Merge pull request #78499 from BastiaanOlij/fix_stereo_decals

Apply reprojection in multiview for our cluster lookup

servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.cpp

@@ -1451,6 +1451,11 @@ void RenderForwardClustered::_pre_opaque_render(RenderDataRD *p_render_data, boo
 	RD::get_singleton()->barrier(RD::BARRIER_MASK_ALL_BARRIERS, RD::BARRIER_MASK_ALL_BARRIERS);
 
 	if (current_cluster_builder) {
+		// Note: when rendering stereoscopic (multiview) we are using our combined frustum projection to create
+		// our cluster data. We use reprojection in the shader to adjust for our left/right eye.
+		// This only works as we don't filter our cluster by depth buffer.
+		// If we ever make this optimisation we should make it optional and only use it in mono.
+		// What we win by filtering out a few lights, we loose by having to do the work double for stereo.
 		current_cluster_builder->begin(p_render_data->scene_data->cam_transform, p_render_data->scene_data->cam_projection, !p_render_data->reflection_probe.is_valid());
 	}
 

servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered.glsl

@@ -735,6 +735,10 @@ void fragment_shader(in SceneData scene_data) {
 #ifdef USE_MULTIVIEW
 	vec3 eye_offset = scene_data.eye_offset[ViewIndex].xyz;
 	vec3 view = -normalize(vertex_interp - eye_offset);
+
+	// UV in our combined frustum space is used for certain screen uv processes where it's
+	// overkill to render separate left and right eye views
+	vec2 combined_uv = (combined_projected.xy / combined_projected.w) * 0.5 + 0.5;
 #else
 	vec3 eye_offset = vec3(0.0, 0.0, 0.0);
 	vec3 view = -normalize(vertex_interp);
@@ -921,8 +925,7 @@ void fragment_shader(in SceneData scene_data) {
 
 	if (implementation_data.volumetric_fog_enabled) {
 #ifdef USE_MULTIVIEW
-		vec2 center_uv = (combined_projected.xy / combined_projected.w) * 0.5 + 0.5;
-		vec4 volumetric_fog = volumetric_fog_process(center_uv, -vertex.z);
+		vec4 volumetric_fog = volumetric_fog_process(combined_uv, -vertex.z);
 #else
 		vec4 volumetric_fog = volumetric_fog_process(screen_uv, -vertex.z);
 #endif
@@ -952,7 +955,11 @@ void fragment_shader(in SceneData scene_data) {
 
 #ifndef MODE_RENDER_DEPTH
 
+#ifdef USE_MULTIVIEW
+	uvec2 cluster_pos = uvec2(combined_uv.xy / scene_data.screen_pixel_size) >> implementation_data.cluster_shift;
+#else
 	uvec2 cluster_pos = uvec2(gl_FragCoord.xy) >> implementation_data.cluster_shift;
+#endif
 	uint cluster_offset = (implementation_data.cluster_width * cluster_pos.y + cluster_pos.x) * (implementation_data.max_cluster_element_count_div_32 + 32);
 
 	uint cluster_z = uint(clamp((-vertex.z / scene_data.z_far) * 32.0, 0.0, 31.0));