feat(autoware_cuda_pointcloud_preprocessor): cuda 12.0 build compatibility

[amadeuszsz]

Description

Latest PR related to NVCC flags aligned compatibility with minimum required version for Autoware. However, downstream projects may use custom CI/CD pipeline or local builds, where CUDA version is below 12.8. This PR adjust flags and make build possible for pipelines with CUDA 12.0+.

Resulting flags for:

• CUDA 12.0 - 12.7: sm_86, sm_87, sm_89, compute_89
• CUDA 12.8 - 12.9: sm_86, sm_87, sm_89, sm_101, sm_120, compute_120
• CUDA 13.0+: sm_86, sm_87, sm_89, sm_110, sm_120, compute_120

Related links

Parent Issue:

autowarefoundation/autoware#6789

How was this PR tested?

Notes for reviewers

Even though this modification keeps flags compatible with user's CUDA version, we can't assure compatibility with CUDA < 12.8 for any incoming features since Autoware is developed under CUDA 12.8 environment.

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

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 18.11%. Comparing base (8d07c62) to head (1f4ac70).

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #12194      +/-   ##
==========================================
- Coverage   18.14%   18.11%   -0.03%     
==========================================
  Files        1841     1843       +2     
  Lines      126693   126897     +204     
  Branches    44501    44671     +170     
==========================================
  Hits        22983    22983              
- Misses      84544    84748     +204     
  Partials    19166    19166              

Flag	Coverage Δ		*Carryforward flag
daily	20.63% <ø> (ø)		Carriedforward from 898718b
daily-cuda	18.14% <ø> (ø)		Carriedforward from 898718b
differential-cuda	0.00% <ø> (?)
total-cuda	18.13% <ø> (ø)		Carriedforward from 898718b

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

☔ 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.

[mojomex]

Possible duplicate of TIER IV branch: tier4#2707

[mojomex]

Change itself LGTM. Is the plan to have 1 PR per package? Otherwise the tier4#2707 PR could be cherry-picked instead.

[amadeuszsz]

Possible duplicate of TIER IV branch: tier4#2707

All sub-projects may need similar change, therefore change in upstream seems more appropriate.
The change is split by multiple PR for easier cherry-picking.

Also linked PR does not come with forward compatibility if build with CUDA < 12.8, so for consistency purposes I would prefer to apply changes from this PR.

memo: I experimented with more robust approaches, but unfortunately not all the environments can discover architecture properly and therefore fallback to sm_50. Current approach is kind of "legacy", but safest.