf.mipmaps = 4;
+			tf.array_layers = 4;
 			tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
 			rb->ssao.depth = RD::get_singleton()->texture_create(tf, RD::TextureView());
 			for (uint32_t i = 0; i < tf.mipmaps; i++) {
-				RID slice = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), rb->ssao.depth, 0, i);
+				RID slice = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), rb->ssao.depth, 0, i, RD::TEXTURE_SLICE_2D_ARRAY);
 				rb->ssao.depth_slices.push_back(slice);
 			}
 		}
 
 		{
 			RD::TextureFormat tf;
-			tf.format = RD::DATA_FORMAT_R8_UNORM;
-			tf.width = ssao_half_size ? rb->width / 2 : rb->width;
-			tf.height = ssao_half_size ? rb->height / 2 : rb->height;
+			tf.format = RD::DATA_FORMAT_R8G8_UNORM;
+			tf.texture_type = RD::TEXTURE_TYPE_2D_ARRAY;
+			tf.width = buffer_width;
+			tf.height = buffer_height;
+			tf.array_layers = 4;
+			tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
+			rb->ssao.ao_deinterleaved = RD::get_singleton()->texture_create(tf, RD::TextureView());
+			for (uint32_t i = 0; i < 4; i++) {
+				RID slice = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), rb->ssao.ao_deinterleaved, i, 0);
+				rb->ssao.ao_deinterleaved_slices.push_back(slice);
+			}
+		}
+
+		{
+			RD::TextureFormat tf;
+			tf.format = RD::DATA_FORMAT_R8G8_UNORM;
+			tf.texture_type = RD::TEXTURE_TYPE_2D_ARRAY;
+			tf.width = buffer_width;
+			tf.height = buffer_height;
+			tf.array_layers = 4;
 			tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
-			rb->ssao.ao[0] = RD::get_singleton()->texture_create(tf, RD::TextureView());
-			rb->ssao.ao[1] = RD::get_singleton()->texture_create(tf, RD::TextureView());
+			rb->ssao.ao_pong = RD::get_singleton()->texture_create(tf, RD::TextureView());
+			for (uint32_t i = 0; i < 4; i++) {
+				RID slice = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), rb->ssao.ao_pong, i, 0);
+				rb->ssao.ao_pong_slices.push_back(slice);
+			}
 		}
 
-		if (ssao_half_size) {
-			//upsample texture
+		{
+			RD::TextureFormat tf;
+			tf.format = RD::DATA_FORMAT_R8_UNORM;
+			tf.width = half_width;
+			tf.height = half_height;
+			tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
+			rb->ssao.importance_map[0] = RD::get_singleton()->texture_create(tf, RD::TextureView());
+			rb->ssao.importance_map[1] = RD::get_singleton()->texture_create(tf, RD::TextureView());
+		}
+		{
 			RD::TextureFormat tf;
 			tf.format = RD::DATA_FORMAT_R8_UNORM;
 			tf.width = rb->width;
 			tf.height = rb->height;
 			tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
-			rb->ssao.ao_full = RD::get_singleton()->texture_create(tf, RD::TextureView());
+			rb->ssao.ao_final = RD::get_singleton()->texture_create(tf, RD::TextureView());
+			_render_buffers_uniform_set_changed(p_render_buffers);
 		}
-
-		_render_buffers_uniform_set_changed(p_render_buffers);
+		ssao_using_half_size = ssao_half_size;
+		uniform_sets_are_invalid = true;
 	}
 
-	storage->get_effects()->generate_ssao(rb->depth_texture, p_normal_buffer, Size2i(rb->width, rb->height), rb->ssao.depth, rb->ssao.depth_slices, rb->ssao.ao[0], rb->ssao.ao_full.is_valid(), rb->ssao.ao[1], rb->ssao.ao_full, env->ssao_intensity, env->ssao_radius, env->ssao_bias, p_projection, ssao_quality, env->ssao_blur, env->ssao_blur_edge_sharpness);
+	EffectsRD::SSAOSettings settings;
+	settings.radius = env->ssao_radius;
+	settings.intensity = env->ssao_intensity;
+	settings.power = env->ssao_power;
+	settings.detail = env->ssao_detail;
+	settings.horizon = env->ssao_horizon;
+	settings.sharpness = env->ssao_sharpness;
+
+	settings.quality = ssao_quality;
+	settings.half_size = ssao_half_size;
+	settings.adaptive_target = ssao_adaptive_target;
+	settings.blur_passes = ssao_blur_passes;
+	settings.fadeout_from = ssao_fadeout_from;
+	settings.fadeout_to = ssao_fadeout_to;
+	settings.screen_size = Size2i(rb->width, rb->height);
+	settings.half_screen_size = Size2i(buffer_width, buffer_height);
+	settings.quarter_size = Size2i(half_width, half_height);
+
+	storage->get_effects()->generate_ssao(rb->depth_texture, p_normal_buffer, rb->ssao.depth, rb->ssao.depth_slices, rb->ssao.ao_deinterleaved, rb->ssao.ao_deinterleaved_slices, rb->ssao.ao_pong, rb->ssao.ao_pong_slices, rb->ssao.ao_final, rb->ssao.importance_map[0], rb->ssao.importance_map[1], p_projection, settings, uniform_sets_are_invalid);
 }
 
 void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(RID p_render_buffers, RID p_environment, RID p_camera_effects, const CameraMatrix &p_projection) {
@@ -5431,9 +5509,9 @@ void RendererSceneRenderRD::_render_buffers_debug_draw(RID p_render_buffers, RID
 		}
 	}
 
-	if (debug_draw == RS::VIEWPORT_DEBUG_DRAW_SSAO && rb->ssao.ao[0].is_valid()) {
+	if (debug_draw == RS::VIEWPORT_DEBUG_DRAW_SSAO && rb->ssao.ao_final.is_valid()) {
 		Size2 rtsize = storage->render_target_get_size(rb->render_target);
-		RID ao_buf = rb->ssao.ao_full.is_valid() ? rb->ssao.ao_full : rb->ssao.ao[0];
+		RID ao_buf = rb->ssao.ao_final;
 		effects->copy_to_fb_rect(ao_buf, storage->render_target_get_rd_framebuffer(rb->render_target), Rect2(Vector2(), rtsize), false, true);
 	}
 
@@ -5621,7 +5699,7 @@ RID RendererSceneRenderRD::render_buffers_get_ao_texture(RID p_render_buffers) {
 	RenderBuffers *rb = render_buffers_owner.getornull(p_render_buffers);
 	ERR_FAIL_COND_V(!rb, RID());
 
-	return rb->ssao.ao_full.is_valid() ? rb->ssao.ao_full : rb->ssao.ao[0];
+	return rb->ssao.ao_final;
 }
 
 RID RendererSceneRenderRD::render_buffers_get_gi_probe_buffer(RID p_render_buffers) {
@@ -8435,7 +8513,7 @@ RendererSceneRenderRD::RendererSceneRenderRD(RendererStorageRD *p_storage) {
 
 	camera_effects_set_dof_blur_bokeh_shape(RS::DOFBokehShape(int(GLOBAL_GET("rendering/quality/depth_of_field/depth_of_field_bokeh_shape"))));
 	camera_effects_set_dof_blur_quality(RS::DOFBlurQuality(int(GLOBAL_GET("rendering/quality/depth_of_field/depth_of_field_bokeh_quality"))), GLOBAL_GET("rendering/quality/depth_of_field/depth_of_field_use_jitter"));
-	environment_set_ssao_quality(RS::EnvironmentSSAOQuality(int(GLOBAL_GET("rendering/quality/ssao/quality"))), GLOBAL_GET("rendering/quality/ssao/half_size"));
+	environment_set_ssao_quality(RS::EnvironmentSSAOQuality(int(GLOBAL_GET("rendering/quality/ssao/quality"))), GLOBAL_GET("rendering/quality/ssao/half_size"), GLOBAL_GET("rendering/quality/ssao/adaptive_target"), GLOBAL_GET("rendering/quality/ssao/blur_passes"), GLOBAL_GET("rendering/quality/ssao/fadeout_from"), GLOBAL_GET("rendering/quality/ssao/fadeout_to"));
 	screen_space_roughness_limiter = GLOBAL_GET("rendering/quality/screen_filters/screen_space_roughness_limiter_enabled");
 	screen_space_roughness_limiter_amount = GLOBAL_GET("rendering/quality/screen_filters/screen_space_roughness_limiter_amount");
 	screen_space_roughness_limiter_limit = GLOBAL_GET("rendering/quality/screen_filters/screen_space_roughness_limiter_limit");
diff --git a/servers/rendering/renderer_rd/renderer_scene_render_rd.h b/servers/rendering/renderer_rd/renderer_scene_render_rd.h
index 3afe6e3f4a..1b0ae61042 100644
--- a/servers/rendering/renderer_rd/renderer_scene_render_rd.h
+++ b/servers/rendering/renderer_rd/renderer_scene_render_rd.h
@@ -737,13 +737,14 @@ private:
 		/// SSAO
 
 		bool ssao_enabled = false;
-		float ssao_radius = 1;
-		float ssao_intensity = 1;
-		float ssao_bias = 0.01;
+		float ssao_radius = 1.0;
+		float ssao_intensity = 2.0;
+		float ssao_power = 1.5;
+		float ssao_detail = 0.5;
+		float ssao_horizon = 0.06;
+		float ssao_sharpness = 0.98;
 		float ssao_direct_light_affect = 0.0;
 		float ssao_ao_channel_affect = 0.0;
-		float ssao_blur_edge_sharpness = 4.0;
-		RS::EnvironmentSSAOBlur ssao_blur = RS::ENV_SSAO_BLUR_3x3;
 
 		/// SSR
 		///
@@ -777,6 +778,12 @@ private:
 
 	RS::EnvironmentSSAOQuality ssao_quality = RS::ENV_SSAO_QUALITY_MEDIUM;
 	bool ssao_half_size = false;
+	bool ssao_using_half_size = false;
+	float ssao_adaptive_target = 0.5;
+	int ssao_blur_passes = 2;
+	float ssao_fadeout_from = 50.0;
+	float ssao_fadeout_to = 300.0;
+
 	bool glow_bicubic_upscale = false;
 	bool glow_high_quality = false;
 	RS::EnvironmentSSRRoughnessQuality ssr_roughness_quality = RS::ENV_SSR_ROUGNESS_QUALITY_LOW;
@@ -861,8 +868,12 @@ private:
 		struct SSAO {
 			RID depth;
 			Vector<RID> depth_slices;
-			RID ao[2];
-			RID ao_full; //when using half-size
+			RID ao_deinterleaved;
+			Vector<RID> ao_deinterleaved_slices;
+			RID ao_pong;
+			Vector<RID> ao_pong_slices;
+			RID ao_final;
+			RID importance_map[2];
 		} ssao;
 
 		struct SSR {
@@ -1567,8 +1578,8 @@ public:
 	virtual void environment_set_volumetric_fog_positional_shadow_shrink_size(int p_shrink_size);
 
 	void environment_set_ssr(RID p_env, bool p_enable, int p_max_steps, float p_fade_int, float p_fade_out, float p_depth_tolerance);
-	void environment_set_ssao(RID p_env, bool p_enable, float p_radius, float p_intensity, float p_bias, float p_light_affect, float p_ao_channel_affect, RS::EnvironmentSSAOBlur p_blur, float p_bilateral_sharpness);
-	void environment_set_ssao_quality(RS::EnvironmentSSAOQuality p_quality, bool p_half_size);
+	void environment_set_ssao(RID p_env, bool p_enable, float p_radius, float p_intensity, float p_power, float p_detail, float p_horizon, float p_sharpness, float p_light_affect, float p_ao_channel_affect);
+	void environment_set_ssao_quality(RS::EnvironmentSSAOQuality p_quality, bool p_half_size, float p_adaptive_target, int p_blur_passes, float p_fadeout_from, float p_fadeout_to);
 	bool environment_is_ssao_enabled(RID p_env) const;
 	float environment_get_ssao_ao_affect(RID p_env) const;
 	float environment_get_ssao_light_affect(RID p_env) const;
diff --git a/servers/rendering/renderer_rd/shaders/SCsub b/servers/rendering/renderer_rd/shaders/SCsub
index cb62882deb..deaa9668df 100644
--- a/servers/rendering/renderer_rd/shaders/SCsub
+++ b/servers/rendering/renderer_rd/shaders/SCsub
@@ -21,8 +21,10 @@ if "RD_GLSL" in env["BUILDERS"]:
     env.RD_GLSL("luminance_reduce.glsl")
     env.RD_GLSL("bokeh_dof.glsl")
     env.RD_GLSL("ssao.glsl")
-    env.RD_GLSL("ssao_minify.glsl")
+    env.RD_GLSL("ssao_downsample.glsl")
+    env.RD_GLSL("ssao_importance_map.glsl")
     env.RD_GLSL("ssao_blur.glsl")
+    env.RD_GLSL("ssao_interleave.glsl")
     env.RD_GLSL("roughness_limiter.glsl")
     env.RD_GLSL("screen_space_reflection.glsl")
     env.RD_GLSL("screen_space_reflection_filter.glsl")
diff --git a/servers/rendering/renderer_rd/shaders/ssao.glsl b/servers/rendering/renderer_rd/shaders/ssao.glsl
index 346338181a..f67965ab49 100644
--- a/servers/rendering/renderer_rd/shaders/ssao.glsl
+++ b/servers/rendering/renderer_rd/shaders/ssao.glsl
@@ -1,249 +1,491 @@
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// Copyright (c) 2016, Intel Corporation
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated
+// documentation files (the "Software"), to deal in the Software without restriction, including without limitation
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to
+// permit persons to whom the Software is furnished to do so, subject to the following conditions:
+// The above copyright notice and this permission notice shall be included in all copies or substantial portions of
+// the Software.
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+// TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// File changes (yyyy-mm-dd)
+// 2016-09-07: filip.strugar@intel.com: first commit
+// 2020-12-05: clayjohn: convert to Vulkan and Godot
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+
 #[compute]
 
 #version 450
 
 VERSION_DEFINES
 
-layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+#define SSAO_ADAPTIVE_TAP_BASE_COUNT 5
+
+#define INTELSSAO_MAIN_DISK_SAMPLE_COUNT (32)
+const vec4 sample_pattern[INTELSSAO_MAIN_DISK_SAMPLE_COUNT] = {
+	vec4(0.78488064, 0.56661671, 1.500000, -0.126083), vec4(0.26022232, -0.29575172, 1.500000, -1.064030), vec4(0.10459357, 0.08372527, 1.110000, -2.730563), vec4(-0.68286800, 0.04963045, 1.090000, -0.498827),
+	vec4(-0.13570161, -0.64190155, 1.250000, -0.532765), vec4(-0.26193795, -0.08205118, 0.670000, -1.783245), vec4(-0.61177456, 0.66664219, 0.710000, -0.044234), vec4(0.43675563, 0.25119025, 0.610000, -1.167283),
+	vec4(0.07884444, 0.86618668, 0.640000, -0.459002), vec4(-0.12790935, -0.29869005, 0.600000, -1.729424), vec4(-0.04031125, 0.02413622, 0.600000, -4.792042), vec4(0.16201244, -0.52851415, 0.790000, -1.067055),
+	vec4(-0.70991218, 0.47301072, 0.640000, -0.335236), vec4(0.03277707, -0.22349690, 0.600000, -1.982384), vec4(0.68921727, 0.36800742, 0.630000, -0.266718), vec4(0.29251814, 0.37775412, 0.610000, -1.422520),
+	vec4(-0.12224089, 0.96582592, 0.600000, -0.426142), vec4(0.11071457, -0.16131058, 0.600000, -2.165947), vec4(0.46562141, -0.59747696, 0.600000, -0.189760), vec4(-0.51548797, 0.11804193, 0.600000, -1.246800),
+	vec4(0.89141309, -0.42090443, 0.600000, 0.028192), vec4(-0.32402530, -0.01591529, 0.600000, -1.543018), vec4(0.60771245, 0.41635221, 0.600000, -0.605411), vec4(0.02379565, -0.08239821, 0.600000, -3.809046),
+	vec4(0.48951152, -0.23657045, 0.600000, -1.189011), vec4(-0.17611565, -0.81696892, 0.600000, -0.513724), vec4(-0.33930185, -0.20732205, 0.600000, -1.698047), vec4(-0.91974425, 0.05403209, 0.600000, 0.062246),
+	vec4(-0.15064627, -0.14949332, 0.600000, -1.896062), vec4(0.53180975, -0.35210401, 0.600000, -0.758838), vec4(0.41487166, 0.81442589, 0.600000, -0.505648), vec4(-0.24106961, -0.32721516, 0.600000, -1.665244)
+};
+
+// these values can be changed (up to SSAO_MAX_TAPS) with no changes required elsewhere; values for 4th and 5th preset are ignored but array needed to avoid compilation errors
+// the actual number of texture samples is two times this value (each "tap" has two symmetrical depth texture samples)
+const int num_taps[5] = { 3, 5, 12, 0, 0 };
+
+#define SSAO_TILT_SAMPLES_ENABLE_AT_QUALITY_PRESET (99) // to disable simply set to 99 or similar
+#define SSAO_TILT_SAMPLES_AMOUNT (0.4)
+//
+#define SSAO_HALOING_REDUCTION_ENABLE_AT_QUALITY_PRESET (1) // to disable simply set to 99 or similar
+#define SSAO_HALOING_REDUCTION_AMOUNT (0.6) // values from 0.0 - 1.0, 1.0 means max weighting (will cause artifacts, 0.8 is more reasonable)
+//
+#define SSAO_NORMAL_BASED_EDGES_ENABLE_AT_QUALITY_PRESET (2) // to disable simply set to 99 or similar
+#define SSAO_NORMAL_BASED_EDGES_DOT_THRESHOLD (0.5) // use 0-0.1 for super-sharp normal-based edges
+//
+#define SSAO_DETAIL_AO_ENABLE_AT_QUALITY_PRESET (1) // whether to use detail; to disable simply set to 99 or similar
+//
+#define SSAO_DEPTH_MIPS_ENABLE_AT_QUALITY_PRESET (2) // !!warning!! the MIP generation on the C++ side will be enabled on quality preset 2 regardless of this value, so if changing here, change the C++ side too
+#define SSAO_DEPTH_MIPS_GLOBAL_OFFSET (-4.3) // best noise/quality/performance tradeoff, found empirically
+//
+// !!warning!! the edge handling is hard-coded to 'disabled' on quality level 0, and enabled above, on the C++ side; while toggling it here will work for
+// testing purposes, it will not yield performance gains (or correct results)
+#define SSAO_DEPTH_BASED_EDGES_ENABLE_AT_QUALITY_PRESET (1)
+//
+#define SSAO_REDUCE_RADIUS_NEAR_SCREEN_BORDER_ENABLE_AT_QUALITY_PRESET (1)
+
+#define SSAO_MAX_TAPS 32
+#define SSAO_MAX_REF_TAPS 512
+#define SSAO_ADAPTIVE_TAP_BASE_COUNT 5
+#define SSAO_ADAPTIVE_TAP_FLEXIBLE_COUNT (SSAO_MAX_TAPS - SSAO_ADAPTIVE_TAP_BASE_COUNT)
+#define SSAO_DEPTH_MIP_LEVELS 4
 
-#define TWO_PI 6.283185307179586476925286766559
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
 
-#ifdef SSAO_QUALITY_HIGH
-#define NUM_SAMPLES (20)
-#endif
+layout(set = 0, binding = 0) uniform sampler2DArray source_depth_mipmaps;
+layout(rgba8, set = 0, binding = 1) uniform restrict readonly image2D source_normal;
+layout(set = 0, binding = 2) uniform Constants { //get into a lower set
+	vec4 rotation_matrices[20];
+}
+constants;
 
-#ifdef SSAO_QUALITY_ULTRA
-#define NUM_SAMPLES (48)
+#ifdef ADAPTIVE
+layout(rg8, set = 1, binding = 0) uniform restrict readonly image2DArray source_ssao;
+layout(set = 1, binding = 1) uniform sampler2D source_importance;
+layout(set = 1, binding = 2, std430) buffer Counter {
+	uint sum;
+}
+counter;
 #endif
 
-#ifdef SSAO_QUALITY_LOW
-#define NUM_SAMPLES (8)
-#endif
+layout(rg8, set = 2, binding = 0) uniform restrict writeonly image2D dest_image;
 
-#if !defined(SSAO_QUALITY_LOW) && !defined(SSAO_QUALITY_HIGH) && !defined(SSAO_QUALITY_ULTRA)
-#define NUM_SAMPLES (12)
-#endif
+// This push_constant is full - 128 bytes - if you need to add more data, consider adding to the uniform buffer instead
+layout(push_constant, binding = 1, std430) uniform Params {
+	ivec2 screen_size;
+	int pass;
+	int quality;
 
-// If using depth mip levels, the log of the maximum pixel offset before we need to switch to a lower
-// miplevel to maintain reasonable spatial locality in the cache
-// If this number is too small (< 3), too many taps will land in the same pixel, and we'll get bad variance that manifests as flashing.
-// If it is too high (> 5), we'll get bad performance because we're not using the MIP levels effectively
-#define LOG_MAX_OFFSET (3)
-
-// This must be less than or equal to the MAX_MIP_LEVEL defined in SSAO.cpp
-#define MAX_MIP_LEVEL (4)
-
-// This is the number of turns around the circle that the spiral pattern makes.  This should be prime to prevent
-// taps from lining up.  This particular choice was tuned for NUM_SAMPLES == 9
-
-const int ROTATIONS[] = int[](
-		1, 1, 2, 3, 2, 5, 2, 3, 2,
-		3, 3, 5, 5, 3, 4, 7, 5, 5, 7,
-		9, 8, 5, 5, 7, 7, 7, 8, 5, 8,
-		11, 12, 7, 10, 13, 8, 11, 8, 7, 14,
-		11, 11, 13, 12, 13, 19, 17, 13, 11, 18,
-		19, 11, 11, 14, 17, 21, 15, 16, 17, 18,
-		13, 17, 11, 17, 19, 18, 25, 18, 19, 19,
-		29, 21, 19, 27, 31, 29, 21, 18, 17, 29,
-		31, 31, 23, 18, 25, 26, 25, 23, 19, 34,
-		19, 27, 21, 25, 39, 29, 17, 21, 27);
-
-//#define NUM_SPIRAL_TURNS (7)
-const int NUM_SPIRAL_TURNS = ROTATIONS[NUM_SAMPLES - 1];
-
-layout(set = 0, binding = 0) uniform sampler2D source_depth_mipmaps;
-layout(r8, set = 1, binding = 0) uniform restrict writeonly image2D dest_image;
-
-#ifndef USE_HALF_SIZE
-layout(set = 2, binding = 0) uniform sampler2D source_depth;
-#endif
+	vec2 half_screen_pixel_size;
+	int size_multiplier;
+	float detail_intensity;
 
-layout(set = 3, binding = 0) uniform sampler2D source_normal;
+	vec2 NDC_to_view_mul;
+	vec2 NDC_to_view_add;
 
-layout(push_constant, binding = 1, std430) uniform Params {
-	ivec2 screen_size;
-	float z_far;
-	float z_near;
+	vec2 pad2;
+	vec2 half_screen_pixel_size_x025;
 
-	bool orthogonal;
-	float intensity_div_r6;
 	float radius;
-	float bias;
-
-	vec4 proj_info;
-	vec2 pixel_size;
-	float proj_scale;
-	uint pad;
+	float intensity;
+	float shadow_power;
+	float shadow_clamp;
+
+	float fade_out_mul;
+	float fade_out_add;
+	float horizon_angle_threshold;
+	float inv_radius_near_limit;
+
+	bool is_orthogonal;
+	float neg_inv_radius;
+	float load_counter_avg_div;
+	float adaptive_sample_limit;
+
+	ivec2 pass_coord_offset;
+	vec2 pass_uv_offset;
 }
 params;
 
-vec3 reconstructCSPosition(vec2 S, float z) {
-	if (params.orthogonal) {
-		return vec3((S.xy * params.proj_info.xy + params.proj_info.zw), z);
+// packing/unpacking for edges; 2 bits per edge mean 4 gradient values (0, 0.33, 0.66, 1) for smoother transitions!
+float pack_edges(vec4 p_edgesLRTB) {
+	p_edgesLRTB = round(clamp(p_edgesLRTB, 0.0, 1.0) * 3.05);
+	return dot(p_edgesLRTB, vec4(64.0 / 255.0, 16.0 / 255.0, 4.0 / 255.0, 1.0 / 255.0));
+}
+
+vec3 NDC_to_view_space(vec2 p_pos, float p_viewspace_depth) {
+	if (params.is_orthogonal) {
+		return vec3((params.NDC_to_view_mul * p_pos.xy + params.NDC_to_view_add), p_viewspace_depth);
 	} else {
-		return vec3((S.xy * params.proj_info.xy + params.proj_info.zw) * z, z);
+		return vec3((params.NDC_to_view_mul * p_pos.xy + params.NDC_to_view_add) * p_viewspace_depth, p_viewspace_depth);
 	}
 }
 
-vec3 getPosition(ivec2 ssP) {
-	vec3 P;
-#ifdef USE_HALF_SIZE
-	P.z = texelFetch(source_depth_mipmaps, ssP, 0).r;
-	P.z = -P.z;
-#else
-	P.z = texelFetch(source_depth, ssP, 0).r;
+// calculate effect radius and fit our screen sampling pattern inside it
+void calculate_radius_parameters(const float p_pix_center_length, const vec2 p_pixel_size_at_center, out float r_lookup_radius, out float r_radius, out float r_fallof_sq) {
+	r_radius = params.radius;
 
-	P.z = P.z * 2.0 - 1.0;
-	if (params.orthogonal) {
-		P.z = ((P.z + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / 2.0;
-	} else {
-		P.z = 2.0 * params.z_near * params.z_far / (params.z_far + params.z_near - P.z * (params.z_far - params.z_near));
-	}
-	P.z = -P.z;
-#endif
-	// Offset to pixel center
-	P = reconstructCSPosition(vec2(ssP) + vec2(0.5), P.z);
-	return P;
+	// when too close, on-screen sampling disk will grow beyond screen size; limit this to avoid closeup temporal artifacts
+	const float too_close_limit = clamp(p_pix_center_length * params.inv_radius_near_limit, 0.0, 1.0) * 0.8 + 0.2;
+
+	r_radius *= too_close_limit;
+
+	// 0.85 is to reduce the radius to allow for more samples on a slope to still stay within influence
+	r_lookup_radius = (0.85 * r_radius) / p_pixel_size_at_center.x;
+
+	// used to calculate falloff (both for AO samples and per-sample weights)
+	r_fallof_sq = -1.0 / (r_radius * r_radius);
 }
 
-/** Returns a unit vector and a screen-space radius for the tap on a unit disk (the caller should scale by the actual disk radius) */
-vec2 tapLocation(int sampleNumber, float spinAngle, out float ssR) {
-	// Radius relative to ssR
-	float alpha = (float(sampleNumber) + 0.5) * (1.0 / float(NUM_SAMPLES));
-	float angle = alpha * (float(NUM_SPIRAL_TURNS) * 6.28) + spinAngle;
+vec4 calculate_edges(const float p_center_z, const float p_left_z, const float p_right_z, const float p_top_z, const float p_bottom_z) {
+	// slope-sensitive depth-based edge detection
+	vec4 edgesLRTB = vec4(p_left_z, p_right_z, p_top_z, p_bottom_z) - p_center_z;
+	vec4 edgesLRTB_slope_adjusted = edgesLRTB + edgesLRTB.yxwz;
+	edgesLRTB = min(abs(edgesLRTB), abs(edgesLRTB_slope_adjusted));
+	return clamp((1.3 - edgesLRTB / (p_center_z * 0.040)), 0.0, 1.0);
+}
 
-	ssR = alpha;
-	return vec2(cos(angle), sin(angle));
+vec3 decode_normal(vec3 p_encoded_normal) {
+	vec3 normal = p_encoded_normal * 2.0 - 1.0;
+	return normal;
 }
 
-/** Read the camera-space position of the point at screen-space pixel ssP + unitOffset * ssR.  Assumes length(unitOffset) == 1 */
-vec3 getOffsetPosition(ivec2 ssP, float ssR) {
-	// Derivation:
-	//  mipLevel = floor(log(ssR / MAX_OFFSET));
+vec3 load_normal(ivec2 p_pos) {
+	vec3 encoded_normal = imageLoad(source_normal, p_pos).xyz;
+	encoded_normal.z = 1.0 - encoded_normal.z;
+	return decode_normal(encoded_normal);
+}
 
-	int mipLevel = clamp(int(floor(log2(ssR))) - LOG_MAX_OFFSET, 0, MAX_MIP_LEVEL);
+vec3 load_normal(ivec2 p_pos, ivec2 p_offset) {
+	vec3 encoded_normal = imageLoad(source_normal, p_pos + p_offset).xyz;
+	encoded_normal.z = 1.0 - encoded_normal.z;
+	return decode_normal(encoded_normal);
+}
 
-	vec3 P;
+// all vectors in viewspace
+float calculate_pixel_obscurance(vec3 p_pixel_normal, vec3 p_hit_delta, float p_fallof_sq) {
+	float length_sq = dot(p_hit_delta, p_hit_delta);
+	float NdotD = dot(p_pixel_normal, p_hit_delta) / sqrt(length_sq);
 
-	// We need to divide by 2^mipLevel to read the appropriately scaled coordinate from a MIP-map.
-	// Manually clamp to the texture size because texelFetch bypasses the texture unit
-	ivec2 mipP = clamp(ssP >> mipLevel, ivec2(0), (params.screen_size >> mipLevel) - ivec2(1));
+	float falloff_mult = max(0.0, length_sq * p_fallof_sq + 1.0);
 
-#ifdef USE_HALF_SIZE
-	P.z = texelFetch(source_depth_mipmaps, mipP, mipLevel).r;
-	P.z = -P.z;
-#else
-	if (mipLevel < 1) {
-		//read from depth buffer
-		P.z = texelFetch(source_depth, mipP, 0).r;
-		P.z = P.z * 2.0 - 1.0;
-		if (params.orthogonal) {
-			P.z = ((P.z + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / 2.0;
-		} else {
-			P.z = 2.0 * params.z_near * params.z_far / (params.z_far + params.z_near - P.z * (params.z_far - params.z_near));
-		}
-		P.z = -P.z;
+	return max(0, NdotD - params.horizon_angle_threshold) * falloff_mult;
+}
 
-	} else {
-		//read from mipmaps
-		P.z = texelFetch(source_depth_mipmaps, mipP, mipLevel - 1).r;
-		P.z = -P.z;
+void SSAO_tap_inner(const int p_quality_level, inout float r_obscurance_sum, inout float r_weight_sum, const vec2 p_sampling_uv, const float p_mip_level, const vec3 p_pix_center_pos, vec3 p_pixel_normal, const float p_fallof_sq, const float p_weight_mod) {
+	// get depth at sample
+	float viewspace_sample_z = textureLod(source_depth_mipmaps, vec3(p_sampling_uv, params.pass), p_mip_level).x;
+
+	// convert to viewspace
+	vec3 hit_pos = NDC_to_view_space(p_sampling_uv.xy, viewspace_sample_z).xyz;
+	vec3 hit_delta = hit_pos - p_pix_center_pos;
+
+	float obscurance = calculate_pixel_obscurance(p_pixel_normal, hit_delta, p_fallof_sq);
+	float weight = 1.0;
+
+	if (p_quality_level >= SSAO_HALOING_REDUCTION_ENABLE_AT_QUALITY_PRESET) {
+		float reduct = max(0, -hit_delta.z);
+		reduct = clamp(reduct * params.neg_inv_radius + 2.0, 0.0, 1.0);
+		weight = SSAO_HALOING_REDUCTION_AMOUNT * reduct + (1.0 - SSAO_HALOING_REDUCTION_AMOUNT);
 	}
-#endif
+	weight *= p_weight_mod;
+	r_obscurance_sum += obscurance * weight;
+	r_weight_sum += weight;
+}
+
+void SSAOTap(const int p_quality_level, inout float r_obscurance_sum, inout float r_weight_sum, const int p_tap_index, const mat2 p_rot_scale, const vec3 p_pix_center_pos, vec3 p_pixel_normal, const vec2 p_normalized_screen_pos, const float p_mip_offset, const float p_fallof_sq, float p_weight_mod, vec2 p_norm_xy, float p_norm_xy_length) {
+	vec2 sample_offset;
+	float sample_pow_2_len;
+
+	// patterns
+	{
+		vec4 new_sample = sample_pattern[p_tap_index];
+		sample_offset = new_sample.xy * p_rot_scale;
+		sample_pow_2_len = new_sample.w; // precalculated, same as: sample_pow_2_len = log2( length( new_sample.xy ) );
+		p_weight_mod *= new_sample.z;
+	}
+
+	// snap to pixel center (more correct obscurance math, avoids artifacts)
+	sample_offset = round(sample_offset);
+
+	// calculate MIP based on the sample distance from the centre, similar to as described
+	// in http://graphics.cs.williams.edu/papers/SAOHPG12/.
+	float mip_level = (p_quality_level < SSAO_DEPTH_MIPS_ENABLE_AT_QUALITY_PRESET) ? (0) : (sample_pow_2_len + p_mip_offset);
+
+	vec2 sampling_uv = sample_offset * params.half_screen_pixel_size + p_normalized_screen_pos;
+
+	SSAO_tap_inner(p_quality_level, r_obscurance_sum, r_weight_sum, sampling_uv, mip_level, p_pix_center_pos, p_pixel_normal, p_fallof_sq, p_weight_mod);
+
+	// for the second tap, just use the mirrored offset
+	vec2 sample_offset_mirrored_uv = -sample_offset;
 
-	// Offset to pixel center
-	P = reconstructCSPosition(vec2(ssP) + vec2(0.5), P.z);
+	// tilt the second set of samples so that the disk is effectively rotated by the normal
+	// effective at removing one set of artifacts, but too expensive for lower quality settings
+	if (p_quality_level >= SSAO_TILT_SAMPLES_ENABLE_AT_QUALITY_PRESET) {
+		float dot_norm = dot(sample_offset_mirrored_uv, p_norm_xy);
+		sample_offset_mirrored_uv -= dot_norm * p_norm_xy_length * p_norm_xy;
+		sample_offset_mirrored_uv = round(sample_offset_mirrored_uv);
+	}
+
+	// snap to pixel center (more correct obscurance math, avoids artifacts)
+	vec2 sampling_mirrored_uv = sample_offset_mirrored_uv * params.half_screen_pixel_size + p_normalized_screen_pos;
 
-	return P;
+	SSAO_tap_inner(p_quality_level, r_obscurance_sum, r_weight_sum, sampling_mirrored_uv, mip_level, p_pix_center_pos, p_pixel_normal, p_fallof_sq, p_weight_mod);
 }
 
-/** Compute the occlusion due to sample with index \a i about the pixel at \a ssC that corresponds
-	to camera-space point \a C with unit normal \a n_C, using maximum screen-space sampling radius \a ssDiskRadius
+// this function is designed to only work with half/half depth at the moment - there's a couple of hardcoded paths that expect pixel/texel size, so it will not work for full res
+void generate_SSAO_shadows_internal(out float r_shadow_term, out vec4 r_edges, out float r_weight, const vec2 p_pos, int p_quality_level, bool p_adaptive_base) {
+	vec2 pos_rounded = trunc(p_pos);
+	uvec2 upos = uvec2(pos_rounded);
+
+	const int number_of_taps = (p_adaptive_base) ? (SSAO_ADAPTIVE_TAP_BASE_COUNT) : (num_taps[p_quality_level]);
+	float pix_z, pix_left_z, pix_top_z, pix_right_z, pix_bottom_z;
+
+	vec4 valuesUL = textureGather(source_depth_mipmaps, vec3(pos_rounded * params.half_screen_pixel_size, params.pass)); // g_ViewspaceDepthSource.GatherRed(g_PointMirrorSampler, pos_rounded * params.half_screen_pixel_size);
+	vec4 valuesBR = textureGather(source_depth_mipmaps, vec3((pos_rounded + vec2(1.0)) * params.half_screen_pixel_size, params.pass)); // g_ViewspaceDepthSource.GatherRed(g_PointMirrorSampler, pos_rounded * params.half_screen_pixel_size, ivec2(1, 1));
+
+	// get this pixel's viewspace depth
+	pix_z = valuesUL.y;
+
+	// get left right top bottom neighbouring pixels for edge detection (gets compiled out on quality_level == 0)
+	pix_left_z = valuesUL.x;
+	pix_top_z = valuesUL.z;
+	pix_right_z = valuesBR.z;
+	pix_bottom_z = valuesBR.x;
 
-	Note that units of H() in the HPG12 paper are meters, not
-	unitless.  The whole falloff/sampling function is therefore
-	unitless.  In this implementation, we factor out (9 / radius).
+	vec2 normalized_screen_pos = pos_rounded * params.half_screen_pixel_size + params.half_screen_pixel_size_x025;
+	vec3 pix_center_pos = NDC_to_view_space(normalized_screen_pos, pix_z);
 
-	Four versions of the falloff function are implemented below
-*/
-float sampleAO(in ivec2 ssC, in vec3 C, in vec3 n_C, in float ssDiskRadius, in float p_radius, in int tapIndex, in float randomPatternRotationAngle) {
-	// Offset on the unit disk, spun for this pixel
-	float ssR;
-	vec2 unitOffset = tapLocation(tapIndex, randomPatternRotationAngle, ssR);
-	ssR *= ssDiskRadius;
+	// Load this pixel's viewspace normal
+	uvec2 full_res_coord = upos * 2 * params.size_multiplier + params.pass_coord_offset.xy;
+	vec3 pixel_normal = load_normal(ivec2(full_res_coord));
 
-	ivec2 ssP = ivec2(ssR * unitOffset) + ssC;
+	//const vec2 pixel_size_at_center = pix_center_pos.z * params.NDC_to_view_mul * params.half_screen_pixel_size; // optimized approximation of:
+	vec2 pixel_size_at_center = NDC_to_view_space(normalized_screen_pos.xy + params.half_screen_pixel_size * 0.5, pix_center_pos.z).xy - pix_center_pos.xy;
 
-	if (any(lessThan(ssP, ivec2(0))) || any(greaterThanEqual(ssP, params.screen_size))) {
-		return 0.0;
+	float pixel_lookup_radius;
+	float fallof_sq;
+
+	// calculate effect radius and fit our screen sampling pattern inside it
+	float viewspace_radius;
+	calculate_radius_parameters(length(pix_center_pos), pixel_size_at_center, pixel_lookup_radius, viewspace_radius, fallof_sq);
+
+	// calculate samples rotation/scaling
+	mat2 rot_scale_matrix;
+	uint pseudo_random_index;
+
+	{
+		vec4 rotation_scale;
+		// reduce effect radius near the screen edges slightly; ideally, one would render a larger depth buffer (5% on each side) instead
+		if (!p_adaptive_base && (p_quality_level >= SSAO_REDUCE_RADIUS_NEAR_SCREEN_BORDER_ENABLE_AT_QUALITY_PRESET)) {
+			float near_screen_border = min(min(normalized_screen_pos.x, 1.0 - normalized_screen_pos.x), min(normalized_screen_pos.y, 1.0 - normalized_screen_pos.y));
+			near_screen_border = clamp(10.0 * near_screen_border + 0.6, 0.0, 1.0);
+			pixel_lookup_radius *= near_screen_border;
+		}
+
+		// load & update pseudo-random rotation matrix
+		pseudo_random_index = uint(pos_rounded.y * 2 + pos_rounded.x) % 5;
+		rotation_scale = constants.rotation_matrices[params.pass * 5 + pseudo_random_index];
+		rot_scale_matrix = mat2(rotation_scale.x * pixel_lookup_radius, rotation_scale.y * pixel_lookup_radius, rotation_scale.z * pixel_lookup_radius, rotation_scale.w * pixel_lookup_radius);
 	}
 
-	// The occluding point in camera space
-	vec3 Q = getOffsetPosition(ssP, ssR);
+	// the main obscurance & sample weight storage
+	float obscurance_sum = 0.0;
+	float weight_sum = 0.0;
 
-	vec3 v = Q - C;
+	// edge mask for between this and left/right/top/bottom neighbour pixels - not used in quality level 0 so initialize to "no edge" (1 is no edge, 0 is edge)
+	vec4 edgesLRTB = vec4(1.0, 1.0, 1.0, 1.0);
 
-	float vv = dot(v, v);
-	float vn = dot(v, n_C);
+	// Move center pixel slightly towards camera to avoid imprecision artifacts due to using of 16bit depth buffer; a lot smaller offsets needed when using 32bit floats
+	pix_center_pos *= 0.9992;
 
-	const float epsilon = 0.01;
-	float radius2 = p_radius * p_radius;
+	if (!p_adaptive_base && (p_quality_level >= SSAO_DEPTH_BASED_EDGES_ENABLE_AT_QUALITY_PRESET)) {
+		edgesLRTB = calculate_edges(pix_z, pix_left_z, pix_right_z, pix_top_z, pix_bottom_z);
+	}
 
-	// A: From the HPG12 paper
-	// Note large epsilon to avoid overdarkening within cracks
-	//return float(vv < radius2) * max((vn - bias) / (epsilon + vv), 0.0) * radius2 * 0.6;
+	// adds a more high definition sharp effect, which gets blurred out (reuses left/right/top/bottom samples that we used for edge detection)
+	if (!p_adaptive_base && (p_quality_level >= SSAO_DETAIL_AO_ENABLE_AT_QUALITY_PRESET)) {
+		// disable in case of quality level 4 (reference)
+		if (p_quality_level != 4) {
+			//approximate neighbouring pixels positions (actually just deltas or "positions - pix_center_pos" )
+			vec3 normalized_viewspace_dir = vec3(pix_center_pos.xy / pix_center_pos.zz, 1.0);
+			vec3 pixel_left_delta = vec3(-pixel_size_at_center.x, 0.0, 0.0) + normalized_viewspace_dir * (pix_left_z - pix_center_pos.z);
+			vec3 pixel_right_delta = vec3(+pixel_size_at_center.x, 0.0, 0.0) + normalized_viewspace_dir * (pix_right_z - pix_center_pos.z);
+			vec3 pixel_top_delta = vec3(0.0, -pixel_size_at_center.y, 0.0) + normalized_viewspace_dir * (pix_top_z - pix_center_pos.z);
+			vec3 pixel_bottom_delta = vec3(0.0, +pixel_size_at_center.y, 0.0) + normalized_viewspace_dir * (pix_bottom_z - pix_center_pos.z);
+
+			const float range_reduction = 4.0f; // this is to avoid various artifacts
+			const float modified_fallof_sq = range_reduction * fallof_sq;
+
+			vec4 additional_obscurance;
+			additional_obscurance.x = calculate_pixel_obscurance(pixel_normal, pixel_left_delta, modified_fallof_sq);
+			additional_obscurance.y = calculate_pixel_obscurance(pixel_normal, pixel_right_delta, modified_fallof_sq);
+			additional_obscurance.z = calculate_pixel_obscurance(pixel_normal, pixel_top_delta, modified_fallof_sq);
+			additional_obscurance.w = calculate_pixel_obscurance(pixel_normal, pixel_bottom_delta, modified_fallof_sq);
+
+			obscurance_sum += params.detail_intensity * dot(additional_obscurance, edgesLRTB);
+		}
+	}
 
-	// B: Smoother transition to zero (lowers contrast, smoothing out corners). [Recommended]
-	float f = max(radius2 - vv, 0.0);
-	return f * f * f * max((vn - params.bias) / (epsilon + vv), 0.0);
+	// Sharp normals also create edges - but this adds to the cost as well
+	if (!p_adaptive_base && (p_quality_level >= SSAO_NORMAL_BASED_EDGES_ENABLE_AT_QUALITY_PRESET)) {
+		vec3 neighbour_normal_left = load_normal(ivec2(full_res_coord), ivec2(-2, 0));
+		vec3 neighbour_normal_right = load_normal(ivec2(full_res_coord), ivec2(2, 0));
+		vec3 neighbour_normal_top = load_normal(ivec2(full_res_coord), ivec2(0, -2));
+		vec3 neighbour_normal_bottom = load_normal(ivec2(full_res_coord), ivec2(0, 2));
 
-	// C: Medium contrast (which looks better at high radii), no division.  Note that the
-	// contribution still falls off with radius^2, but we've adjusted the rate in a way that is
-	// more computationally efficient and happens to be aesthetically pleasing.
-	// return 4.0 * max(1.0 - vv * invRadius2, 0.0) * max(vn - bias, 0.0);
+		const float dot_threshold = SSAO_NORMAL_BASED_EDGES_DOT_THRESHOLD;
 
-	// D: Low contrast, no division operation
-	// return 2.0 * float(vv < radius * radius) * max(vn - bias, 0.0);
-}
+		vec4 normal_edgesLRTB;
+		normal_edgesLRTB.x = clamp((dot(pixel_normal, neighbour_normal_left) + dot_threshold), 0.0, 1.0);
+		normal_edgesLRTB.y = clamp((dot(pixel_normal, neighbour_normal_right) + dot_threshold), 0.0, 1.0);
+		normal_edgesLRTB.z = clamp((dot(pixel_normal, neighbour_normal_top) + dot_threshold), 0.0, 1.0);
+		normal_edgesLRTB.w = clamp((dot(pixel_normal, neighbour_normal_bottom) + dot_threshold), 0.0, 1.0);
 
-void main() {
-	// Pixel being shaded
-	ivec2 ssC = ivec2(gl_GlobalInvocationID.xy);
-	if (any(greaterThanEqual(ssC, params.screen_size))) { //too large, do nothing
-		return;
+		edgesLRTB *= normal_edgesLRTB;
 	}
 
-	// World space point being shaded
-	vec3 C = getPosition(ssC);
+	const float global_mip_offset = SSAO_DEPTH_MIPS_GLOBAL_OFFSET;
+	float mip_offset = (p_quality_level < SSAO_DEPTH_MIPS_ENABLE_AT_QUALITY_PRESET) ? (0) : (log2(pixel_lookup_radius) + global_mip_offset);
 
-#ifdef USE_HALF_SIZE
-	vec3 n_C = texelFetch(source_normal, ssC << 1, 0).xyz * 2.0 - 1.0;
-#else
-	vec3 n_C = texelFetch(source_normal, ssC, 0).xyz * 2.0 - 1.0;
+	// Used to tilt the second set of samples so that the disk is effectively rotated by the normal
+	// effective at removing one set of artifacts, but too expensive for lower quality settings
+	vec2 norm_xy = vec2(pixel_normal.x, pixel_normal.y);
+	float norm_xy_length = length(norm_xy);
+	norm_xy /= vec2(norm_xy_length, -norm_xy_length);
+	norm_xy_length *= SSAO_TILT_SAMPLES_AMOUNT;
+
+	// standard, non-adaptive approach
+	if ((p_quality_level != 3) || p_adaptive_base) {
+		for (int i = 0; i < number_of_taps; i++) {
+			SSAOTap(p_quality_level, obscurance_sum, weight_sum, i, rot_scale_matrix, pix_center_pos, pixel_normal, normalized_screen_pos, mip_offset, fallof_sq, 1.0, norm_xy, norm_xy_length);
+		}
+	}
+#ifdef ADAPTIVE
+	else {
+		// add new ones if needed
+		vec2 full_res_uv = normalized_screen_pos + params.pass_uv_offset.xy;
+		float importance = textureLod(source_importance, full_res_uv, 0.0).x;
+
+		// this is to normalize SSAO_DETAIL_AO_AMOUNT across all pixel regardless of importance
+		obscurance_sum *= (SSAO_ADAPTIVE_TAP_BASE_COUNT / float(SSAO_MAX_TAPS)) + (importance * SSAO_ADAPTIVE_TAP_FLEXIBLE_COUNT / float(SSAO_MAX_TAPS));
+
+		// load existing base values
+		vec2 base_values = imageLoad(source_ssao, ivec3(upos, params.pass)).xy;
+		weight_sum += base_values.y * float(SSAO_ADAPTIVE_TAP_BASE_COUNT * 4.0);
+		obscurance_sum += (base_values.x) * weight_sum;
+
+		// increase importance around edges
+		float edge_count = dot(1.0 - edgesLRTB, vec4(1.0, 1.0, 1.0, 1.0));
+
+		float avg_total_importance = float(counter.sum) * params.load_counter_avg_div;
+
+		float importance_limiter = clamp(params.adaptive_sample_limit / avg_total_importance, 0.0, 1.0);
+		importance *= importance_limiter;
+
+		float additional_sample_count = SSAO_ADAPTIVE_TAP_FLEXIBLE_COUNT * importance;
+
+		const float blend_range = 3.0;
+		const float blend_range_inv = 1.0 / blend_range;
+
+		additional_sample_count += 0.5;
+		uint additional_samples = uint(additional_sample_count);
+		uint additional_samples_to = min(SSAO_MAX_TAPS, additional_samples + SSAO_ADAPTIVE_TAP_BASE_COUNT);
+
+		for (uint i = SSAO_ADAPTIVE_TAP_BASE_COUNT; i < additional_samples_to; i++) {
+			additional_sample_count -= 1.0f;
+			float weight_mod = clamp(additional_sample_count * blend_range_inv, 0.0, 1.0);
+			SSAOTap(p_quality_level, obscurance_sum, weight_sum, int(i), rot_scale_matrix, pix_center_pos, pixel_normal, normalized_screen_pos, mip_offset, fallof_sq, weight_mod, norm_xy, norm_xy_length);
+		}
+	}
 #endif
-	n_C = normalize(n_C);
-	n_C.y = -n_C.y; //because this code reads flipped
 
-	// Hash function used in the HPG12 AlchemyAO paper
-	float randomPatternRotationAngle = mod(float((3 * ssC.x ^ ssC.y + ssC.x * ssC.y) * 10), TWO_PI);
+	// early out for adaptive base - just output weight (used for the next pass)
+	if (p_adaptive_base) {
+		float obscurance = obscurance_sum / weight_sum;
+
+		r_shadow_term = obscurance;
+		r_edges = vec4(0.0);
+		r_weight = weight_sum;
+		return;
+	}
+
+	// calculate weighted average
+	float obscurance = obscurance_sum / weight_sum;
 
-	// Reconstruct normals from positions. These will lead to 1-pixel black lines
-	// at depth discontinuities, however the blur will wipe those out so they are not visible
-	// in the final image.
+	// calculate fadeout (1 close, gradient, 0 far)
+	float fade_out = clamp(pix_center_pos.z * params.fade_out_mul + params.fade_out_add, 0.0, 1.0);
 
-	// Choose the screen-space sample radius
-	// proportional to the projected area of the sphere
+	// Reduce the SSAO shadowing if we're on the edge to remove artifacts on edges (we don't care for the lower quality one)
+	if (!p_adaptive_base && (p_quality_level >= SSAO_DEPTH_BASED_EDGES_ENABLE_AT_QUALITY_PRESET)) {
+		// when there's more than 2 opposite edges, start fading out the occlusion to reduce aliasing artifacts
+		float edge_fadeout_factor = clamp((1.0 - edgesLRTB.x - edgesLRTB.y) * 0.35, 0.0, 1.0) + clamp((1.0 - edgesLRTB.z - edgesLRTB.w) * 0.35, 0.0, 1.0);
 
-	float ssDiskRadius = -params.proj_scale * params.radius;
-	if (!params.orthogonal) {
-		ssDiskRadius = -params.proj_scale * params.radius / C.z;
+		fade_out *= clamp(1.0 - edge_fadeout_factor, 0.0, 1.0);
 	}
-	float sum = 0.0;
-	for (int i = 0; i < NUM_SAMPLES; ++i) {
-		sum += sampleAO(ssC, C, n_C, ssDiskRadius, params.radius, i, randomPatternRotationAngle);
+
+	// same as a bove, but a lot more conservative version
+	// fade_out *= clamp( dot( edgesLRTB, vec4( 0.9, 0.9, 0.9, 0.9 ) ) - 2.6 , 0.0, 1.0);
+
+	// strength
+	obscurance = params.intensity * obscurance;
+
+	// clamp
+	obscurance = min(obscurance, params.shadow_clamp);
+
+	// fadeout
+	obscurance *= fade_out;
+
+	// conceptually switch to occlusion with the meaning being visibility (grows with visibility, occlusion == 1 implies full visibility),
+	// to be in line with what is more commonly used.
+	float occlusion = 1.0 - obscurance;
+
+	// modify the gradient
+	// note: this cannot be moved to a later pass because of loss of precision after storing in the render target
+	occlusion = pow(clamp(occlusion, 0.0, 1.0), params.shadow_power);
+
+	// outputs!
+	r_shadow_term = occlusion; // Our final 'occlusion' term (0 means fully occluded, 1 means fully lit)
+	r_edges = edgesLRTB; // These are used to prevent blurring across edges, 1 means no edge, 0 means edge, 0.5 means half way there, etc.
+	r_weight = weight_sum;
+}
+
+void main() {
+	float out_shadow_term;
+	float out_weight;
+	vec4 out_edges;
+	ivec2 ssC = ivec2(gl_GlobalInvocationID.xy);
+	if (any(greaterThanEqual(ssC, params.screen_size))) { //too large, do nothing
+		return;
 	}
 
-	float A = max(0.0, 1.0 - sum * params.intensity_div_r6 * (5.0 / float(NUM_SAMPLES)));
+	vec2 uv = vec2(gl_GlobalInvocationID) + vec2(0.5);
+#ifdef SSAO_BASE
+	generate_SSAO_shadows_internal(out_shadow_term, out_edges, out_weight, uv, params.quality, true);
+
+	imageStore(dest_image, ivec2(gl_GlobalInvocationID.xy), vec4(out_shadow_term, out_weight / (float(SSAO_ADAPTIVE_TAP_BASE_COUNT) * 4.0), 0.0, 0.0));
+#else
+	generate_SSAO_shadows_internal(out_shadow_term, out_edges, out_weight, uv, params.quality, false); // pass in quality levels
+	if (params.quality == 0) {
+		out_edges = vec4(1.0);
+	}
 
-	imageStore(dest_image, ssC, vec4(A));
+	imageStore(dest_image, ivec2(gl_GlobalInvocationID.xy), vec4(out_shadow_term, pack_edges(out_edges), 0.0, 0.0));
+#endif
 }
diff --git a/servers/rendering/renderer_rd/shaders/ssao_blur.glsl b/servers/rendering/renderer_rd/shaders/ssao_blur.glsl
index 3e63e3cb59..510a777048 100644
--- a/servers/rendering/renderer_rd/shaders/ssao_blur.glsl
+++ b/servers/rendering/renderer_rd/shaders/ssao_blur.glsl
@@ -1,3 +1,22 @@
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// Copyright (c) 2016, Intel Corporation
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated
+// documentation files (the "Software"), to deal in the Software without restriction, including without limitation
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to
+// permit persons to whom the Software is furnished to do so, subject to the following conditions:
+// The above copyright notice and this permission notice shall be included in all copies or substantial portions of
+// the Software.
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+// TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// File changes (yyyy-mm-dd)
+// 2016-09-07: filip.strugar@intel.com: first commit
+// 2020-12-05: clayjohn: convert to Vulkan and Godot
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+
 #[compute]
 
 #version 450
@@ -7,147 +26,129 @@ VERSION_DEFINES
 layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
 
 layout(set = 0, binding = 0) uniform sampler2D source_ssao;
-layout(set = 1, binding = 0) uniform sampler2D source_depth;
-#ifdef MODE_UPSCALE
-layout(set = 2, binding = 0) uniform sampler2D source_depth_mipmaps;
-#endif
 
-layout(r8, set = 3, binding = 0) uniform restrict writeonly image2D dest_image;
-
-//////////////////////////////////////////////////////////////////////////////////////////////
-// Tunable Parameters:
+layout(rg8, set = 1, binding = 0) uniform restrict writeonly image2D dest_image;
 
 layout(push_constant, binding = 1, std430) uniform Params {
-	float edge_sharpness; /** Increase to make depth edges crisper. Decrease to reduce flicker. */
-	int filter_scale;
-	float z_far;
-	float z_near;
-	bool orthogonal;
-	uint pad0;
-	uint pad1;
-	uint pad2;
-	ivec2 axis; /** (1, 0) or (0, 1) */
-	ivec2 screen_size;
+	float edge_sharpness;
+	float pad;
+	vec2 half_screen_pixel_size;
 }
 params;
 
-/** Filter radius in pixels. This will be multiplied by SCALE. */
-#define R (4)
+vec4 unpack_edges(float p_packed_val) {
+	uint packed_val = uint(p_packed_val * 255.5);
+	vec4 edgesLRTB;
+	edgesLRTB.x = float((packed_val >> 6) & 0x03) / 3.0;
+	edgesLRTB.y = float((packed_val >> 4) & 0x03) / 3.0;
+	edgesLRTB.z = float((packed_val >> 2) & 0x03) / 3.0;
+	edgesLRTB.w = float((packed_val >> 0) & 0x03) / 3.0;
+
+	return clamp(edgesLRTB + params.edge_sharpness, 0.0, 1.0);
+}
+
+void add_sample(float p_ssao_value, float p_edge_value, inout float r_sum, inout float r_sum_weight) {
+	float weight = p_edge_value;
+
+	r_sum += (weight * p_ssao_value);
+	r_sum_weight += weight;
+}
+
+#ifdef MODE_WIDE
+vec2 sample_blurred_wide(vec2 p_coord) {
+	vec2 vC = textureLodOffset(source_ssao, vec2(p_coord), 0.0, ivec2(0, 0)).xy;
+	vec2 vL = textureLodOffset(source_ssao, vec2(p_coord), 0.0, ivec2(-2, 0)).xy;
+	vec2 vT = textureLodOffset(source_ssao, vec2(p_coord), 0.0, ivec2(0, -2)).xy;
+	vec2 vR = textureLodOffset(source_ssao, vec2(p_coord), 0.0, ivec2(2, 0)).xy;
+	vec2 vB = textureLodOffset(source_ssao, vec2(p_coord), 0.0, ivec2(0, 2)).xy;
+
+	float packed_edges = vC.y;
+	vec4 edgesLRTB = unpack_edges(packed_edges);
+	edgesLRTB.x *= unpack_edges(vL.y).y;
+	edgesLRTB.z *= unpack_edges(vT.y).w;
+	edgesLRTB.y *= unpack_edges(vR.y).x;
+	edgesLRTB.w *= unpack_edges(vB.y).z;
+
+	float ssao_value = vC.x;
+	float ssao_valueL = vL.x;
+	float ssao_valueT = vT.x;
+	float ssao_valueR = vR.x;
+	float ssao_valueB = vB.x;
+
+	float sum_weight = 0.8f;
+	float sum = ssao_value * sum_weight;
+
+	add_sample(ssao_valueL, edgesLRTB.x, sum, sum_weight);
+	add_sample(ssao_valueR, edgesLRTB.y, sum, sum_weight);
+	add_sample(ssao_valueT, edgesLRTB.z, sum, sum_weight);
+	add_sample(ssao_valueB, edgesLRTB.w, sum, sum_weight);
+
+	float ssao_avg = sum / sum_weight;
+
+	ssao_value = ssao_avg;
+
+	return vec2(ssao_value, packed_edges);
+}
+#endif
+
+#ifdef MODE_SMART
+vec2 sample_blurred(vec3 p_pos, vec2 p_coord) {
+	float packed_edges = texelFetch(source_ssao, ivec2(p_pos.xy), 0).y;
+	vec4 edgesLRTB = unpack_edges(packed_edges);
+
+	vec4 valuesUL = textureGather(source_ssao, vec2(p_coord - params.half_screen_pixel_size * 0.5));
+	vec4 valuesBR = textureGather(source_ssao, vec2(p_coord + params.half_screen_pixel_size * 0.5));
+
+	float ssao_value = valuesUL.y;
+	float ssao_valueL = valuesUL.x;
+	float ssao_valueT = valuesUL.z;
+	float ssao_valueR = valuesBR.z;
+	float ssao_valueB = valuesBR.x;
+
+	float sum_weight = 0.5;
+	float sum = ssao_value * sum_weight;
+
+	add_sample(ssao_valueL, edgesLRTB.x, sum, sum_weight);
+	add_sample(ssao_valueR, edgesLRTB.y, sum, sum_weight);
+
+	add_sample(ssao_valueT, edgesLRTB.z, sum, sum_weight);
+	add_sample(ssao_valueB, edgesLRTB.w, sum, sum_weight);
 
-//////////////////////////////////////////////////////////////////////////////////////////////
+	float ssao_avg = sum / sum_weight;
 
-// Gaussian coefficients
-const float gaussian[R + 1] =
-		//float[](0.356642, 0.239400, 0.072410, 0.009869);
-		//float[](0.398943, 0.241971, 0.053991, 0.004432, 0.000134);  // stddev = 1.0
-		float[](0.153170, 0.144893, 0.122649, 0.092902, 0.062970); // stddev = 2.0
-//float[](0.111220, 0.107798, 0.098151, 0.083953, 0.067458, 0.050920, 0.036108); // stddev = 3.0
+	ssao_value = ssao_avg;
+
+	return vec2(ssao_value, packed_edges);
+}
+#endif
 
 void main() {
 	// Pixel being shaded
 	ivec2 ssC = ivec2(gl_GlobalInvocationID.xy);
-	if (any(greaterThanEqual(ssC, params.screen_size))) { //too large, do nothing
-		return;
-	}
-
-#ifdef MODE_UPSCALE
-
-	//closest one should be the same pixel, but check nearby just in case
-	float depth = texelFetch(source_depth, ssC, 0).r;
-
-	depth = depth * 2.0 - 1.0;
-	if (params.orthogonal) {
-		depth = ((depth + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / 2.0;
-	} else {
-		depth = 2.0 * params.z_near * params.z_far / (params.z_far + params.z_near - depth * (params.z_far - params.z_near));
-	}
-
-	vec2 pixel_size = 1.0 / vec2(params.screen_size);
-	vec2 closest_uv = vec2(ssC) * pixel_size + pixel_size * 0.5;
-	vec2 from_uv = closest_uv;
-	vec2 ps2 = pixel_size; // * 2.0;
-
-	float closest_depth = abs(textureLod(source_depth_mipmaps, closest_uv, 0.0).r - depth);
-
-	vec2 offsets[4] = vec2[](vec2(ps2.x, 0), vec2(-ps2.x, 0), vec2(0, ps2.y), vec2(0, -ps2.y));
-	for (int i = 0; i < 4; i++) {
-		vec2 neighbour = from_uv + offsets[i];
-		float neighbour_depth = abs(textureLod(source_depth_mipmaps, neighbour, 0.0).r - depth);
-		if (neighbour_depth < closest_depth) {
-			closest_uv = neighbour;
-			closest_depth = neighbour_depth;
-		}
-	}
-
-	float visibility = textureLod(source_ssao, closest_uv, 0.0).r;
-	imageStore(dest_image, ssC, vec4(visibility));
-#else
 
-	float depth = texelFetch(source_depth, ssC, 0).r;
+#ifdef MODE_NON_SMART
 
-#ifdef MODE_FULL_SIZE
-	depth = depth * 2.0 - 1.0;
+	vec2 halfPixel = params.half_screen_pixel_size * 0.5f;
 
-	if (params.orthogonal) {
-		depth = ((depth + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / 2.0;
-	} else {
-		depth = 2.0 * params.z_near * params.z_far / (params.z_far + params.z_near - depth * (params.z_far - params.z_near));
-	}
+	vec2 uv = (vec2(gl_GlobalInvocationID.xy) + vec2(0.5, 0.5)) * params.half_screen_pixel_size;
 
-#endif
-	float depth_divide = 1.0 / params.z_far;
-
-	//depth *= depth_divide;
-
-	/*
-	if (depth > params.z_far * 0.999) {
-		discard; //skybox
-	}
-	*/
-
-	float sum = texelFetch(source_ssao, ssC, 0).r;
-
-	// Base weight for depth falloff.  Increase this for more blurriness,
-	// decrease it for better edge discrimination
-	float BASE = gaussian[0];
-	float totalWeight = BASE;
-	sum *= totalWeight;
-
-	ivec2 clamp_limit = params.screen_size - ivec2(1);
-
-	for (int r = -R; r <= R; ++r) {
-		// We already handled the zero case above.  This loop should be unrolled and the static branch optimized out,
-		// so the IF statement has no runtime cost
-		if (r != 0) {
-			ivec2 ppos = ssC + params.axis * (r * params.filter_scale);
-			float value = texelFetch(source_ssao, clamp(ppos, ivec2(0), clamp_limit), 0).r;
-			ivec2 rpos = clamp(ppos, ivec2(0), clamp_limit);
-
-			float temp_depth = texelFetch(source_depth, rpos, 0).r;
-#ifdef MODE_FULL_SIZE
-			temp_depth = temp_depth * 2.0 - 1.0;
-			if (params.orthogonal) {
-				temp_depth = ((temp_depth + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / 2.0;
-			} else {
-				temp_depth = 2.0 * params.z_near * params.z_far / (params.z_far + params.z_near - temp_depth * (params.z_far - params.z_near));
-			}
-			//temp_depth *= depth_divide;
-#endif
-			// spatial domain: offset gaussian tap
-			float weight = 0.3 + gaussian[abs(r)];
-			//weight *= max(0.0, dot(temp_normal, normal));
+	vec2 centre = textureLod(source_ssao, vec2(uv), 0.0).xy;
 
-			// range domain (the "bilateral" weight). As depth difference increases, decrease weight.
-			weight *= max(0.0, 1.0 - params.edge_sharpness * abs(temp_depth - depth));
+	vec4 vals;
+	vals.x = textureLod(source_ssao, vec2(uv + vec2(-halfPixel.x * 3, -halfPixel.y)), 0.0).x;
+	vals.y = textureLod(source_ssao, vec2(uv + vec2(+halfPixel.x, -halfPixel.y * 3)), 0.0).x;
+	vals.z = textureLod(source_ssao, vec2(uv + vec2(-halfPixel.x, +halfPixel.y * 3)), 0.0).x;
+	vals.w = textureLod(source_ssao, vec2(uv + vec2(+halfPixel.x * 3, +halfPixel.y)), 0.0).x;
 
-			sum += value * weight;
-			totalWeight += weight;
-		}
-	}
+	vec2 sampled = vec2(dot(vals, vec4(0.2)) + centre.x * 0.2, centre.y);
 
-	const float epsilon = 0.0001;
-	float visibility = sum / (totalWeight + epsilon);
+#else
+#ifdef MODE_SMART
+	vec2 sampled = sample_blurred(vec3(gl_GlobalInvocationID), (vec2(gl_GlobalInvocationID.xy) + vec2(0.5, 0.5)) * params.half_screen_pixel_size);
+#else // MODE_WIDE
+	vec2 sampled = sample_blurred_wide((vec2(gl_GlobalInvocationID.xy) + vec2(0.5, 0.5)) * params.half_screen_pixel_size);
+#endif
 
-	imageStore(dest_image, ssC, vec4(visibility));
 #endif
+	imageStore(dest_image, ivec2(ssC), vec4(sampled, 0.0, 0.0));
 }
diff --git a/servers/rendering/renderer_rd/shaders/ssao_downsample.glsl b/servers/rendering/renderer_rd/shaders/ssao_downsample.glsl
new file mode 100644
index 0000000000..cb2d31f70d
--- /dev/null
+++ b/servers/rendering/renderer_rd/shaders/ssao_downsample.glsl
@@ -0,0 +1,206 @@
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// Copyright (c) 2016, Intel Corporation
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated
+// documentation files (the "Software"), to deal in the Software without restriction, including without limitation
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to
+// permit persons to whom the Software is furnished to do so, subject to the following conditions:
+// The above copyright notice and this permission notice shall be included in all copies or substantial portions of
+// the Software.
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+// TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// File changes (yyyy-mm-dd)
+// 2016-09-07: filip.strugar@intel.com: first commit
+// 2020-12-05: clayjohn: convert to Vulkan and Godot
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+
+#[compute]
+
+#version 450
+
+VERSION_DEFINES
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+layout(push_constant, binding = 1, std430) uniform Params {
+	vec2 pixel_size;
+	float z_far;
+	float z_near;
+	bool orthogonal;
+	float radius_sq;
+	uvec2 pad;
+}
+params;
+
+layout(set = 0, binding = 0) uniform sampler2D source_depth;
+
+layout(r16f, set = 1, binding = 0) uniform restrict writeonly image2DArray dest_image0; //rename
+#ifdef GENERATE_MIPS
+layout(r16f, set = 2, binding = 0) uniform restrict writeonly image2DArray dest_image1;
+layout(r16f, set = 2, binding = 1) uniform restrict writeonly image2DArray dest_image2;
+layout(r16f, set = 2, binding = 2) uniform restrict writeonly image2DArray dest_image3;
+#endif
+
+vec4 screen_space_to_view_space_depth(vec4 p_depth) {
+	if (params.orthogonal) {
+		vec4 depth = p_depth * 2.0 - 1.0;
+		return ((depth + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / 2.0;
+	}
+
+	float depth_linearize_mul = params.z_near;
+	float depth_linearize_add = params.z_far;
+
+	// Optimised version of "-cameraClipNear / (cameraClipFar - projDepth * (cameraClipFar - cameraClipNear)) * cameraClipFar"
+
+	// Set your depth_linearize_mul and depth_linearize_add to:
+	// depth_linearize_mul = ( cameraClipFar * cameraClipNear) / ( cameraClipFar - cameraClipNear );
+	// depth_linearize_add = cameraClipFar / ( cameraClipFar - cameraClipNear );
+
+	return depth_linearize_mul / (depth_linearize_add - p_depth);
+}
+
+float screen_space_to_view_space_depth(float p_depth) {
+	if (params.orthogonal) {
+		float depth = p_depth * 2.0 - 1.0;
+		return ((depth + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / (2.0 * params.z_far);
+	}
+
+	float depth_linearize_mul = params.z_near;
+	float depth_linearize_add = params.z_far;
+
+	return depth_linearize_mul / (depth_linearize_add - p_depth);
+}
+
+#ifdef GENERATE_MIPS
+
+shared float depth_buffer[4][8][8];
+
+float mip_smart_average(vec4 p_depths) {
+	float closest = min(min(p_depths.x, p_depths.y), min(p_depths.z, p_depths.w));
+	float fallof_sq = -1.0f / params.radius_sq;
+	vec4 dists = p_depths - closest.xxxx;
+	vec4 weights = clamp(dists * dists * fallof_sq + 1.0, 0.0, 1.0);
+	return dot(weights, p_depths) / dot(weights, vec4(1.0, 1.0, 1.0, 1.0));
+}
+
+void prepare_depths_and_mips(vec4 p_samples, uvec2 p_output_coord, uvec2 p_gtid) {
+	p_samples = screen_space_to_view_space_depth(p_samples);
+
+	depth_buffer[0][p_gtid.x][p_gtid.y] = p_samples.w;
+	depth_buffer[1][p_gtid.x][p_gtid.y] = p_samples.z;
+	depth_buffer[2][p_gtid.x][p_gtid.y] = p_samples.x;
+	depth_buffer[3][p_gtid.x][p_gtid.y] = p_samples.y;
+
+	imageStore(dest_image0, ivec3(p_output_coord.x, p_output_coord.y, 0), vec4(p_samples.w));
+	imageStore(dest_image0, ivec3(p_output_coord.x, p_output_coord.y, 1), vec4(p_samples.z));
+	imageStore(dest_image0, ivec3(p_output_coord.x, p_output_coord.y, 2), vec4(p_samples.x));
+	imageStore(dest_image0, ivec3(p_output_coord.x, p_output_coord.y, 3), vec4(p_samples.y));
+
+	uint depth_array_index = 2 * (p_gtid.y % 2) + (p_gtid.x % 2);
+	uvec2 depth_array_offset = ivec2(p_gtid.x % 2, p_gtid.y % 2);
+	ivec2 buffer_coord = ivec2(p_gtid) - ivec2(depth_array_offset);
+
+	p_output_coord /= 2;
+	groupMemoryBarrier();
+	barrier();
+
+	// if (still_alive) <-- all threads alive here
+	{
+		float sample_00 = depth_buffer[depth_array_index][buffer_coord.x + 0][buffer_coord.y + 0];
+		float sample_01 = depth_buffer[depth_array_index][buffer_coord.x + 0][buffer_coord.y + 1];
+		float sample_10 = depth_buffer[depth_array_index][buffer_coord.x + 1][buffer_coord.y + 0];
+		float sample_11 = depth_buffer[depth_array_index][buffer_coord.x + 1][buffer_coord.y + 1];
+
+		float avg = mip_smart_average(vec4(sample_00, sample_01, sample_10, sample_11));
+		imageStore(dest_image1, ivec3(p_output_coord.x, p_output_coord.y, depth_array_index), vec4(avg));
+		depth_buffer[depth_array_index][buffer_coord.x][buffer_coord.y] = avg;
+	}
+
+	bool still_alive = p_gtid.x % 4 == depth_array_offset.x && p_gtid.y % 4 == depth_array_offset.y;
+
+	p_output_coord /= 2;
+	groupMemoryBarrier();
+	barrier();
+
+	if (still_alive) {
+		float sample_00 = depth_buffer[depth_array_index][buffer_coord.x + 0][buffer_coord.y + 0];
+		float sample_01 = depth_buffer[depth_array_index][buffer_coord.x + 0][buffer_coord.y + 2];
+		float sample_10 = depth_buffer[depth_array_index][buffer_coord.x + 2][buffer_coord.y + 0];
+		float sample_11 = depth_buffer[depth_array_index][buffer_coord.x + 2][buffer_coord.y + 2];
+
+		float avg = mip_smart_average(vec4(sample_00, sample_01, sample_10, sample_11));
+		imageStore(dest_image2, ivec3(p_output_coord.x, p_output_coord.y, depth_array_index), vec4(avg));
+		depth_buffer[depth_array_index][buffer_coord.x][buffer_coord.y] = avg;
+	}
+
+	still_alive = p_gtid.x % 8 == depth_array_offset.x && depth_array_offset.y % 8 == depth_array_offset.y;
+
+	p_output_coord /= 2;
+	groupMemoryBarrier();
+	barrier();
+
+	if (still_alive) {
+		float sample_00 = depth_buffer[depth_array_index][buffer_coord.x + 0][buffer_coord.y + 0];
+		float sample_01 = depth_buffer[depth_array_index][buffer_coord.x + 0][buffer_coord.y + 4];
+		float sample_10 = depth_buffer[depth_array_index][buffer_coord.x + 4][buffer_coord.y + 0];
+		float sample_11 = depth_buffer[depth_array_index][buffer_coord.x + 4][buffer_coord.y + 4];
+
+		float avg = mip_smart_average(vec4(sample_00, sample_01, sample_10, sample_11));
+		imageStore(dest_image3, ivec3(p_output_coord.x, p_output_coord.y, depth_array_index), vec4(avg));
+	}
+}
+#else
+#ifndef USE_HALF_BUFFERS
+void prepare_depths(vec4 p_samples, uvec2 p_tid) {
+	p_samples = screen_space_to_view_space_depth(p_samples);
+
+	imageStore(dest_image0, ivec3(p_tid, 0), vec4(p_samples.w));
+	imageStore(dest_image0, ivec3(p_tid, 1), vec4(p_samples.z));
+	imageStore(dest_image0, ivec3(p_tid, 2), vec4(p_samples.x));
+	imageStore(dest_image0, ivec3(p_tid, 3), vec4(p_samples.y));
+}
+#endif
+#endif
+
+void main() {
+#ifdef USE_HALF_BUFFERS
+#ifdef USE_HALF_SIZE
+	float sample_00 = texelFetch(source_depth, ivec2(4 * gl_GlobalInvocationID.x + 0, 4 * gl_GlobalInvocationID.y + 0), 0).x;
+	float sample_11 = texelFetch(source_depth, ivec2(4 * gl_GlobalInvocationID.x + 2, 4 * gl_GlobalInvocationID.y + 2), 0).x;
+#else
+	float sample_00 = texelFetch(source_depth, ivec2(2 * gl_GlobalInvocationID.x + 0, 2 * gl_GlobalInvocationID.y + 0), 0).x;
+	float sample_11 = texelFetch(source_depth, ivec2(2 * gl_GlobalInvocationID.x + 1, 2 * gl_GlobalInvocationID.y + 1), 0).x;
+#endif
+	sample_00 = screen_space_to_view_space_depth(sample_00);
+	sample_11 = screen_space_to_view_space_depth(sample_11);
+
+	imageStore(dest_image0, ivec3(gl_GlobalInvocationID.xy, 0), vec4(sample_00));
+	imageStore(dest_image0, ivec3(gl_GlobalInvocationID.xy, 3), vec4(sample_11));
+#else //!USE_HALF_BUFFERS
+#ifdef USE_HALF_SIZE
+	ivec2 depth_buffer_coord = 4 * ivec2(gl_GlobalInvocationID.xy);
+	ivec2 output_coord = ivec2(gl_GlobalInvocationID);
+
+	vec2 uv = (vec2(depth_buffer_coord) + 0.5f) * params.pixel_size;
+	vec4 samples;
+	samples.x = textureLodOffset(source_depth, uv, 0, ivec2(0, 2)).x;
+	samples.y = textureLodOffset(source_depth, uv, 0, ivec2(2, 2)).x;
+	samples.z = textureLodOffset(source_depth, uv, 0, ivec2(2, 0)).x;
+	samples.w = textureLodOffset(source_depth, uv, 0, ivec2(0, 0)).x;
+#else
+	ivec2 depth_buffer_coord = 2 * ivec2(gl_GlobalInvocationID.xy);
+	ivec2 output_coord = ivec2(gl_GlobalInvocationID);
+
+	vec2 uv = (vec2(depth_buffer_coord) + 0.5f) * params.pixel_size;
+	vec4 samples = textureGather(source_depth, uv);
+#endif
+#ifdef GENERATE_MIPS
+	prepare_depths_and_mips(samples, output_coord, gl_LocalInvocationID.xy);
+#else
+	prepare_depths(samples, gl_GlobalInvocationID.xy);
+#endif
+#endif
+}
diff --git a/servers/rendering/renderer_rd/shaders/ssao_importance_map.glsl b/servers/rendering/renderer_rd/shaders/ssao_importance_map.glsl
new file mode 100644
index 0000000000..6aa7624261
--- /dev/null
+++ b/servers/rendering/renderer_rd/shaders/ssao_importance_map.glsl
@@ -0,0 +1,126 @@
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// Copyright (c) 2016, Intel Corporation
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated
+// documentation files (the "Software"), to deal in the Software without restriction, including without limitation
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to
+// permit persons to whom the Software is furnished to do so, subject to the following conditions:
+// The above copyright notice and this permission notice shall be included in all copies or substantial portions of
+// the Software.
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+// TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// File changes (yyyy-mm-dd)
+// 2016-09-07: filip.strugar@intel.com: first commit
+// 2020-12-05: clayjohn: convert to Vulkan and Godot
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+
+#[compute]
+
+#version 450
+
+VERSION_DEFINES
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+#ifdef GENERATE_MAP
+layout(set = 0, binding = 0) uniform sampler2DArray source_ssao;
+#else
+layout(set = 0, binding = 0) uniform sampler2D source_importance;
+#endif
+layout(r8, set = 1, binding = 0) uniform restrict writeonly image2D dest_image;
+
+#ifdef PROCESS_MAPB
+layout(set = 2, binding = 0, std430) buffer Counter {
+	uint sum;
+}
+counter;
+#endif
+
+layout(push_constant, binding = 1, std430) uniform Params {
+	vec2 half_screen_pixel_size;
+	float intensity;
+	float power;
+}
+params;
+
+void main() {
+	// Pixel being shaded
+	ivec2 ssC = ivec2(gl_GlobalInvocationID.xy);
+
+#ifdef GENERATE_MAP
+	// importance map stuff
+	uvec2 base_position = ssC * 2;
+
+	vec2 base_uv = (vec2(base_position) + vec2(0.5f, 0.5f)) * params.half_screen_pixel_size;
+
+	float avg = 0.0;
+	float minV = 1.0;
+	float maxV = 0.0;
+	for (int i = 0; i < 4; i++) {
+		vec4 vals = textureGather(source_ssao, vec3(base_uv, i));
+
+		// apply the same modifications that would have been applied in the main shader
+		vals = params.intensity * vals;
+
+		vals = 1 - vals;
+
+		vals = pow(clamp(vals, 0.0, 1.0), vec4(params.power));
+
+		avg += dot(vec4(vals.x, vals.y, vals.z, vals.w), vec4(1.0 / 16.0, 1.0 / 16.0, 1.0 / 16.0, 1.0 / 16.0));
+
+		maxV = max(maxV, max(max(vals.x, vals.y), max(vals.z, vals.w)));
+		minV = min(minV, min(min(vals.x, vals.y), min(vals.z, vals.w)));
+	}
+
+	float min_max_diff = maxV - minV;
+
+	imageStore(dest_image, ssC, vec4(pow(clamp(min_max_diff * 2.0, 0.0, 1.0), 0.8)));
+#endif
+
+#ifdef PROCESS_MAPA
+	vec2 uv = (vec2(ssC) + 0.5f) * params.half_screen_pixel_size * 2.0;
+
+	float centre = textureLod(source_importance, uv, 0.0).x;
+
+	vec2 half_pixel = params.half_screen_pixel_size;
+
+	vec4 vals;
+	vals.x = textureLod(source_importance, uv + vec2(-half_pixel.x * 3, -half_pixel.y), 0.0).x;
+	vals.y = textureLod(source_importance, uv + vec2(+half_pixel.x, -half_pixel.y * 3), 0.0).x;
+	vals.z = textureLod(source_importance, uv + vec2(+half_pixel.x * 3, +half_pixel.y), 0.0).x;
+	vals.w = textureLod(source_importance, uv + vec2(-half_pixel.x, +half_pixel.y * 3), 0.0).x;
+
+	float avg = dot(vals, vec4(0.25, 0.25, 0.25, 0.25));
+
+	imageStore(dest_image, ssC, vec4(avg));
+#endif
+
+#ifdef PROCESS_MAPB
+	vec2 uv = (vec2(ssC) + 0.5f) * params.half_screen_pixel_size * 2.0;
+
+	float centre = textureLod(source_importance, uv, 0.0).x;
+
+	vec2 half_pixel = params.half_screen_pixel_size;
+
+	vec4 vals;
+	vals.x = textureLod(source_importance, uv + vec2(-half_pixel.x, -half_pixel.y * 3), 0.0).x;
+	vals.y = textureLod(source_importance, uv + vec2(+half_pixel.x * 3, -half_pixel.y), 0.0).x;
+	vals.z = textureLod(source_importance, uv + vec2(+half_pixel.x, +half_pixel.y * 3), 0.0).x;
+	vals.w = textureLod(source_importance, uv + vec2(-half_pixel.x * 3, +half_pixel.y), 0.0).x;
+
+	float avg = dot(vals, vec4(0.25, 0.25, 0.25, 0.25));
+
+	imageStore(dest_image, ssC, vec4(avg));
+
+	// sum the average; to avoid overflowing we assume max AO resolution is not bigger than 16384x16384; so quarter res (used here) will be 4096x4096, which leaves us with 8 bits per pixel
+	uint sum = uint(clamp(avg, 0.0, 1.0) * 255.0 + 0.5);
+
+	// save every 9th to avoid InterlockedAdd congestion - since we're blurring, this is good enough; compensated by multiplying load_counter_avg_div by 9
+	if (((ssC.x % 3) + (ssC.y % 3)) == 0) {
+		atomicAdd(counter.sum, sum);
+	}
+#endif
+}
diff --git a/servers/rendering/renderer_rd/shaders/ssao_interleave.glsl b/servers/rendering/renderer_rd/shaders/ssao_interleave.glsl
new file mode 100644
index 0000000000..4fdf334aa5
--- /dev/null
+++ b/servers/rendering/renderer_rd/shaders/ssao_interleave.glsl
@@ -0,0 +1,119 @@
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// Copyright (c) 2016, Intel Corporation
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated
+// documentation files (the "Software"), to deal in the Software without restriction, including without limitation
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to
+// permit persons to whom the Software is furnished to do so, subject to the following conditions:
+// The above copyright notice and this permission notice shall be included in all copies or substantial portions of
+// the Software.
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+// TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// File changes (yyyy-mm-dd)
+// 2016-09-07: filip.strugar@intel.com: first commit
+// 2020-12-05: clayjohn: convert to Vulkan and Godot
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+#[compute]
+
+#version 450
+
+VERSION_DEFINES
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+layout(rgba8, set = 0, binding = 0) uniform restrict writeonly image2D dest_image;
+layout(set = 1, binding = 0) uniform sampler2DArray source_texture;
+
+layout(push_constant, binding = 1, std430) uniform Params {
+	float inv_sharpness;
+	uint size_modifier;
+	vec2 pixel_size;
+}
+params;
+
+vec4 unpack_edges(float p_packed_val) {
+	uint packed_val = uint(p_packed_val * 255.5);
+	vec4 edgesLRTB;
+	edgesLRTB.x = float((packed_val >> 6) & 0x03) / 3.0;
+	edgesLRTB.y = float((packed_val >> 4) & 0x03) / 3.0;
+	edgesLRTB.z = float((packed_val >> 2) & 0x03) / 3.0;
+	edgesLRTB.w = float((packed_val >> 0) & 0x03) / 3.0;
+
+	return clamp(edgesLRTB + params.inv_sharpness, 0.0, 1.0);
+}
+
+void main() {
+	ivec2 ssC = ivec2(gl_GlobalInvocationID.xy);
+	if (any(greaterThanEqual(ssC, ivec2(1.0 / params.pixel_size)))) { //too large, do nothing
+		return;
+	}
+
+#ifdef MODE_SMART
+	float ao;
+	uvec2 pix_pos = uvec2(gl_GlobalInvocationID.xy);
+	vec2 uv = (gl_GlobalInvocationID.xy + vec2(0.5)) * params.pixel_size;
+
+	// calculate index in the four deinterleaved source array texture
+	int mx = int(pix_pos.x % 2);
+	int my = int(pix_pos.y % 2);
+	int index_center = mx + my * 2; // center index
+	int index_horizontal = (1 - mx) + my * 2; // neighbouring, horizontal
+	int index_vertical = mx + (1 - my) * 2; // neighbouring, vertical
+	int index_diagonal = (1 - mx) + (1 - my) * 2; // diagonal
+
+	vec2 center_val = texelFetch(source_texture, ivec3(pix_pos / uvec2(params.size_modifier), index_center), 0).xy;
+
+	ao = center_val.x;
+
+	vec4 edgesLRTB = unpack_edges(center_val.y);
+
+	// convert index shifts to sampling offsets
+	float fmx = float(mx);
+	float fmy = float(my);
+
+	// in case of an edge, push sampling offsets away from the edge (towards pixel center)
+	float fmxe = (edgesLRTB.y - edgesLRTB.x);
+	float fmye = (edgesLRTB.w - edgesLRTB.z);
+
+	// calculate final sampling offsets and sample using bilinear filter
+	vec2 uv_horizontal = (gl_GlobalInvocationID.xy + vec2(0.5) + vec2(fmx + fmxe - 0.5, 0.5 - fmy)) * params.pixel_size;
+	float ao_horizontal = textureLod(source_texture, vec3(uv_horizontal, index_horizontal), 0.0).x;
+	vec2 uv_vertical = (gl_GlobalInvocationID.xy + vec2(0.5) + vec2(0.5 - fmx, fmy - 0.5 + fmye)) * params.pixel_size;
+	float ao_vertical = textureLod(source_texture, vec3(uv_vertical, index_vertical), 0.0).x;
+	vec2 uv_diagonal = (gl_GlobalInvocationID.xy + vec2(0.5) + vec2(fmx - 0.5 + fmxe, fmy - 0.5 + fmye)) * params.pixel_size;
+	float ao_diagonal = textureLod(source_texture, vec3(uv_diagonal, index_diagonal), 0.0).x;
+
+	// reduce weight for samples near edge - if the edge is on both sides, weight goes to 0
+	vec4 blendWeights;
+	blendWeights.x = 1.0;
+	blendWeights.y = (edgesLRTB.x + edgesLRTB.y) * 0.5;
+	blendWeights.z = (edgesLRTB.z + edgesLRTB.w) * 0.5;
+	blendWeights.w = (blendWeights.y + blendWeights.z) * 0.5;
+
+	// calculate weighted average
+	float blendWeightsSum = dot(blendWeights, vec4(1.0, 1.0, 1.0, 1.0));
+	ao = dot(vec4(ao, ao_horizontal, ao_vertical, ao_diagonal), blendWeights) / blendWeightsSum;
+
+	imageStore(dest_image, ivec2(gl_GlobalInvocationID.xy), vec4(ao));
+#else // !MODE_SMART
+
+	vec2 uv = (gl_GlobalInvocationID.xy + vec2(0.5)) * params.pixel_size;
+#ifdef MODE_HALF
+	float a = textureLod(source_texture, vec3(uv, 0), 0.0).x;
+	float d = textureLod(source_texture, vec3(uv, 3), 0.0).x;
+	float avg = (a + d) * 0.5;
+
+#else
+	float a = textureLod(source_texture, vec3(uv, 0), 0.0).x;
+	float b = textureLod(source_texture, vec3(uv, 1), 0.0).x;
+	float c = textureLod(source_texture, vec3(uv, 2), 0.0).x;
+	float d = textureLod(source_texture, vec3(uv, 3), 0.0).x;
+	float avg = (a + b + c + d) * 0.25;
+
+#endif
+	imageStore(dest_image, ivec2(gl_GlobalInvocationID.xy), vec4(avg));
+#endif
+}
diff --git a/servers/rendering/renderer_rd/shaders/ssao_minify.glsl b/servers/rendering/renderer_rd/shaders/ssao_minify.glsl
deleted file mode 100644
index 263fca386f..0000000000
--- a/servers/rendering/renderer_rd/shaders/ssao_minify.glsl
+++ /dev/null
@@ -1,45 +0,0 @@
-#[compute]
-
-#version 450
-
-VERSION_DEFINES
-
-layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
-
-layout(push_constant, binding = 1, std430) uniform Params {
-	vec2 pixel_size;
-	float z_far;
-	float z_near;
-	ivec2 source_size;
-	bool orthogonal;
-	uint pad;
-}
-params;
-
-#ifdef MINIFY_START
-layout(set = 0, binding = 0) uniform sampler2D source_texture;
-#else
-layout(r32f, set = 0, binding = 0) uniform restrict readonly image2D source_image;
-#endif
-layout(r32f, set = 1, binding = 0) uniform restrict writeonly image2D dest_image;
-
-void main() {
-	ivec2 pos = ivec2(gl_GlobalInvocationID.xy);
-
-	if (any(greaterThan(pos, params.source_size >> 1))) { //too large, do nothing
-		return;
-	}
-
-#ifdef MINIFY_START
-	float depth = texelFetch(source_texture, pos << 1, 0).r * 2.0 - 1.0;
-	if (params.orthogonal) {
-		depth = ((depth + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / 2.0;
-	} else {
-		depth = 2.0 * params.z_near * params.z_far / (params.z_far + params.z_near - depth * (params.z_far - params.z_near));
-	}
-#else
-	float depth = imageLoad(source_image, pos << 1).r;
-#endif
-
-	imageStore(dest_image, pos, vec4(depth));
-}