【Hackathon 10th Spring No.45】FastDeploy 支持在 T4/V100 硬件的编译 -part

[playaswd]

Motivation

Support FastDeploy compilation on NVIDIA T4/V100 (SM75/SM70).

Modifications

• Update custom_ops/setup_ops.py to gate architecture-specific CUDA sources/macros and deduplicate source entries.
• Update custom_ops/gpu_ops/cpp_extensions.cc to register/export SM-specific ops with compile-time guards.

Usage or Command

# Build
python -m pip install -U pybind11 pybind11-global
MAX_JOBS=8 python setup.py bdist_wheel

# Install
python -m pip install -U dist/fastdeploy_gpu-*.whl

# Verify
python -c "import fastdeploy; print(fastdeploy.__file__)"
python -m pip show fastdeploy-gpu

Accuracy Tests

• This PR does not change model forward numerical logic.
• It changes build/source selection and import-time compatibility guards only.
• Build and install were verified on SM86/SM75/SM70 environments; the HPU CI error path (paddle.compat missing) is addressed by compatibility fallback.

Checklist

• PR description sections are complete and non-empty.
• Formatting checks (pre-commit) passed for modified files.
• A new commit has been pushed and CI can be re-run on the latest head.

[paddle-bot]

Thanks for your contribution!

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (develop@0b4c1cb). Learn more about missing BASE report.

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #6488   +/-   ##
==========================================
  Coverage           ?   73.67%           
==========================================
  Files              ?      399           
  Lines              ?    55827           
  Branches           ?     8802           
==========================================
  Hits               ?    41132           
  Misses             ?    11781           
  Partials           ?     2914           

Flag	Coverage Δ
GPU	73.67% <ø> (?)

Flags with carried forward coverage won't be shown. 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.

[mitu626]

lgtm

[luotao1]

请解决下冲突

[cloudforge1]

Note: the current HEAD has diverged from the state that received lgtm (Mar 9).

Post-review additions include get_compile_parallelism() (+26 lines) which is unrelated to Task 45 scope (compile guards for T4/V100). The reviewed diff was a clean compile-guard PR; the current diff is a compile-guard PR + build parallelism refactor.

May be worth re-reviewing against the current state, or reverting the out-of-scope additions to match the reviewed version.

[luotao1]

@mattheliu 已经找 @qingqing01 @Jiang-Jia-Jun 问过，该PR可以先合入，部分完成该赛题。

[luotao1]

@playaswd 感谢各位荣获第十期黑客松奖项！为确保奖金顺利发放，请获奖选手于7月10日（本周五）18:00前完成问卷填写。信息仅用于本次打款，请确保准确无误。

问卷链接：https://paddle.wjx.cn/vm/YeLn0Ov.aspx#

填写后请务必加入微信群，后续通知将统一在群内同步