Closed ZhenshengLee closed 2 years ago
see https://github.com/ros2/demos/pull/482 https://github.com/ros2/demos/issues/583
https://github.com/ros-perception/vision_opencv/pull/441
I don't see the priority, cause it doesn't improve the performance.
But in rclcpp-intra-process, type conversion is avoided, see following
https://github.com/ros2/rclcpp/pull/1849 https://github.com/ros2/ros2_documentation/issues/2377 https://github.com/ros2/rclcpp/issues/1664
this make it possible to improve cuda performance, because cudamemcpy canbe avoided too!
see this example of intra-process-cuda-image-pipeline https://github.com/NVIDIA-ISAAC-ROS/ros2_examples/tree/humble/rclcpp/type_adaptation/accelerated_pipeline
Anyway , rclcpp-intra-process is not for ros2_shm_msgs.
zero-copy can be achieved with intra-process since dashing(stable in foxy), the type-adaption feature is mostly for saving costs in the GPU processing of image/pointclouds. this issue will be merged into #18
closing this.
https://github.com/ros-infrastructure/rep/blob/master/rep-2007.rst