ros indigo rviz performance displaying pointclouds 2x faster when compiled from_source

edit

Hi there,

as we started migrating our code to indigo yesterday we had performance issues, especially when working with pointclouds. The code compiled with catkin_make runs just with 1Hz, whereas using rosbuild we achieve 30Hz. We found out, related to the following topic:

http://answers.ros.org/question/71965...

that we need to enable compiler optimization by setting catkin_make -DCMAKE_BUILD_TYPE=Release and the code runs at the expected framerate of 30Hz.

Now to the topic:

Displaying just a pointcloud from the asus xtion in rviz installed from the repository gives us a performance of 14Hz on our desktop.

• Intel i7-3770 3.4 GHz quad-core
• Nvidia GeForce GTX660
• 16GB ram
• Ubuntu 14.04 - 64bit
• ros-indigo

Knowing the above (catkin compiled code runs 3x slower) we build rviz from source with the following compiler options:

• set(CMAKE_BUILD_TYPE RelWithDebInfo)
• set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=gnu++0x -Wall -msse2 -msse3 -mssse3 -msse4 -msse4.2")

This gives us a performance of 30Hz displaying a pointcloud, resulting in a much smoother usage of rviz.

Evaluating the same on a laptop:

• Intel i7-4600M, 2.1Ghz dual core
• Intel HD Graphics 4600
• 8 GB RAM
• Ubuntu 14.04 - 64bit
• ros indigo

rviz from repository: 4Hz vs. rviz from source 9Hz

EDIT I prepared a table containing all build tests I have done:

This leads to the question: Are the ros-indigo packages really compiled with compiler optimizations turned on e.g. as Release?

Have a nice day!