Merge pull request #45852 from reduz/make-servers-truly-thread-safe

Make Servers truly Thread Safe

core/config/project_settings.cpp

@@ -1244,8 +1244,8 @@ ProjectSettings::ProjectSettings() {
 
 	custom_prop_info["display/window/handheld/orientation"] = PropertyInfo(Variant::STRING, "display/window/handheld/orientation", PROPERTY_HINT_ENUM, "landscape,portrait,reverse_landscape,reverse_portrait,sensor_landscape,sensor_portrait,sensor");
 	custom_prop_info["rendering/threads/thread_model"] = PropertyInfo(Variant::INT, "rendering/threads/thread_model", PROPERTY_HINT_ENUM, "Single-Unsafe,Single-Safe,Multi-Threaded");
-	custom_prop_info["physics/2d/thread_model"] = PropertyInfo(Variant::INT, "physics/2d/thread_model", PROPERTY_HINT_ENUM, "Single-Unsafe,Single-Safe,Multi-Threaded");
-	custom_prop_info["rendering/quality/intended_usage/framebuffer_allocation"] = PropertyInfo(Variant::INT, "rendering/quality/intended_usage/framebuffer_allocation", PROPERTY_HINT_ENUM, "2D,2D Without Sampling,3D,3D Without Effects");
+	GLOBAL_DEF("physics/2d/run_on_thread", false);
+	GLOBAL_DEF("physics/3d/run_on_thread", false);
 
 	GLOBAL_DEF("debug/settings/profiler/max_functions", 16384);
 	custom_prop_info["debug/settings/profiler/max_functions"] = PropertyInfo(Variant::INT, "debug/settings/profiler/max_functions", PROPERTY_HINT_RANGE, "128,65535,1");

core/templates/command_queue_mt.h

@@ -498,6 +498,11 @@ class CommandQueueMT {
 		flush_one();
 	}
 
+	_FORCE_INLINE_ void flush_if_pending() {
+		if (unlikely(read_ptr_and_epoch != write_ptr_and_epoch)) {
+			flush_all();
+		}
+	}
 	void flush_all() {
 		//ERR_FAIL_COND(sync);
 		lock();

core/templates/rid_owner.h

@@ -79,8 +79,7 @@ class RID_Alloc : public RID_AllocBase {
 
 	SpinLock spin_lock;
 
-public:
-	RID make_rid(const T &p_value) {
+	_FORCE_INLINE_ RID _allocate_rid(const T *p_initializer) {
 		if (THREAD_SAFE) {
 			spin_lock.lock();
 		}
@@ -115,15 +114,22 @@ class RID_Alloc : public RID_AllocBase {
 		uint32_t free_chunk = free_index / elements_in_chunk;
 		uint32_t free_element = free_index % elements_in_chunk;
 
-		T *ptr = &chunks[free_chunk][free_element];
-		memnew_placement(ptr, T(p_value));
+		if (p_initializer) {
+			T *ptr = &chunks[free_chunk][free_element];
+			memnew_placement(ptr, T(*p_initializer));
+		}
 
-		uint32_t validator = (uint32_t)(_gen_id() & 0xFFFFFFFF);
+		uint32_t validator = (uint32_t)(_gen_id() & 0x7FFFFFFF);
 		uint64_t id = validator;
 		id <<= 32;
 		id |= free_index;
 
 		validator_chunks[free_chunk][free_element] = validator;
+
+		if (!p_initializer) {
+			validator_chunks[free_chunk][free_element] |= 0x80000000; //mark uninitialized bit
+		}
+
 		alloc_count++;
 
 		if (THREAD_SAFE) {
@@ -133,7 +139,20 @@ class RID_Alloc : public RID_AllocBase {
 		return _make_from_id(id);
 	}
 
-	_FORCE_INLINE_ T *getornull(const RID &p_rid) {
+public:
+	RID make_rid(const T &p_value) {
+		return _allocate_rid(&p_value);
+	}
+
+	//allocate but don't initialize, use initialize_rid afterwards
+	RID allocate_rid() {
+		return _allocate_rid(nullptr);
+	}
+
+	_FORCE_INLINE_ T *getornull(const RID &p_rid, bool p_initialize = false) {
+		if (p_rid == RID()) {
+			return nullptr;
+		}
 		if (THREAD_SAFE) {
 			spin_lock.lock();
 		}
@@ -151,10 +170,32 @@ class RID_Alloc : public RID_AllocBase {
 		uint32_t idx_element = idx % elements_in_chunk;
 
 		uint32_t validator = uint32_t(id >> 32);
-		if (unlikely(validator_chunks[idx_chunk][idx_element] != validator)) {
+
+		if (unlikely(p_initialize)) {
+			if (unlikely(!(validator_chunks[idx_chunk][idx_element] & 0x80000000))) {
+				if (THREAD_SAFE) {
+					spin_lock.unlock();
+				}
+				ERR_FAIL_V_MSG(nullptr, "Initializing already initialized RID");
+			}
+
+			if (unlikely((validator_chunks[idx_chunk][idx_element] & 0x7FFFFFFF) != validator)) {
+				if (THREAD_SAFE) {
+					spin_lock.unlock();
+				}
+				ERR_FAIL_V_MSG(nullptr, "Attempting to initialize the wrong RID");
+				return nullptr;
+			}
+
+			validator_chunks[idx_chunk][idx_element] &= 0x7FFFFFFF; //initialized
+
+		} else if (unlikely(validator_chunks[idx_chunk][idx_element] != validator)) {
 			if (THREAD_SAFE) {
 				spin_lock.unlock();
 			}
+			if (validator_chunks[idx_chunk][idx_element] & 0x80000000) {
+				ERR_FAIL_V_MSG(nullptr, "Attempting to use an uninitialized RID");
+			}
 			return nullptr;
 		}
 
@@ -166,6 +207,11 @@ class RID_Alloc : public RID_AllocBase {
 
 		return ptr;
 	}
+	void initialize_rid(RID p_rid, const T &p_value) {
+		T *mem = getornull(p_rid, true);
+		ERR_FAIL_COND(!mem);
+		memnew_placement(mem, T(p_value));
+	}
 
 	_FORCE_INLINE_ bool owns(const RID &p_rid) {
 		if (THREAD_SAFE) {
@@ -186,7 +232,7 @@ class RID_Alloc : public RID_AllocBase {
 
 		uint32_t validator = uint32_t(id >> 32);
 
-		bool owned = validator_chunks[idx_chunk][idx_element] == validator;
+		bool owned = (validator_chunks[idx_chunk][idx_element] & 0x7FFFFFFF) == validator;
 
 		if (THREAD_SAFE) {
 			spin_lock.unlock();
@@ -213,7 +259,12 @@ class RID_Alloc : public RID_AllocBase {
 		uint32_t idx_element = idx % elements_in_chunk;
 
 		uint32_t validator = uint32_t(id >> 32);
-		if (unlikely(validator_chunks[idx_chunk][idx_element] != validator)) {
+		if (unlikely(validator_chunks[idx_chunk][idx_element] & 0x80000000)) {
+			if (THREAD_SAFE) {
+				spin_lock.unlock();
+			}
+			ERR_FAIL_MSG("Attempted to free an uninitialized or invalid RID");
+		} else if (unlikely(validator_chunks[idx_chunk][idx_element] != validator)) {
 			if (THREAD_SAFE) {
 				spin_lock.unlock();
 			}
@@ -330,6 +381,14 @@ class RID_PtrOwner {
 		return alloc.make_rid(p_ptr);
 	}
 
+	_FORCE_INLINE_ RID allocate_rid() {
+		return alloc.allocate_rid();
+	}
+
+	_FORCE_INLINE_ void initialize_rid(RID p_rid, T *p_ptr) {
+		alloc.initialize_rid(p_rid, p_ptr);
+	}
+
 	_FORCE_INLINE_ T *getornull(const RID &p_rid) {
 		T **ptr = alloc.getornull(p_rid);
 		if (unlikely(!ptr)) {
@@ -338,6 +397,12 @@ class RID_PtrOwner {
 		return *ptr;
 	}
 
+	_FORCE_INLINE_ void replace(const RID &p_rid, T *p_new_ptr) {
+		T **ptr = alloc.getornull(p_rid);
+		ERR_FAIL_COND(!ptr);
+		*ptr = p_new_ptr;
+	}
+
 	_FORCE_INLINE_ bool owns(const RID &p_rid) {
 		return alloc.owns(p_rid);
 	}
@@ -379,6 +444,14 @@ class RID_Owner {
 		return alloc.make_rid(p_ptr);
 	}
 
+	_FORCE_INLINE_ RID allocate_rid() {
+		return alloc.allocate_rid();
+	}
+
+	_FORCE_INLINE_ void initialize_rid(RID p_rid, const T &p_ptr) {
+		alloc.initialize_rid(p_rid, p_ptr);
+	}
+
 	_FORCE_INLINE_ T *getornull(const RID &p_rid) {
 		return alloc.getornull(p_rid);
 	}

main/main.cpp

@@ -71,7 +71,6 @@
 #include "servers/physics_server_3d.h"
 #include "servers/register_server_types.h"
 #include "servers/rendering/rendering_server_default.h"
-#include "servers/rendering/rendering_server_wrap_mt.h"
 #include "servers/text_server.h"
 #include "servers/xr_server.h"
 
@@ -1570,12 +1569,7 @@ Error Main::setup2(Thread::ID p_main_tid_override) {
 
 	/* Initialize Visual Server */
 
-	rendering_server = memnew(RenderingServerDefault);
-	if (OS::get_singleton()->get_render_thread_mode() != OS::RENDER_THREAD_UNSAFE) {
-		rendering_server = memnew(RenderingServerWrapMT(rendering_server,
-				OS::get_singleton()->get_render_thread_mode() ==
-						OS::RENDER_SEPARATE_THREAD));
-	}
+	rendering_server = memnew(RenderingServerDefault(OS::get_singleton()->get_render_thread_mode() == OS::RENDER_SEPARATE_THREAD));
 
 	rendering_server->init();
 	rendering_server->set_render_loop_enabled(!disable_render_loop);
@@ -2451,6 +2445,7 @@ bool Main::iteration() {
 	for (int iters = 0; iters < advance.physics_steps; ++iters) {
 		uint64_t physics_begin = OS::get_singleton()->get_ticks_usec();
 
+		PhysicsServer3D::get_singleton()->sync();
 		PhysicsServer3D::get_singleton()->flush_queries();
 
 		PhysicsServer2D::get_singleton()->sync();
@@ -2465,6 +2460,7 @@ bool Main::iteration() {
 
 		message_queue->flush();
 
+		PhysicsServer3D::get_singleton()->end_sync();
 		PhysicsServer3D::get_singleton()->step(physics_step * time_scale);
 
 		PhysicsServer2D::get_singleton()->end_sync();

modules/bullet/bullet_physics_server.cpp

@@ -1059,16 +1059,16 @@ real_t BulletPhysicsServer3D::soft_body_get_linear_stiffness(RID p_body) {
 	return body->get_linear_stiffness();
 }
 
-void BulletPhysicsServer3D::soft_body_set_areaAngular_stiffness(RID p_body, real_t p_stiffness) {
+void BulletPhysicsServer3D::soft_body_set_angular_stiffness(RID p_body, real_t p_stiffness) {
 	SoftBodyBullet *body = soft_body_owner.getornull(p_body);
 	ERR_FAIL_COND(!body);
-	body->set_areaAngular_stiffness(p_stiffness);
+	body->set_angular_stiffness(p_stiffness);
 }
 
-real_t BulletPhysicsServer3D::soft_body_get_areaAngular_stiffness(RID p_body) {
+real_t BulletPhysicsServer3D::soft_body_get_angular_stiffness(RID p_body) {
 	SoftBodyBullet *body = soft_body_owner.getornull(p_body);
 	ERR_FAIL_COND_V(!body, 0.f);
-	return body->get_areaAngular_stiffness();
+	return body->get_angular_stiffness();
 }
 
 void BulletPhysicsServer3D::soft_body_set_volume_stiffness(RID p_body, real_t p_stiffness) {

modules/bullet/bullet_physics_server.h

@@ -298,8 +298,8 @@ class BulletPhysicsServer3D : public PhysicsServer3D {
 	virtual void soft_body_set_linear_stiffness(RID p_body, real_t p_stiffness) override;
 	virtual real_t soft_body_get_linear_stiffness(RID p_body) override;
 
-	virtual void soft_body_set_areaAngular_stiffness(RID p_body, real_t p_stiffness) override;
-	virtual real_t soft_body_get_areaAngular_stiffness(RID p_body) override;
+	virtual void soft_body_set_angular_stiffness(RID p_body, real_t p_stiffness) override;
+	virtual real_t soft_body_get_angular_stiffness(RID p_body) override;
 
 	virtual void soft_body_set_volume_stiffness(RID p_body, real_t p_stiffness) override;
 	virtual real_t soft_body_get_volume_stiffness(RID p_body) override;

modules/bullet/config.py

@@ -1,5 +1,6 @@
 def can_build(env, platform):
-    return True
+    # API Changed and bullet is disabled at the moment
+    return False
 
 
 def configure(env):

modules/bullet/soft_body_bullet.cpp

@@ -259,10 +259,10 @@ void SoftBodyBullet::set_linear_stiffness(real_t p_val) {
 	}
 }
 
-void SoftBodyBullet::set_areaAngular_stiffness(real_t p_val) {
-	areaAngular_stiffness = p_val;
+void SoftBodyBullet::set_angular_stiffness(real_t p_val) {
+	angular_stiffness = p_val;
 	if (bt_soft_body) {
-		mat0->m_kAST = areaAngular_stiffness;
+		mat0->m_kAST = angular_stiffness;
 	}
 }
 
@@ -409,7 +409,7 @@ void SoftBodyBullet::setup_soft_body() {
 	bt_soft_body->generateBendingConstraints(2, mat0);
 
 	mat0->m_kLST = linear_stiffness;
-	mat0->m_kAST = areaAngular_stiffness;
+	mat0->m_kAST = angular_stiffness;
 	mat0->m_kVST = volume_stiffness;
 
 	// Clusters allow to have Soft vs Soft collision but doesn't work well right now

modules/bullet/soft_body_bullet.h

@@ -67,7 +67,7 @@ class SoftBodyBullet : public CollisionObjectBullet {
 	int simulation_precision = 5;
 	real_t total_mass = 1.;
 	real_t linear_stiffness = 0.5; // [0,1]
-	real_t areaAngular_stiffness = 0.5; // [0,1]
+	real_t angular_stiffness = 0.5; // [0,1]
 	real_t volume_stiffness = 0.5; // [0,1]
 	real_t pressure_coefficient = 0.; // [-inf,+inf]
 	real_t pose_matching_coefficient = 0.; // [0,1]
@@ -129,8 +129,8 @@ class SoftBodyBullet : public CollisionObjectBullet {
 	void set_linear_stiffness(real_t p_val);
 	_FORCE_INLINE_ real_t get_linear_stiffness() const { return linear_stiffness; }
 
-	void set_areaAngular_stiffness(real_t p_val);
-	_FORCE_INLINE_ real_t get_areaAngular_stiffness() const { return areaAngular_stiffness; }
+	void set_angular_stiffness(real_t p_val);
+	_FORCE_INLINE_ real_t get_angular_stiffness() const { return angular_stiffness; }
 
 	void set_volume_stiffness(real_t p_val);
 	_FORCE_INLINE_ real_t get_volume_stiffness() const { return volume_stiffness; }

modules/csg/csg_shape.cpp

@@ -45,7 +45,8 @@ void CSGShape3D::set_use_collision(bool p_enable) {
 
 	if (use_collision) {
 		root_collision_shape.instance();
-		root_collision_instance = PhysicsServer3D::get_singleton()->body_create(PhysicsServer3D::BODY_MODE_STATIC);
+		root_collision_instance = PhysicsServer3D::get_singleton()->body_create();
+		PhysicsServer3D::get_singleton()->body_set_mode(root_collision_instance, PhysicsServer3D::BODY_MODE_STATIC);
 		PhysicsServer3D::get_singleton()->body_set_state(root_collision_instance, PhysicsServer3D::BODY_STATE_TRANSFORM, get_global_transform());
 		PhysicsServer3D::get_singleton()->body_add_shape(root_collision_instance, root_collision_shape->get_rid());
 		PhysicsServer3D::get_singleton()->body_set_space(root_collision_instance, get_world_3d()->get_space());
@@ -494,7 +495,8 @@ void CSGShape3D::_notification(int p_what) {
 
 		if (use_collision && is_root_shape()) {
 			root_collision_shape.instance();
-			root_collision_instance = PhysicsServer3D::get_singleton()->body_create(PhysicsServer3D::BODY_MODE_STATIC);
+			root_collision_instance = PhysicsServer3D::get_singleton()->body_create();
+			PhysicsServer3D::get_singleton()->body_set_mode(root_collision_instance, PhysicsServer3D::BODY_MODE_STATIC);
 			PhysicsServer3D::get_singleton()->body_set_state(root_collision_instance, PhysicsServer3D::BODY_STATE_TRANSFORM, get_global_transform());
 			PhysicsServer3D::get_singleton()->body_add_shape(root_collision_instance, root_collision_shape->get_rid());
 			PhysicsServer3D::get_singleton()->body_set_space(root_collision_instance, get_world_3d()->get_space());

modules/gridmap/grid_map.cpp

@@ -286,7 +286,8 @@ void GridMap::set_cell_item(const Vector3i &p_position, int p_item, int p_rot) {
 		//create octant because it does not exist
 		Octant *g = memnew(Octant);
 		g->dirty = true;
-		g->static_body = PhysicsServer3D::get_singleton()->body_create(PhysicsServer3D::BODY_MODE_STATIC);
+		g->static_body = PhysicsServer3D::get_singleton()->body_create();
+		PhysicsServer3D::get_singleton()->body_set_mode(g->static_body, PhysicsServer3D::BODY_MODE_STATIC);
 		PhysicsServer3D::get_singleton()->body_attach_object_instance_id(g->static_body, get_instance_id());
 		PhysicsServer3D::get_singleton()->body_set_collision_layer(g->static_body, collision_layer);
 		PhysicsServer3D::get_singleton()->body_set_collision_mask(g->static_body, collision_mask);
@@ -491,7 +492,7 @@ bool GridMap::_octant_update(const OctantKey &p_key) {
 			Octant::MultimeshInstance mmi;
 
 			RID mm = RS::get_singleton()->multimesh_create();
-			RS::get_singleton()->multimesh_allocate(mm, E->get().size(), RS::MULTIMESH_TRANSFORM_3D);
+			RS::get_singleton()->multimesh_allocate_data(mm, E->get().size(), RS::MULTIMESH_TRANSFORM_3D);
 			RS::get_singleton()->multimesh_set_mesh(mm, mesh_library->get_item_mesh(E->key())->get_rid());
 
 			int idx = 0;

platform/uwp/os_uwp.cpp

@@ -44,7 +44,6 @@
 #include "platform/windows/windows_terminal_logger.h"
 #include "servers/audio_server.h"
 #include "servers/rendering/rendering_server_default.h"
-#include "servers/rendering/rendering_server_wrap_mt.h"
 
 #include <ppltasks.h>
 #include <wrl.h>

platform/windows/os_windows.cpp

@@ -45,7 +45,6 @@
 #include "platform/windows/display_server_windows.h"
 #include "servers/audio_server.h"
 #include "servers/rendering/rendering_server_default.h"
-#include "servers/rendering/rendering_server_wrap_mt.h"
 #include "windows_terminal_logger.h"
 
 #include <avrt.h>

scene/2d/cpu_particles_2d.cpp

@@ -60,7 +60,7 @@ void CPUParticles2D::set_amount(int p_amount) {
 	}
 
 	particle_data.resize((8 + 4 + 4) * p_amount);
-	RS::get_singleton()->multimesh_allocate(multimesh, p_amount, RS::MULTIMESH_TRANSFORM_2D, true, true);
+	RS::get_singleton()->multimesh_allocate_data(multimesh, p_amount, RS::MULTIMESH_TRANSFORM_2D, true, true);
 
 	particle_order.resize(p_amount);
 }