Optimize and fix backbuffer gaussian blur

servers/rendering/renderer_rd/effects_rd.cpp

@@ -482,12 +482,11 @@ void EffectsRD::set_color(RID p_dest_texture, const Color &p_color, const Rect2i
 	RD::get_singleton()->compute_list_end();
 }
 
-void EffectsRD::gaussian_blur(RID p_source_rd_texture, RID p_texture, RID p_back_texture, const Rect2i &p_region, bool p_8bit_dst) {
+void EffectsRD::gaussian_blur(RID p_source_rd_texture, RID p_texture, const Rect2i &p_region, bool p_8bit_dst) {
 	ERR_FAIL_COND_MSG(prefer_raster_effects, "Can't use the compute version of the gaussian blur with the mobile renderer.");
 
 	memset(&copy.push_constant, 0, sizeof(CopyPushConstant));
 
-	uint32_t base_flags = 0;
 	copy.push_constant.section[0] = p_region.position.x;
 	copy.push_constant.section[1] = p_region.position.y;
 	copy.push_constant.section[2] = p_region.size.width;
@@ -497,23 +496,12 @@ void EffectsRD::gaussian_blur(RID p_source_rd_texture, RID p_texture, RID p_back
 	RD::DrawListID compute_list = RD::get_singleton()->compute_list_begin();
 	RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, copy.pipelines[p_8bit_dst ? COPY_MODE_GAUSSIAN_COPY_8BIT : COPY_MODE_GAUSSIAN_COPY]);
 	RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_source_rd_texture), 0);
-	RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_back_texture), 3);
-
-	copy.push_constant.flags = base_flags | COPY_FLAG_HORIZONTAL;
-	RD::get_singleton()->compute_list_set_push_constant(compute_list, &copy.push_constant, sizeof(CopyPushConstant));
-
-	RD::get_singleton()->compute_list_dispatch_threads(compute_list, p_region.size.width, p_region.size.height, 1);
-
-	RD::get_singleton()->compute_list_add_barrier(compute_list);
-
-	//VERTICAL
-	RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_back_texture), 0);
 	RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_texture), 3);
 
-	copy.push_constant.flags = base_flags;
 	RD::get_singleton()->compute_list_set_push_constant(compute_list, &copy.push_constant, sizeof(CopyPushConstant));
 
 	RD::get_singleton()->compute_list_dispatch_threads(compute_list, p_region.size.width, p_region.size.height, 1);
+
 	RD::get_singleton()->compute_list_end();
 }
 
@@ -2344,8 +2332,8 @@ EffectsRD::EffectsRD(bool p_prefer_raster_effects) {
 		Vector<String> copy_modes;
 		copy_modes.push_back("\n#define MODE_GAUSSIAN_BLUR\n");
 		copy_modes.push_back("\n#define MODE_GAUSSIAN_BLUR\n#define DST_IMAGE_8BIT\n");
-		copy_modes.push_back("\n#define MODE_GAUSSIAN_GLOW\n");
-		copy_modes.push_back("\n#define MODE_GAUSSIAN_GLOW\n#define GLOW_USE_AUTO_EXPOSURE\n");
+		copy_modes.push_back("\n#define MODE_GAUSSIAN_BLUR\n#define MODE_GLOW\n");
+		copy_modes.push_back("\n#define MODE_GAUSSIAN_BLUR\n#define MODE_GLOW\n#define GLOW_USE_AUTO_EXPOSURE\n");
 		copy_modes.push_back("\n#define MODE_SIMPLE_COPY\n");
 		copy_modes.push_back("\n#define MODE_SIMPLE_COPY\n#define DST_IMAGE_8BIT\n");
 		copy_modes.push_back("\n#define MODE_SIMPLE_COPY_DEPTH\n");

servers/rendering/renderer_rd/effects_rd.h

@@ -899,7 +899,7 @@ public:
 	void copy_depth_to_rect(RID p_source_rd_texture, RID p_dest_framebuffer, const Rect2i &p_rect, bool p_flip_y = false);
 	void copy_depth_to_rect_and_linearize(RID p_source_rd_texture, RID p_dest_texture, const Rect2i &p_rect, bool p_flip_y, float p_z_near, float p_z_far);
 	void copy_to_atlas_fb(RID p_source_rd_texture, RID p_dest_framebuffer, const Rect2 &p_uv_rect, RD::DrawListID p_draw_list, bool p_flip_y = false, bool p_panorama = false);
-	void gaussian_blur(RID p_source_rd_texture, RID p_texture, RID p_back_texture, const Rect2i &p_region, bool p_8bit_dst = false);
+	void gaussian_blur(RID p_source_rd_texture, RID p_texture, const Rect2i &p_region, bool p_8bit_dst = false);
 	void set_color(RID p_dest_texture, const Color &p_color, const Rect2i &p_region, bool p_8bit_dst = false);
 	void gaussian_glow(RID p_source_rd_texture, RID p_back_texture, const Size2i &p_size, float p_strength = 1.0, bool p_high_quality = false, bool p_first_pass = false, float p_luminance_cap = 16.0, float p_exposure = 1.0, float p_bloom = 0.0, float p_hdr_bleed_threshold = 1.0, float p_hdr_bleed_scale = 1.0, RID p_auto_exposure = RID(), float p_auto_exposure_grey = 1.0);
 	void gaussian_glow_raster(RID p_source_rd_texture, RID p_framebuffer_half, RID p_rd_texture_half, RID p_dest_framebuffer, const Vector2 &p_pixel_size, float p_strength = 1.0, bool p_high_quality = false, bool p_first_pass = false, float p_luminance_cap = 16.0, float p_exposure = 1.0, float p_bloom = 0.0, float p_hdr_bleed_threshold = 1.0, float p_hdr_bleed_scale = 1.0, RID p_auto_exposure = RID(), float p_auto_exposure_grey = 1.0);

servers/rendering/renderer_rd/renderer_storage_rd.cpp

@@ -7516,10 +7516,6 @@ void RendererStorageRD::_clear_render_target(RenderTarget *rt) {
 	if (rt->backbuffer.is_valid()) {
 		RD::get_singleton()->free(rt->backbuffer);
 		rt->backbuffer = RID();
-		for (int i = 0; i < rt->backbuffer_mipmaps.size(); i++) {
-			//just erase copies, since the rest are erased by dependency
-			RD::get_singleton()->free(rt->backbuffer_mipmaps[i].mipmap_copy);
-		}
 		rt->backbuffer_mipmaps.clear();
 		rt->backbuffer_uniform_set = RID(); //chain deleted
 	}
@@ -7636,7 +7632,9 @@ void RendererStorageRD::_create_render_target_backbuffer(RenderTarget *rt) {
 	tf.mipmaps = mipmaps_required;
 
 	rt->backbuffer = RD::get_singleton()->texture_create(tf, RD::TextureView());
+	RD::get_singleton()->set_resource_name(rt->backbuffer, "Render Target Back Buffer");
 	rt->backbuffer_mipmap0 = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), rt->backbuffer, 0, 0);
+	RD::get_singleton()->set_resource_name(rt->backbuffer_mipmap0, "Back Buffer slice mipmap 0");
 
 	{
 		Vector<RID> fb_tex;
@@ -7651,23 +7649,10 @@ void RendererStorageRD::_create_render_target_backbuffer(RenderTarget *rt) {
 	}
 	//create mipmaps
 	for (uint32_t i = 1; i < mipmaps_required; i++) {
-		RenderTarget::BackbufferMipmap mm;
-		{
-			mm.mipmap = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), rt->backbuffer, 0, i);
-		}
-
-		{
-			Size2 mm_size = Image::get_image_mipmap_size(tf.width, tf.height, Image::FORMAT_RGBA8, i);
+		RID mipmap = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), rt->backbuffer, 0, i);
+		RD::get_singleton()->set_resource_name(mipmap, "Back Buffer slice mip: " + itos(i));
 
-			RD::TextureFormat mmtf = tf;
-			mmtf.width = mm_size.width;
-			mmtf.height = mm_size.height;
-			mmtf.mipmaps = 1;
-
-			mm.mipmap_copy = RD::get_singleton()->texture_create(mmtf, RD::TextureView());
-		}
-
-		rt->backbuffer_mipmaps.push_back(mm);
+		rt->backbuffer_mipmaps.push_back(mipmap);
 	}
 }
 
@@ -8111,7 +8096,7 @@ void RendererStorageRD::render_target_copy_to_back_buffer(RID p_render_target, c
 	if (!p_gen_mipmaps) {
 		return;
 	}
-
+	RD::get_singleton()->draw_command_begin_label("Gaussian Blur Mipmaps");
 	//then mipmap blur
 	RID prev_texture = rt->color; //use color, not backbuffer, as bb has mipmaps.
 
@@ -8121,10 +8106,11 @@ void RendererStorageRD::render_target_copy_to_back_buffer(RID p_render_target, c
 		region.size.x = MAX(1, region.size.x >> 1);
 		region.size.y = MAX(1, region.size.y >> 1);
 
-		const RenderTarget::BackbufferMipmap &mm = rt->backbuffer_mipmaps[i];
-		effects->gaussian_blur(prev_texture, mm.mipmap, mm.mipmap_copy, region, true);
-		prev_texture = mm.mipmap;
+		RID mipmap = rt->backbuffer_mipmaps[i];
+		effects->gaussian_blur(prev_texture, mipmap, region, true);
+		prev_texture = mipmap;
 	}
+	RD::get_singleton()->draw_command_end_label();
 }
 
 void RendererStorageRD::render_target_clear_back_buffer(RID p_render_target, const Rect2i &p_region, const Color &p_color) {
@@ -8164,7 +8150,7 @@ void RendererStorageRD::render_target_gen_back_buffer_mipmaps(RID p_render_targe
 			return; //nothing to do
 		}
 	}
-
+	RD::get_singleton()->draw_command_begin_label("Gaussian Blur Mipmaps2");
 	//then mipmap blur
 	RID prev_texture = rt->backbuffer_mipmap0;
 
@@ -8174,10 +8160,11 @@ void RendererStorageRD::render_target_gen_back_buffer_mipmaps(RID p_render_targe
 		region.size.x = MAX(1, region.size.x >> 1);
 		region.size.y = MAX(1, region.size.y >> 1);
 
-		const RenderTarget::BackbufferMipmap &mm = rt->backbuffer_mipmaps[i];
-		effects->gaussian_blur(prev_texture, mm.mipmap, mm.mipmap_copy, region, true);
-		prev_texture = mm.mipmap;
+		RID mipmap = rt->backbuffer_mipmaps[i];
+		effects->gaussian_blur(prev_texture, mipmap, region, true);
+		prev_texture = mipmap;
 	}
+	RD::get_singleton()->draw_command_end_label();
 }
 
 RID RendererStorageRD::render_target_get_framebuffer_uniform_set(RID p_render_target) {

servers/rendering/renderer_rd/renderer_storage_rd.h

@@ -1169,12 +1169,7 @@ private:
 		RID backbuffer_fb;
 		RID backbuffer_mipmap0;
 
-		struct BackbufferMipmap {
-			RID mipmap;
-			RID mipmap_copy;
-		};
-
-		Vector<BackbufferMipmap> backbuffer_mipmaps;
+		Vector<RID> backbuffer_mipmaps;
 
 		RID framebuffer_uniform_set;
 		RID backbuffer_uniform_set;

servers/rendering/renderer_rd/shaders/copy.glsl

@@ -61,7 +61,7 @@ layout(rgba8, set = 3, binding = 0) uniform restrict writeonly image2D dest_buff
 layout(rgba32f, set = 3, binding = 0) uniform restrict writeonly image2D dest_buffer;
 #endif
 
-#ifdef MODE_GAUSSIAN_GLOW
+#ifdef MODE_GAUSSIAN_BLUR
 shared vec4 local_cache[256];
 shared vec4 temp_cache[128];
 #endif
@@ -70,7 +70,7 @@ void main() {
 	// Pixel being shaded
 	ivec2 pos = ivec2(gl_GlobalInvocationID.xy);
 
-#ifndef MODE_GAUSSIAN_GLOW // Glow needs the extra threads
+#ifndef MODE_GAUSSIAN_BLUR // Gaussian blur needs the extra threads
 	if (any(greaterThanEqual(pos, params.section.zw))) { //too large, do nothing
 		return;
 	}
@@ -92,35 +92,11 @@ void main() {
 
 #ifdef MODE_GAUSSIAN_BLUR
 
-	//Simpler blur uses SIGMA2 for the gaussian kernel for a stronger effect
-
-	if (bool(params.flags & FLAG_HORIZONTAL)) {
-		ivec2 base_pos = (pos + params.section.xy) << 1;
-		vec4 color = texelFetch(source_color, base_pos + ivec2(0, 0), 0) * 0.214607;
-		color += texelFetch(source_color, base_pos + ivec2(1, 0), 0) * 0.189879;
-		color += texelFetch(source_color, base_pos + ivec2(2, 0), 0) * 0.131514;
-		color += texelFetch(source_color, base_pos + ivec2(3, 0), 0) * 0.071303;
-		color += texelFetch(source_color, base_pos + ivec2(-1, 0), 0) * 0.189879;
-		color += texelFetch(source_color, base_pos + ivec2(-2, 0), 0) * 0.131514;
-		color += texelFetch(source_color, base_pos + ivec2(-3, 0), 0) * 0.071303;
-		imageStore(dest_buffer, pos + params.target, color);
-	} else {
-		ivec2 base_pos = (pos + params.section.xy);
-		vec4 color = texelFetch(source_color, base_pos + ivec2(0, 0), 0) * 0.38774;
-		color += texelFetch(source_color, base_pos + ivec2(0, 1), 0) * 0.24477;
-		color += texelFetch(source_color, base_pos + ivec2(0, 2), 0) * 0.06136;
-		color += texelFetch(source_color, base_pos + ivec2(0, -1), 0) * 0.24477;
-		color += texelFetch(source_color, base_pos + ivec2(0, -2), 0) * 0.06136;
-		imageStore(dest_buffer, pos + params.target, color);
-	}
-#endif
-
-#ifdef MODE_GAUSSIAN_GLOW
-
 	// First pass copy texture into 16x16 local memory for every 8x8 thread block
 	vec2 quad_center_uv = clamp(vec2(gl_GlobalInvocationID.xy + gl_LocalInvocationID.xy - 3.5) / params.section.zw, vec2(0.5 / params.section.zw), vec2(1.0 - 1.5 / params.section.zw));
 	uint dest_index = gl_LocalInvocationID.x * 2 + gl_LocalInvocationID.y * 2 * 16;
 
+#ifdef MODE_GLOW
 	if (bool(params.flags & FLAG_HIGH_QUALITY_GLOW)) {
 		vec2 quad_offset_uv = clamp((vec2(gl_GlobalInvocationID.xy + gl_LocalInvocationID.xy - 3.0)) / params.section.zw, vec2(0.5 / params.section.zw), vec2(1.0 - 1.5 / params.section.zw));
 
@@ -128,12 +104,15 @@ void main() {
 		local_cache[dest_index + 1] = (textureLod(source_color, quad_center_uv + vec2(1.0 / params.section.z, 0.0), 0) + textureLod(source_color, quad_offset_uv + vec2(1.0 / params.section.z, 0.0), 0)) * 0.5;
 		local_cache[dest_index + 16] = (textureLod(source_color, quad_center_uv + vec2(0.0, 1.0 / params.section.w), 0) + textureLod(source_color, quad_offset_uv + vec2(0.0, 1.0 / params.section.w), 0)) * 0.5;
 		local_cache[dest_index + 16 + 1] = (textureLod(source_color, quad_center_uv + vec2(1.0 / params.section.zw), 0) + textureLod(source_color, quad_offset_uv + vec2(1.0 / params.section.zw), 0)) * 0.5;
-	} else {
+	} else
+#endif
+	{
 		local_cache[dest_index] = textureLod(source_color, quad_center_uv, 0);
 		local_cache[dest_index + 1] = textureLod(source_color, quad_center_uv + vec2(1.0 / params.section.z, 0.0), 0);
 		local_cache[dest_index + 16] = textureLod(source_color, quad_center_uv + vec2(0.0, 1.0 / params.section.w), 0);
 		local_cache[dest_index + 16 + 1] = textureLod(source_color, quad_center_uv + vec2(1.0 / params.section.zw), 0);
 	}
+#ifdef MODE_GLOW
 	if (bool(params.flags & FLAG_GLOW_FIRST_PASS)) {
 		// Tonemap initial samples to reduce weight of fireflies: https://graphicrants.blogspot.com/2013/12/tone-mapping.html
 		local_cache[dest_index] /= 1.0 + dot(local_cache[dest_index].rgb, vec3(0.299, 0.587, 0.114));
@@ -141,29 +120,33 @@ void main() {
 		local_cache[dest_index + 16] /= 1.0 + dot(local_cache[dest_index + 16].rgb, vec3(0.299, 0.587, 0.114));
 		local_cache[dest_index + 16 + 1] /= 1.0 + dot(local_cache[dest_index + 16 + 1].rgb, vec3(0.299, 0.587, 0.114));
 	}
-
+	const float kernel[4] = { 0.174938, 0.165569, 0.140367, 0.106595 };
+#else
+	// Simpler blur uses SIGMA2 for the gaussian kernel for a stronger effect.
+	const float kernel[4] = { 0.214607, 0.189879, 0.131514, 0.071303 };
+#endif
 	memoryBarrierShared();
 	barrier();
 
 	// Horizontal pass. Needs to copy into 8x16 chunk of local memory so vertical pass has full resolution
 	uint read_index = gl_LocalInvocationID.x + gl_LocalInvocationID.y * 32 + 4;
 	vec4 color_top = vec4(0.0);
-	color_top += local_cache[read_index] * 0.174938;
-	color_top += local_cache[read_index + 1] * 0.165569;
-	color_top += local_cache[read_index + 2] * 0.140367;
-	color_top += local_cache[read_index + 3] * 0.106595;
-	color_top += local_cache[read_index - 1] * 0.165569;
-	color_top += local_cache[read_index - 2] * 0.140367;
-	color_top += local_cache[read_index - 3] * 0.106595;
+	color_top += local_cache[read_index] * kernel[0];
+	color_top += local_cache[read_index + 1] * kernel[1];
+	color_top += local_cache[read_index + 2] * kernel[2];
+	color_top += local_cache[read_index + 3] * kernel[3];
+	color_top += local_cache[read_index - 1] * kernel[1];
+	color_top += local_cache[read_index - 2] * kernel[2];
+	color_top += local_cache[read_index - 3] * kernel[3];
 
 	vec4 color_bottom = vec4(0.0);
-	color_bottom += local_cache[read_index + 16] * 0.174938;
-	color_bottom += local_cache[read_index + 1 + 16] * 0.165569;
-	color_bottom += local_cache[read_index + 2 + 16] * 0.140367;
-	color_bottom += local_cache[read_index + 3 + 16] * 0.106595;
-	color_bottom += local_cache[read_index - 1 + 16] * 0.165569;
-	color_bottom += local_cache[read_index - 2 + 16] * 0.140367;
-	color_bottom += local_cache[read_index - 3 + 16] * 0.106595;
+	color_bottom += local_cache[read_index + 16] * kernel[0];
+	color_bottom += local_cache[read_index + 1 + 16] * kernel[1];
+	color_bottom += local_cache[read_index + 2 + 16] * kernel[2];
+	color_bottom += local_cache[read_index + 3 + 16] * kernel[3];
+	color_bottom += local_cache[read_index - 1 + 16] * kernel[1];
+	color_bottom += local_cache[read_index - 2 + 16] * kernel[2];
+	color_bottom += local_cache[read_index - 3 + 16] * kernel[3];
 
 	// rotate samples to take advantage of cache coherency
 	uint write_index = gl_LocalInvocationID.y * 2 + gl_LocalInvocationID.x * 16;
@@ -174,18 +157,24 @@ void main() {
 	memoryBarrierShared();
 	barrier();
 
+	// If destination outside of texture, can stop doing work now
+	if (any(greaterThanEqual(pos, params.section.zw))) {
+		return;
+	}
+
 	// Vertical pass
 	uint index = gl_LocalInvocationID.y + gl_LocalInvocationID.x * 16 + 4;
 	vec4 color = vec4(0.0);
 
-	color += temp_cache[index] * 0.174938;
-	color += temp_cache[index + 1] * 0.165569;
-	color += temp_cache[index + 2] * 0.140367;
-	color += temp_cache[index + 3] * 0.106595;
-	color += temp_cache[index - 1] * 0.165569;
-	color += temp_cache[index - 2] * 0.140367;
-	color += temp_cache[index - 3] * 0.106595;
+	color += temp_cache[index] * kernel[0];
+	color += temp_cache[index + 1] * kernel[1];
+	color += temp_cache[index + 2] * kernel[2];
+	color += temp_cache[index + 3] * kernel[3];
+	color += temp_cache[index - 1] * kernel[1];
+	color += temp_cache[index - 2] * kernel[2];
+	color += temp_cache[index - 3] * kernel[3];
 
+#ifdef MODE_GLOW
 	if (bool(params.flags & FLAG_GLOW_FIRST_PASS)) {
 		// Undo tonemap to restore range: https://graphicrants.blogspot.com/2013/12/tone-mapping.html
 		color /= 1.0 - dot(color.rgb, vec3(0.299, 0.587, 0.114));
@@ -205,7 +194,7 @@ void main() {
 
 		color = min(color * feedback, vec4(params.glow_luminance_cap));
 	}
-
+#endif
 	imageStore(dest_buffer, pos + params.target, color);
 
 #endif