Avoid allocating instance buffer for PointLight2D shadows when there are no instances.

Also avoid the render loop if there are no occluders. We still have to open the render pass to do a clear though.

servers/rendering/renderer_rd/renderer_canvas_render_rd.cpp

@@ -1056,7 +1056,7 @@ void RendererCanvasRenderRD::light_update_shadow(RID p_rid, int p_shadow_index,
 	// Then, upload all the occluder transforms to a shared buffer.
 	// We only do this for the first light so we can avoid uploading the same
 	// Transforms over and over again.
-	if (p_shadow_index == 0) {
+	if (p_shadow_index == 0 && occluder_count > 0) {
 		static thread_local LocalVector<float> transforms;
 		transforms.clear();
 		transforms.resize(occluder_count * 8);
@@ -1076,41 +1076,43 @@ void RendererCanvasRenderRD::light_update_shadow(RID p_rid, int p_shadow_index,
 	Rect2i rect(0, p_shadow_index * 2, state.shadow_texture_size, 2);
 	RD::DrawListID draw_list = RD::get_singleton()->draw_list_begin(state.shadow_fb, RD::DRAW_CLEAR_ALL, cc, 1.0f, 0, rect);
 
-	RD::get_singleton()->draw_list_bind_render_pipeline(draw_list, shadow_render.render_pipelines[SHADOW_RENDER_MODE_POSITIONAL_SHADOW]);
-	RD::get_singleton()->draw_list_bind_uniform_set(draw_list, state.shadow_ocluder_uniform_set, 0);
-
-	for (int i = 0; i < 4; i++) {
-		Rect2i sub_rect((state.shadow_texture_size / 4) * i, p_shadow_index * 2, (state.shadow_texture_size / 4), 2);
-		RD::get_singleton()->draw_list_set_viewport(draw_list, sub_rect);
-
-		static const Vector2 directions[4] = { Vector2(1, 0), Vector2(0, 1), Vector2(-1, 0), Vector2(0, -1) };
-		static const Vector4 rotations[4] = { Vector4(0, -1, 1, 0), Vector4(-1, 0, 0, -1), Vector4(0, 1, -1, 0), Vector4(1, 0, 0, 1) };
-
-		PositionalShadowRenderPushConstant push_constant;
-		_update_transform_2d_to_mat2x4(p_light_xform, push_constant.modelview);
-		push_constant.direction[0] = directions[i].x;
-		push_constant.direction[1] = directions[i].y;
-		push_constant.rotation[0] = rotations[i].x;
-		push_constant.rotation[1] = rotations[i].y;
-		push_constant.rotation[2] = rotations[i].z;
-		push_constant.rotation[3] = rotations[i].w;
-		push_constant.z_far = p_far;
-		push_constant.z_near = p_near;
-
-		for (uint32_t j = 0; j < occluders.size(); j++) {
-			OccluderPolygon *co = occluders[j];
-
-			push_constant.pad = occluder_indices[j];
-			push_constant.cull_mode = uint32_t(co->cull_mode);
-
-			// The slowest part about this whole function is that we have to draw the occluders one by one, 4 times.
-			// We can optimize this so that all occluders draw at once if we store vertices and indices in a giant
-			// SSBO and just save an index into that SSBO for each occluder.
-			RD::get_singleton()->draw_list_bind_vertex_array(draw_list, co->vertex_array);
-			RD::get_singleton()->draw_list_bind_index_array(draw_list, co->index_array);
-			RD::get_singleton()->draw_list_set_push_constant(draw_list, &push_constant, sizeof(PositionalShadowRenderPushConstant));
-
-			RD::get_singleton()->draw_list_draw(draw_list, true);
+	if (state.shadow_occluder_buffer.is_valid()) {
+		RD::get_singleton()->draw_list_bind_render_pipeline(draw_list, shadow_render.render_pipelines[SHADOW_RENDER_MODE_POSITIONAL_SHADOW]);
+		RD::get_singleton()->draw_list_bind_uniform_set(draw_list, state.shadow_ocluder_uniform_set, 0);
+
+		for (int i = 0; i < 4; i++) {
+			Rect2i sub_rect((state.shadow_texture_size / 4) * i, p_shadow_index * 2, (state.shadow_texture_size / 4), 2);
+			RD::get_singleton()->draw_list_set_viewport(draw_list, sub_rect);
+
+			static const Vector2 directions[4] = { Vector2(1, 0), Vector2(0, 1), Vector2(-1, 0), Vector2(0, -1) };
+			static const Vector4 rotations[4] = { Vector4(0, -1, 1, 0), Vector4(-1, 0, 0, -1), Vector4(0, 1, -1, 0), Vector4(1, 0, 0, 1) };
+
+			PositionalShadowRenderPushConstant push_constant;
+			_update_transform_2d_to_mat2x4(p_light_xform, push_constant.modelview);
+			push_constant.direction[0] = directions[i].x;
+			push_constant.direction[1] = directions[i].y;
+			push_constant.rotation[0] = rotations[i].x;
+			push_constant.rotation[1] = rotations[i].y;
+			push_constant.rotation[2] = rotations[i].z;
+			push_constant.rotation[3] = rotations[i].w;
+			push_constant.z_far = p_far;
+			push_constant.z_near = p_near;
+
+			for (uint32_t j = 0; j < occluders.size(); j++) {
+				OccluderPolygon *co = occluders[j];
+
+				push_constant.pad = occluder_indices[j];
+				push_constant.cull_mode = uint32_t(co->cull_mode);
+
+				// The slowest part about this whole function is that we have to draw the occluders one by one, 4 times.
+				// We can optimize this so that all occluders draw at once if we store vertices and indices in a giant
+				// SSBO and just save an index into that SSBO for each occluder.
+				RD::get_singleton()->draw_list_bind_vertex_array(draw_list, co->vertex_array);
+				RD::get_singleton()->draw_list_bind_index_array(draw_list, co->index_array);
+				RD::get_singleton()->draw_list_set_push_constant(draw_list, &push_constant, sizeof(PositionalShadowRenderPushConstant));
+
+				RD::get_singleton()->draw_list_draw(draw_list, true);
+			}
 		}
 	}
 	RD::get_singleton()->draw_list_end();