Code optimization discussion about smacc2 HOT 9 CLOSED

Valgrind output from sm_atomic_performance_test

[INFO] [launch]: Default logging verbosity is set to INFO
[INFO] [sm_atomic_performance_test_node-1]: process started with pid [179876]
[sm_atomic_performance_test_node-1] ==179876== Memcheck, a memory error detector
[sm_atomic_performance_test_node-1] ==179876== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
[sm_atomic_performance_test_node-1] ==179876== Using Valgrind-3.15.0 and LibVEX; rerun with -h for copyright info
[sm_atomic_performance_test_node-1] ==179876== Command: /home/deniss/workspace/robosoft-ai/smacc_rolling/install/sm_atomic_performance_test/lib/sm_atomic_performance_test/sm_atomic_performance_test_node --ros-args --log-level FATAL --ros-args -r __node:=sm_atomic_performance_test --params-file /home/deniss/workspace/robosoft-ai/smacc_rolling/install/sm_atomic_performance_test/share/sm_atomic_performance_test/config/sm_atomic_config.yaml
[sm_atomic_performance_test_node-1] ==179876== 
[sm_atomic_performance_test_node-1] ==179876== Conditional jump or move depends on uninitialised value(s)
[sm_atomic_performance_test_node-1] ==179876==    at 0x4B1B4D8: smacc2::SignalDetector::pollOnce() (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/smacc2/libsmacc2.so)
[sm_atomic_performance_test_node-1] ==179876==    by 0x4B1D8BC: smacc2::SignalDetector::pollingLoop() (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/smacc2/libsmacc2.so)
[sm_atomic_performance_test_node-1] ==179876==    by 0x15F4E6: void smacc2::run<sm_atomic_performance_test::SmAtomicPerformanceTest>() (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/sm_atomic_performance_test/sm_atomic_performance_test_node)
[sm_atomic_performance_test_node-1] ==179876==    by 0x155AFD: main (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/sm_atomic_performance_test/sm_atomic_performance_test_node)
[sm_atomic_performance_test_node-1] ==179876== 
[sm_atomic_performance_test_node-1] ==179876== Conditional jump or move depends on uninitialised value(s)
[sm_atomic_performance_test_node-1] ==179876==    at 0x4B1B4F0: smacc2::SignalDetector::pollOnce() (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/smacc2/libsmacc2.so)
[sm_atomic_performance_test_node-1] ==179876==    by 0x4B1D8BC: smacc2::SignalDetector::pollingLoop() (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/smacc2/libsmacc2.so)
[sm_atomic_performance_test_node-1] ==179876==    by 0x15F4E6: void smacc2::run<sm_atomic_performance_test::SmAtomicPerformanceTest>() (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/sm_atomic_performance_test/sm_atomic_performance_test_node)
[sm_atomic_performance_test_node-1] ==179876==    by 0x155AFD: main (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/sm_atomic_performance_test/sm_atomic_performance_test_node)
[sm_atomic_performance_test_node-1] ==179876==

from smacc2.

We may also have a look to the cpu performance using callgrind:
https://www.valgrind.org/docs/manual/cl-manual.html

And also using the graphical tool to analyze its output files: kcachegrind

With that tool we can see what are the functions that are taking longer.
Please, lets do tests in release and also using the log-level=FATAL

from smacc2.

Valgrind output from sm_atomic_performance_test

[INFO] [launch]: Default logging verbosity is set to INFO
[INFO] [sm_atomic_performance_test_node-1]: process started with pid [179876]
[sm_atomic_performance_test_node-1] ==179876== Memcheck, a memory error detector
[sm_atomic_performance_test_node-1] ==179876== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
[sm_atomic_performance_test_node-1] ==179876== Using Valgrind-3.15.0 and LibVEX; rerun with -h for copyright info
[sm_atomic_performance_test_node-1] ==179876== Command: /home/deniss/workspace/robosoft-ai/smacc_rolling/install/sm_atomic_performance_test/lib/sm_atomic_performance_test/sm_atomic_performance_test_node --ros-args --log-level FATAL --ros-args -r __node:=sm_atomic_performance_test --params-file /home/deniss/workspace/robosoft-ai/smacc_rolling/install/sm_atomic_performance_test/share/sm_atomic_performance_test/config/sm_atomic_config.yaml
[sm_atomic_performance_test_node-1] ==179876== 
[sm_atomic_performance_test_node-1] ==179876== Conditional jump or move depends on uninitialised value(s)
[sm_atomic_performance_test_node-1] ==179876==    at 0x4B1B4D8: smacc2::SignalDetector::pollOnce() (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/smacc2/libsmacc2.so)
[sm_atomic_performance_test_node-1] ==179876==    by 0x4B1D8BC: smacc2::SignalDetector::pollingLoop() (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/smacc2/libsmacc2.so)
[sm_atomic_performance_test_node-1] ==179876==    by 0x15F4E6: void smacc2::run<sm_atomic_performance_test::SmAtomicPerformanceTest>() (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/sm_atomic_performance_test/sm_atomic_performance_test_node)
[sm_atomic_performance_test_node-1] ==179876==    by 0x155AFD: main (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/sm_atomic_performance_test/sm_atomic_performance_test_node)
[sm_atomic_performance_test_node-1] ==179876== 
[sm_atomic_performance_test_node-1] ==179876== Conditional jump or move depends on uninitialised value(s)
[sm_atomic_performance_test_node-1] ==179876==    at 0x4B1B4F0: smacc2::SignalDetector::pollOnce() (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/smacc2/libsmacc2.so)
[sm_atomic_performance_test_node-1] ==179876==    by 0x4B1D8BC: smacc2::SignalDetector::pollingLoop() (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/smacc2/libsmacc2.so)
[sm_atomic_performance_test_node-1] ==179876==    by 0x15F4E6: void smacc2::run<sm_atomic_performance_test::SmAtomicPerformanceTest>() (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/sm_atomic_performance_test/sm_atomic_performance_test_node)
[sm_atomic_performance_test_node-1] ==179876==    by 0x155AFD: main (in /home/deniss/workspace/robosoft-ai/smacc_rolling/build/sm_atomic_performance_test/sm_atomic_performance_test_node)
[sm_atomic_performance_test_node-1] ==179876==

Concerning this, it sounds to me to some variable initialized to the default zero value. Nonetheless, it would be nice to find the way to get which line of code it refers. (I see here that the file and the line of code and the file should be provided )

from smacc2.

• The struct/class issue it would be nice to solve it, but lets follow the priorities on the pending tasks because does not look critical.

Everything should be at the current moment struct. Probably class is makes more sematic sense because we are inheriting from SmaccState, but the visibility of the methods should be updated if we moved to class.

from smacc2.

dec@dec-Wild-Dog-Pro:~/workspace/rolling_ws$ ros2 launch sm_atomic_performance_test_a_1 sm_atomic_performance_test_a_1.launch
[INFO] [launch]: All log files can be found below /home/dec/.ros/log/2021-10-02-17-23-31-140880-dec-Wild-Dog-Pro-38894
[INFO] [launch]: Default logging verbosity is set to INFO
[INFO] [sm_atomic_performance_test_a_1_node-1]: process started with pid [38899]
[sm_atomic_performance_test_a_1_node-1] 1633220611.378020 [0] sm_atomic_: using network interface enp0s31f6 (udp/192.168.1.27) selected arbitrarily from: enp0s31f6, docker0
[sm_atomic_performance_test_a_1_node-1] Waited 10001 ms
[sm_atomic_performance_test_a_1_node-1] Number of iterations 1952
[INFO] [sm_atomic_performance_test_a_1_node-1]: process has finished cleanly [pid 38899]

195.2 hz

from smacc2.

[sm_atomic_performance_test_a_1_node-1] Waited 10005.4 ms
[sm_atomic_performance_test_a_1_node-1] Number of iterations 1948
[INFO] [sm_atomic_performance_test_a_1_node-1]: process has finished cleanly [pid 40469]

close to 200hz
194.8hz

from smacc2.

The key outputs would be...

• A short description of the analysis setup (Callgrind + Kcachegrind) in a readme (so that it can be repeated)
  

• The output files
  

• A cursory analysis of the functions that are taking the most time.
  

from smacc2.

https://baptiste-wicht.com/posts/2011/09/profile-c-application-with-callgrind-kcachegrind.html

from smacc2.

Ok, first pattern found. When testing sm_atomic_performance_test_a_1. If the RTA is on, which then acts as a subscriber, the rate dramatically decreases.
On Declans Machine, with the RTA, 37 interations. Without the RTA, 2946. Corresponding to 294.6Hz.

So bottom line, always make sure the SMACC2_RTA is not running.

from smacc2.