fix(autoware_tensorrt_plugins): avoid tv::zeros and tv::empty

[veqcc]

Description

This PR removes tv::zeros and tv::empty.
This is because they call cudaMalloc inside tensorview, which leads to engine build error on NVIDIA DRIVE AGX Thor.

• in implicit_gemm_plugin.cpp , the tv::zeros call is just moved to the constructor
• in get_indices_pairs_implicit_gemm_plugin.cpp , it is a little bit complex
  • Calc additional workspace size for tv::zeros and tv::empty
  • Call tv::from_blob instead of them at runtime
  • Pre-allocate for thrust workspace

Related links

Parent Issue:

• Link

How was this PR tested?

I have checked compile/engine-build/execution worked well on both x86-64 machine and DRIVE Thor.
Through TIER IV internal evaluator, there is no regression on inference precision.

Notes for reviewers

None.

Interface changes

None.

Effects on system behavior

None.

[github-actions]

Thank you for contributing to the Autoware project!

🚧 If your pull request is in progress, switch it to draft mode.

Please ensure:

• You've checked our contribution guidelines.
• Your PR follows our pull request guidelines.
• All required CI checks pass before marking the PR ready for review.

[codecov]

Codecov Report

❌ Patch coverage is 0% with 34 lines in your changes missing coverage. Please review.
✅ Project coverage is 0.00%. Comparing base (f983578) to head (a08a58d).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
...ins/src/get_indices_pairs_implicit_gemm_plugin.cpp	0.00%	31 Missing ⚠️
...ware_tensorrt_plugins/src/implicit_gemm_plugin.cpp	0.00%	3 Missing ⚠️

❗ There is a different number of reports uploaded between BASE (f983578) and HEAD (a08a58d). Click for more details.

HEAD has 1 upload less than BASE

Flag	BASE (f983578)	HEAD (a08a58d)
daily	1	0

Additional details and impacted files

@@             Coverage Diff             @@
##             main   #12378       +/-   ##
===========================================
- Coverage   18.73%    0.00%   -18.74%     
===========================================
  Files        1904       96     -1808     
  Lines      129969     3497   -126472     
  Branches    43951        0    -43951     
===========================================
- Hits        24355        0    -24355     
+ Misses      85621     3497    -82124     
+ Partials    19993        0    -19993     

Flag	Coverage Δ
daily	?
full-suite	0.00% <0.00%> (-18.74%)	⬇️

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[veqcc]

@amadeuszsz
Thank you for your reviews!! I have addressed all your comments 👍

[amadeuszsz]

LGTM!

[amadeuszsz]

@veqcc
Before merging, could you please make sure that BEVFusion works after latest changes? Checking x86 (desktop) is enough.

[veqcc]

@amadeuszsz

Before merging, could you please make sure that BEVFusion works after latest changes? Checking x86 (desktop) is enough.

I have tested the following on x86-64 machine

• source build success
  colcon build --symlink-install --cmake-args -DCMAKE_BUILD_TYPE=Release --packages-up-to autoware_bevfusion
• engine build success
  ros2 launch autoware_bevfusion bevfusion.launch.xml build_only:=true
• launch success
  ros2 launch autoware_bevfusion bevfusion.launch.xml + ros2 bag play some_rosbag and /perception/detection/objects is published

[veqcc]

@amadeuszsz
I have also checked there is no regression by this PR through TIER IV internal evaluator!

[KSeangTan]

LGTM overall, please address comments accordingly