feat(trtexec_vendor): add vendor package for trtexec

[wep21]

Description

a vendor package for trtexec

Related links

Tests performed

Please check if you can build tensorrt engine from onnx file with trtexec

colcon build --cmake-args -DCMAKE_BUILD_TYPE=Release  --symlink-install --continue-on-error --packages-up-to trtexec_vendor lidar_centerpoint
source install/setup.bash
trtexec --onnx=$(ros2 pkg prefix lidar_centerpoint --share)/data/pts_backbone_neck_head_centerpoint.onnx --shapes=spatial_features:1x32x560x560 --saveEngine=$(ros2 pkg prefix lidar_centerpoint --share)/data/pts_backbone_neck_head_centerpoint.engine --buildOnly

Notes for reviewers

Pre-review checklist for the PR author

The PR author must check the checkboxes below when creating the PR.

• I've confirmed the contribution guidelines.
• The PR follows the pull request guidelines.

In-review checklist for the PR reviewers

The PR reviewers must check the checkboxes below before approval.

• The PR follows the pull request guidelines.
• The PR has been properly tested.
• The PR has been reviewed by the code owners.

Post-review checklist for the PR author

The PR author must check the checkboxes below before merging.

• There are no open discussions or they are tracked via tickets.
• The PR is ready for merge.

After all checkboxes are checked, anyone who has write access can merge the PR.

[kenji-miyake]

If you plan to release this package, I think it's better to manage this package in a separate repository and add it to build_depends.repos.
@wep21 Could you tell me your thought about it?
cc @xmfcx @mitsudome-r @yukkysaito

[wep21]

@kenji-miyake I won't make this package released because ros build farm cannot handle cuda libraries.
I will create a separate repository if it is preferable.

[codecov]

Codecov Report

Base: 10.45% // Head: 10.44% // Decreases project coverage by -0.01% ⚠️

Coverage data is based on head (51d1eab) compared to base (03b4741).
Patch has no changes to coverable lines.

❗ Current head 51d1eab differs from pull request most recent head a2e8969. Consider uploading reports for the commit a2e8969 to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1821      +/-   ##
==========================================
- Coverage   10.45%   10.44%   -0.02%     
==========================================
  Files        1231     1221      -10     
  Lines       87879    87744     -135     
  Branches    20501    20498       -3     
==========================================
- Hits         9191     9164      -27     
+ Misses      69175    69106      -69     
+ Partials     9513     9474      -39     

Flag	Coverage Δ		*Carryforward flag
differential	∅ <ø> (?)
total	10.42% <ø> (-0.02%)	⬇️	Carriedforward from fb01e98

*This pull request uses carry forward flags. Click here to find out more.

Impacted Files	Coverage Δ
...n/motion_utils/test/src/resample/test_resample.cpp	25.66% <0.00%> (-0.28%)	⬇️
...ils/include/motion_utils/trajectory/trajectory.hpp	78.81% <0.00%> (-0.11%)	⬇️
perception/tensorrt_yolo/src/nodelet.cpp	0.00% <0.00%> (ø)
perception/lidar_centerpoint/src/node.cpp	0.00% <0.00%> (ø)
control/shift_decider/src/shift_decider.cpp	0.00% <0.00%> (ø)
planning/route_handler/src/route_handler.cpp	0.00% <0.00%> (ø)
planning/rtc_interface/src/rtc_interface.cpp	0.00% <0.00%> (ø)
perception/tensorrt_yolo/lib/src/trt_yolo.cpp	0.00% <0.00%> (ø)
localization/ekf_localizer/src/ekf_localizer.cpp	0.00% <0.00%> (ø)
planning/obstacle_avoidance_planner/src/node.cpp	0.00% <0.00%> (ø)
... and 119 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[kenji-miyake]

@wep21 I see, thanks!
Considering that CUDA is only used in autoware.universe, I feel it's reasonable to manage this package in this repository. 👍

If it's okay for @yukke42, I think you can merge this.

[yukke42]

I have checked that almost the same performance is reproduced using engine files built with the following trtexec command.

trtexec \
    --onnx=$(ros2 pkg prefix lidar_centerpoint --share)/data/pts_voxel_encoder_centerpoint.onnx \
    --saveEngine=$(ros2 pkg prefix lidar_centerpoint --share)/data/pts_voxel_encoder_centerpoint.engine \
    --shapes=input_features:40000x32x9 \
    --optShapes=input_features:40000x32x9 \
    --workspace=1024 \
    --fp16 \
    --buildOnly

trtexec \
    --onnx=$(ros2 pkg prefix lidar_centerpoint --share)/data/pts_backbone_neck_head_centerpoint.onnx \
    --saveEngine=$(ros2 pkg prefix lidar_centerpoint --share)/data/pts_backbone_neck_head_centerpoint.engine \
    --shapes=spatial_features:1x32x560x560 \
    --optShapes=spatial_features:1x32x560x560 \
    --workspace=1024 \
    --fp16 \
    --buildOnly

But, it takes more than 3 times than the lidar_centerpoint package implementation to build two TensorRT engine files.
lidar_centerpoint package: about 1 min
trtexec: about 3 min 30 sec

[yukke42]

I have checked that trtexec can convert onnx files and lidar_centerpoint can load engine files.