Add dependency detection improvements to the render graph.

- Buffers changing their usage are no longer treated as write usage unless the API requires it.
- Draw lists are not treated as being dependent on each other if their regions do not intersect despite both being write commands.
- Particles were tweaked to use different unused buffers to reduce dependencies.

5 files changed, 39 insertions, 26 deletions

diff --git a/servers/rendering/renderer_rd/effects/copy_effects.cpp b/servers/rendering/renderer_rd/effects/copy_effects.cpp
index c7a7532d76..808c82f712 100644
--- a/servers/rendering/renderer_rd/effects/copy_effects.cpp
+++ b/servers/rendering/renderer_rd/effects/copy_effects.cpp
@@ -1002,15 +1002,19 @@ void CopyEffects::copy_cubemap_to_dp(RID p_source_rd_texture, RID p_dst_framebuf
 	MaterialStorage *material_storage = MaterialStorage::get_singleton();
 	ERR_FAIL_NULL(material_storage);
 
+	Rect2i screen_rect;
+	float atlas_width = p_dst_size.width / p_rect.size.width;
+	float atlas_height = p_dst_size.height / p_rect.size.height;
+	screen_rect.position.x = (int32_t)(Math::round(p_rect.position.x * atlas_width));
+	screen_rect.position.y = (int32_t)(Math::round(p_rect.position.y * atlas_height));
+	screen_rect.size.width = (int32_t)(Math::round(p_dst_size.width));
+	screen_rect.size.height = (int32_t)(Math::round(p_dst_size.height));
+
 	CopyToDPPushConstant push_constant;
-	push_constant.screen_rect[0] = p_rect.position.x;
-	push_constant.screen_rect[1] = p_rect.position.y;
-	push_constant.screen_rect[2] = p_rect.size.width;
-	push_constant.screen_rect[3] = p_rect.size.height;
 	push_constant.z_far = p_z_far;
 	push_constant.z_near = p_z_near;
-	push_constant.texel_size[0] = 1.0f / p_dst_size.x;
-	push_constant.texel_size[1] = 1.0f / p_dst_size.y;
+	push_constant.texel_size[0] = 1.0f / p_dst_size.width;
+	push_constant.texel_size[1] = 1.0f / p_dst_size.height;
 	push_constant.texel_size[0] *= p_dp_flip ? -1.0f : 1.0f; // Encode dp flip as x size sign
 
 	// setup our uniforms
@@ -1021,7 +1025,7 @@ void CopyEffects::copy_cubemap_to_dp(RID p_source_rd_texture, RID p_dst_framebuf
 	RID shader = cube_to_dp.shader.version_get_shader(cube_to_dp.shader_version, 0);
 	ERR_FAIL_COND(shader.is_null());
 
-	RD::DrawListID draw_list = RD::get_singleton()->draw_list_begin(p_dst_framebuffer, RD::INITIAL_ACTION_DISCARD, RD::FINAL_ACTION_DISCARD, RD::INITIAL_ACTION_LOAD, RD::FINAL_ACTION_STORE);
+	RD::DrawListID draw_list = RD::get_singleton()->draw_list_begin(p_dst_framebuffer, RD::INITIAL_ACTION_DISCARD, RD::FINAL_ACTION_DISCARD, RD::INITIAL_ACTION_LOAD, RD::FINAL_ACTION_STORE, Vector<Color>(), 1.0f, 0, screen_rect);
 	RD::get_singleton()->draw_list_bind_render_pipeline(draw_list, cube_to_dp.pipeline.get_render_pipeline(RD::INVALID_ID, RD::get_singleton()->framebuffer_get_format(p_dst_framebuffer)));
 	RD::get_singleton()->draw_list_bind_uniform_set(draw_list, uniform_set_cache->get_cache(shader, 0, u_source_rd_texture), 0);
 	RD::get_singleton()->draw_list_bind_index_array(draw_list, material_storage->get_quad_index_array());
diff --git a/servers/rendering/renderer_rd/effects/copy_effects.h b/servers/rendering/renderer_rd/effects/copy_effects.h
index 014f78e2b9..b6fc95d0f1 100644
--- a/servers/rendering/renderer_rd/effects/copy_effects.h
+++ b/servers/rendering/renderer_rd/effects/copy_effects.h
@@ -217,7 +217,6 @@ private:
 		float z_far;
 		float z_near;
 		float texel_size[2];
-		float screen_rect[4];
 	};
 
 	struct CopyToDP {
diff --git a/servers/rendering/renderer_rd/shaders/effects/cube_to_dp.glsl b/servers/rendering/renderer_rd/shaders/effects/cube_to_dp.glsl
index 3fb93dda35..b45e310b61 100644
--- a/servers/rendering/renderer_rd/shaders/effects/cube_to_dp.glsl
+++ b/servers/rendering/renderer_rd/shaders/effects/cube_to_dp.glsl
@@ -8,7 +8,6 @@ layout(push_constant, std430) uniform Params {
 	float z_far;
 	float z_near;
 	vec2 texel_size;
-	vec4 screen_rect;
 }
 params;
 
@@ -17,8 +16,7 @@ layout(location = 0) out vec2 uv_interp;
 void main() {
 	vec2 base_arr[4] = vec2[](vec2(0.0, 0.0), vec2(0.0, 1.0), vec2(1.0, 1.0), vec2(1.0, 0.0));
 	uv_interp = base_arr[gl_VertexIndex];
-	vec2 screen_pos = uv_interp * params.screen_rect.zw + params.screen_rect.xy;
-	gl_Position = vec4(screen_pos * 2.0 - 1.0, 0.0, 1.0);
+	gl_Position = vec4(uv_interp * 2.0 - 1.0, 0.0, 1.0);
 }
 
 #[fragment]
@@ -35,7 +33,6 @@ layout(push_constant, std430) uniform Params {
 	float z_far;
 	float z_near;
 	vec2 texel_size;
-	vec4 screen_rect;
 }
 params;
 
diff --git a/servers/rendering/renderer_rd/storage_rd/particles_storage.cpp b/servers/rendering/renderer_rd/storage_rd/particles_storage.cpp
index 314cbf9aa9..6efb8c176f 100644
--- a/servers/rendering/renderer_rd/storage_rd/particles_storage.cpp
+++ b/servers/rendering/renderer_rd/storage_rd/particles_storage.cpp
@@ -306,9 +306,14 @@ void ParticlesStorage::_particles_free_data(Particles *particles) {
 		particles->emission_storage_buffer = RID();
 	}
 
-	if (particles->unused_storage_buffer.is_valid()) {
-		RD::get_singleton()->free(particles->unused_storage_buffer);
-		particles->unused_storage_buffer = RID();
+	if (particles->unused_emission_storage_buffer.is_valid()) {
+		RD::get_singleton()->free(particles->unused_emission_storage_buffer);
+		particles->unused_emission_storage_buffer = RID();
+	}
+
+	if (particles->unused_trail_storage_buffer.is_valid()) {
+		RD::get_singleton()->free(particles->unused_trail_storage_buffer);
+		particles->unused_trail_storage_buffer = RID();
 	}
 
 	if (RD::get_singleton()->uniform_set_is_valid(particles->particles_material_uniform_set)) {
@@ -534,9 +539,15 @@ void ParticlesStorage::_particles_allocate_emission_buffer(Particles *particles)
 	}
 }
 
-void ParticlesStorage::_particles_ensure_unused_buffer(Particles *particles) {
-	if (particles->unused_storage_buffer.is_null()) {
-		particles->unused_storage_buffer = RD::get_singleton()->storage_buffer_create(sizeof(uint32_t) * 4);
+void ParticlesStorage::_particles_ensure_unused_emission_buffer(Particles *particles) {
+	if (particles->unused_emission_storage_buffer.is_null()) {
+		particles->unused_emission_storage_buffer = RD::get_singleton()->storage_buffer_create(sizeof(uint32_t) * 4);
+	}
+}
+
+void ParticlesStorage::_particles_ensure_unused_trail_buffer(Particles *particles) {
+	if (particles->unused_trail_storage_buffer.is_null()) {
+		particles->unused_trail_storage_buffer = RD::get_singleton()->storage_buffer_create(sizeof(uint32_t) * 4);
 	}
 }
 
@@ -763,8 +774,8 @@ void ParticlesStorage::_particles_process(Particles *p_particles, double p_delta
 			if (p_particles->emission_storage_buffer.is_valid()) {
 				u.append_id(p_particles->emission_storage_buffer);
 			} else {
-				_particles_ensure_unused_buffer(p_particles);
-				u.append_id(p_particles->unused_storage_buffer);
+				_particles_ensure_unused_emission_buffer(p_particles);
+				u.append_id(p_particles->unused_emission_storage_buffer);
 			}
 			uniforms.push_back(u);
 		}
@@ -779,8 +790,8 @@ void ParticlesStorage::_particles_process(Particles *p_particles, double p_delta
 				}
 				u.append_id(sub_emitter->emission_storage_buffer);
 			} else {
-				_particles_ensure_unused_buffer(p_particles);
-				u.append_id(p_particles->unused_storage_buffer);
+				_particles_ensure_unused_emission_buffer(p_particles);
+				u.append_id(p_particles->unused_emission_storage_buffer);
 			}
 			uniforms.push_back(u);
 		}
@@ -1481,8 +1492,8 @@ void ParticlesStorage::update_particles() {
 					if (particles->trail_bind_pose_buffer.is_valid()) {
 						u.append_id(particles->trail_bind_pose_buffer);
 					} else {
-						_particles_ensure_unused_buffer(particles);
-						u.append_id(particles->unused_storage_buffer);
+						_particles_ensure_unused_trail_buffer(particles);
+						u.append_id(particles->unused_trail_storage_buffer);
 					}
 					uniforms.push_back(u);
 				}
diff --git a/servers/rendering/renderer_rd/storage_rd/particles_storage.h b/servers/rendering/renderer_rd/storage_rd/particles_storage.h
index 33f44f3045..98c693f477 100644
--- a/servers/rendering/renderer_rd/storage_rd/particles_storage.h
+++ b/servers/rendering/renderer_rd/storage_rd/particles_storage.h
@@ -247,7 +247,8 @@ private:
 		ParticleEmissionBuffer *emission_buffer = nullptr;
 		RID emission_storage_buffer;
 
-		RID unused_storage_buffer;
+		RID unused_emission_storage_buffer;
+		RID unused_trail_storage_buffer;
 
 		HashSet<RID> collisions;
 
@@ -265,7 +266,8 @@ private:
 
 	void _particles_process(Particles *p_particles, double p_delta);
 	void _particles_allocate_emission_buffer(Particles *particles);
-	void _particles_ensure_unused_buffer(Particles *particles);
+	void _particles_ensure_unused_emission_buffer(Particles *particles);
+	void _particles_ensure_unused_trail_buffer(Particles *particles);
 	void _particles_free_data(Particles *particles);
 	void _particles_update_buffers(Particles *particles);