Merge pull request #19505 from BastiaanOlij/improve_stereo_render

Improvement for stereo render, reuse culling and lighting for both eyes

2 files changed, 91 insertions, 13 deletions

diff --git a/servers/visual/visual_server_scene.cpp b/servers/visual/visual_server_scene.cpp
index 04dcde1365..43a225b370 100644
--- a/servers/visual/visual_server_scene.cpp
+++ b/servers/visual/visual_server_scene.cpp
@@ -1674,7 +1674,8 @@ void VisualServerScene::render_camera(RID p_camera, RID p_scenario, Size2 p_view
 		} break;
 	}
 
-	_render_scene(camera->transform, camera_matrix, ortho, camera->env, camera->visible_layers, p_scenario, p_shadow_atlas, RID(), -1);
+	_prepare_scene(camera->transform, camera_matrix, ortho, camera->env, camera->visible_layers, p_scenario, p_shadow_atlas, RID());
+	_render_scene(camera->transform, camera_matrix, ortho, camera->env, p_scenario, p_shadow_atlas, RID(), -1);
 }
 
 void VisualServerScene::render_camera(Ref<ARVRInterface> &p_interface, ARVRInterface::Eyes p_eye, RID p_camera, RID p_scenario, Size2 p_viewport_size, RID p_shadow_atlas) {
@@ -1684,7 +1685,6 @@ void VisualServerScene::render_camera(Ref<ARVRInterface> &p_interface, ARVRInter
 	ERR_FAIL_COND(!camera);
 
 	/* SETUP CAMERA, we are ignoring type and FOV here */
-	bool ortho = false;
 	float aspect = p_viewport_size.width / (float)p_viewport_size.height;
 	CameraMatrix camera_matrix = p_interface->get_projection_for_eye(p_eye, aspect, camera->znear, camera->zfar);
 
@@ -1693,10 +1693,79 @@ void VisualServerScene::render_camera(Ref<ARVRInterface> &p_interface, ARVRInter
 	Transform world_origin = ARVRServer::get_singleton()->get_world_origin();
 	Transform cam_transform = p_interface->get_transform_for_eye(p_eye, world_origin);
 
-	_render_scene(cam_transform, camera_matrix, ortho, camera->env, camera->visible_layers, p_scenario, p_shadow_atlas, RID(), -1);
+	// For stereo render we only prepare for our left eye and then reuse the outcome for our right eye
+	if (p_eye == ARVRInterface::EYE_LEFT) {
+		///@TODO possibly move responsibility for this into our ARVRServer or ARVRInterface?
+
+		// Center our transform, we assume basis is equal.
+		Transform mono_transform = cam_transform;
+		Transform right_transform = p_interface->get_transform_for_eye(ARVRInterface::EYE_RIGHT, world_origin);
+		mono_transform.origin += right_transform.origin;
+		mono_transform.origin *= 0.5;
+
+		// We need to combine our projection frustums for culling.
+		// Ideally we should use our clipping planes for this and combine them,
+		// however our shadow map logic uses our projection matrix.
+		// Note: as our left and right frustums should be mirrored, we don't need our right projection matrix.
+
+		// - get some base values we need
+		float eye_dist = (mono_transform.origin - cam_transform.origin).length();
+		float z_near = camera_matrix.get_z_near(); // get our near plane
+		float z_far = camera_matrix.get_z_far(); // get our far plane
+		float width = (2.0 * z_near) / camera_matrix.matrix[0][0];
+		float x_shift = width * camera_matrix.matrix[2][0];
+		float height = (2.0 * z_near) / camera_matrix.matrix[1][1];
+		float y_shift = width * camera_matrix.matrix[2][1];
+
+		// printf("Eye_dist = %f, Near = %f, Far = %f, Width = %f, Shift = %f\n", eye_dist, z_near, z_far, width, x_shift);
+
+		// - calculate our near plane size (horizontal only, right_near is mirrored)
+		float left_near = -eye_dist - ((width - x_shift) * 0.5);
+
+		// - calculate our far plane size (horizontal only, right_far is mirrored)
+		float left_far = -eye_dist - (z_far * (width - x_shift) * 0.5 / z_near);
+		float left_far_right_eye = eye_dist - (z_far * (width + x_shift) * 0.5 / z_near);
+		if (left_far > left_far_right_eye) {
+			// on displays smaller then double our iod, the right eye far frustrum can overtake the left eyes.
+			left_far = left_far_right_eye;
+		}
+
+		// - figure out required z-shift
+		float slope = (left_far - left_near) / (z_far - z_near);
+		float z_shift = (left_near / slope) - z_near;
+
+		// - figure out new vertical near plane size (this will be slightly oversized thanks to our z-shift)
+		float top_near = (height + y_shift) * 0.5;
+		top_near += y_shift * z_shift;
+		float bottom_near = -(height - y_shift) * 0.5;
+		bottom_near -= y_shift * z_shift;
+
+		// printf("Left_near = %f, Left_far = %f, Top_near = %f, Bottom_near = %f, Z_shift = %f\n", left_near, left_far, top_near, bottom_near, z_shift);
+
+		// - generate our frustum
+		CameraMatrix combined_matrix;
+		combined_matrix.set_frustum(left_near, -left_near, bottom_near, top_near, z_near + z_shift, z_far + z_shift);
+
+		// and finally move our camera back
+		Transform apply_z_shift;
+		apply_z_shift.origin = Vector3(0.0, 0.0, z_shift); // z negative is forward so this moves it backwards
+		mono_transform *= apply_z_shift;
+
+		// now prepare our scene with our adjusted transform projection matrix
+		_prepare_scene(mono_transform, combined_matrix, false, camera->env, camera->visible_layers, p_scenario, p_shadow_atlas, RID());
+	} else if (p_eye == ARVRInterface::EYE_MONO) {
+		// For mono render, prepare as per usual
+		_prepare_scene(cam_transform, camera_matrix, false, camera->env, camera->visible_layers, p_scenario, p_shadow_atlas, RID());
+	}
+
+	// And render our scene...
+	_render_scene(cam_transform, camera_matrix, false, camera->env, p_scenario, p_shadow_atlas, RID(), -1);
 };
 
-void VisualServerScene::_render_scene(const Transform p_cam_transform, const CameraMatrix &p_cam_projection, bool p_cam_orthogonal, RID p_force_environment, uint32_t p_visible_layers, RID p_scenario, RID p_shadow_atlas, RID p_reflection_probe, int p_reflection_probe_pass) {
+void VisualServerScene::_prepare_scene(const Transform p_cam_transform, const CameraMatrix &p_cam_projection, bool p_cam_orthogonal, RID p_force_environment, uint32_t p_visible_layers, RID p_scenario, RID p_shadow_atlas, RID p_reflection_probe) {
+	// Note, in stereo rendering:
+	// - p_cam_transform will be a transform in the middle of our two eyes
+	// - p_cam_projection is a wider frustrum that encompasses both eyes
 
 	Scenario *scenario = scenario_owner.getornull(p_scenario);
 
@@ -1713,7 +1782,7 @@ void VisualServerScene::_render_scene(const Transform p_cam_transform, const Cam
 	float z_far = p_cam_projection.get_z_far();
 
 	/* STEP 2 - CULL */
-	int cull_count = scenario->octree.cull_convex(planes, instance_cull_result, MAX_INSTANCE_CULL);
+	instance_cull_count = scenario->octree.cull_convex(planes, instance_cull_result, MAX_INSTANCE_CULL);
 	light_cull_count = 0;
 
 	reflection_probe_cull_count = 0;
@@ -1731,7 +1800,7 @@ void VisualServerScene::_render_scene(const Transform p_cam_transform, const Cam
 
 	/* STEP 4 - REMOVE FURTHER CULLED OBJECTS, ADD LIGHTS */
 
-	for (int i = 0; i < cull_count; i++) {
+	for (int i = 0; i < instance_cull_count; i++) {
 
 		Instance *ins = instance_cull_result[i];
 
@@ -1857,8 +1926,8 @@ void VisualServerScene::_render_scene(const Transform p_cam_transform, const Cam
 
 		if (!keep) {
 			// remove, no reason to keep
-			cull_count--;
-			SWAP(instance_cull_result[i], instance_cull_result[cull_count]);
+			instance_cull_count--;
+			SWAP(instance_cull_result[i], instance_cull_result[instance_cull_count]);
 			i--;
 			ins->last_render_pass = 0; // make invalid
 		} else {
@@ -1870,7 +1939,7 @@ void VisualServerScene::_render_scene(const Transform p_cam_transform, const Cam
 	/* STEP 5 - PROCESS LIGHTS */
 
 	RID *directional_light_ptr = &light_instance_cull_result[light_cull_count];
-	int directional_light_count = 0;
+	directional_light_count = 0;
 
 	// directional lights
 	{
@@ -2007,6 +2076,11 @@ void VisualServerScene::_render_scene(const Transform p_cam_transform, const Cam
 			}
 		}
 	}
+}
+
+void VisualServerScene::_render_scene(const Transform p_cam_transform, const CameraMatrix &p_cam_projection, bool p_cam_orthogonal, RID p_force_environment, RID p_scenario, RID p_shadow_atlas, RID p_reflection_probe, int p_reflection_probe_pass) {
+
+	Scenario *scenario = scenario_owner.getornull(p_scenario);
 
 	/* ENVIRONMENT */
 
@@ -2018,9 +2092,9 @@ void VisualServerScene::_render_scene(const Transform p_cam_transform, const Cam
 	else
 		environment = scenario->fallback_environment;
 
-	/* STEP 6 - PROCESS GEOMETRY AND DRAW SCENE*/
+	/* PROCESS GEOMETRY AND DRAW SCENE */
 
-	VSG::scene_render->render_scene(p_cam_transform, p_cam_projection, p_cam_orthogonal, (RasterizerScene::InstanceBase **)instance_cull_result, cull_count, light_instance_cull_result, light_cull_count + directional_light_count, reflection_probe_instance_cull_result, reflection_probe_cull_count, environment, p_shadow_atlas, scenario->reflection_atlas, p_reflection_probe, p_reflection_probe_pass);
+	VSG::scene_render->render_scene(p_cam_transform, p_cam_projection, p_cam_orthogonal, (RasterizerScene::InstanceBase **)instance_cull_result, instance_cull_count, light_instance_cull_result, light_cull_count + directional_light_count, reflection_probe_instance_cull_result, reflection_probe_cull_count, environment, p_shadow_atlas, scenario->reflection_atlas, p_reflection_probe, p_reflection_probe_pass);
 }
 
 void VisualServerScene::render_empty_scene(RID p_scenario, RID p_shadow_atlas) {
@@ -2093,7 +2167,8 @@ bool VisualServerScene::_render_reflection_probe_step(Instance *p_instance, int
 			shadow_atlas = scenario->reflection_probe_shadow_atlas;
 		}
 
-		_render_scene(xform, cm, false, RID(), VSG::storage->reflection_probe_get_cull_mask(p_instance->base), p_instance->scenario->self, shadow_atlas, reflection_probe->instance, p_step);
+		_prepare_scene(xform, cm, false, RID(), VSG::storage->reflection_probe_get_cull_mask(p_instance->base), p_instance->scenario->self, shadow_atlas, reflection_probe->instance);
+		_render_scene(xform, cm, false, RID(), p_instance->scenario->self, shadow_atlas, reflection_probe->instance, p_step);
 
 	} else {
 		//do roughness postprocess step until it believes it's done
diff --git a/servers/visual/visual_server_scene.h b/servers/visual/visual_server_scene.h
index 109cdf711c..12d732724a 100644
--- a/servers/visual/visual_server_scene.h
+++ b/servers/visual/visual_server_scene.h
@@ -434,11 +434,13 @@ public:
 		}
 	};
 
+	int instance_cull_count;
 	Instance *instance_cull_result[MAX_INSTANCE_CULL];
 	Instance *instance_shadow_cull_result[MAX_INSTANCE_CULL]; //used for generating shadowmaps
 	Instance *light_cull_result[MAX_LIGHTS_CULLED];
 	RID light_instance_cull_result[MAX_LIGHTS_CULLED];
 	int light_cull_count;
+	int directional_light_count;
 	RID reflection_probe_instance_cull_result[MAX_REFLECTION_PROBES_CULLED];
 	int reflection_probe_cull_count;
 
@@ -483,7 +485,8 @@ public:
 
 	_FORCE_INLINE_ void _light_instance_update_shadow(Instance *p_instance, const Transform p_cam_transform, const CameraMatrix &p_cam_projection, bool p_cam_orthogonal, RID p_shadow_atlas, Scenario *p_scenario);
 
-	void _render_scene(const Transform p_cam_transform, const CameraMatrix &p_cam_projection, bool p_cam_orthogonal, RID p_force_environment, uint32_t p_visible_layers, RID p_scenario, RID p_shadow_atlas, RID p_reflection_probe, int p_reflection_probe_pass);
+	void _prepare_scene(const Transform p_cam_transform, const CameraMatrix &p_cam_projection, bool p_cam_orthogonal, RID p_force_environment, uint32_t p_visible_layers, RID p_scenario, RID p_shadow_atlas, RID p_reflection_probe);
+	void _render_scene(const Transform p_cam_transform, const CameraMatrix &p_cam_projection, bool p_cam_orthogonal, RID p_force_environment, RID p_scenario, RID p_shadow_atlas, RID p_reflection_probe, int p_reflection_probe_pass);
 	void render_empty_scene(RID p_scenario, RID p_shadow_atlas);
 
 	void render_camera(RID p_camera, RID p_scenario, Size2 p_viewport_size, RID p_shadow_atlas);