Make nvCOMP linkage configurable

[gerashegalov]

Summary

• add CUDF_NVCOMP_LINKAGE=AUTO|STATIC|SHARED
• preserve the existing target-selection behavior under the default AUTO mode
• fail during configuration when an explicitly requested nvCOMP target is unavailable

Why

The cuDF Java distribution already packages the shared nvCOMP library. Downstream builds that otherwise force static dependencies currently embed a second copy of nvCOMP in libcudf.so while also shipping libnvcomp.so.

This option lets those distributions choose the packaged shared library without changing the default linkage policy for other cuDF consumers.

Related to NVIDIA/cudf-spark#15145.

Validation

• configured and built Release libcudf.so with both STATIC and SHARED on CUDA 12.9.1 (sm_75)
• verified the shared variant has DT_NEEDED: libnvcomp.so.5
• verified Java load order libnvcomp.so, libcudf.so, libcudfjni.so

Measured on otherwise identical one-architecture builds:

nvCOMP linkage	libcudf.so	ZIP level 6
static	711,163,904 B	359,193,871 B
shared	691,845,960 B	344,989,786 B

This removes 18.42 MiB uncompressed and 13.55 MiB after DEFLATE.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 7bf28175-8094-422d-925a-716487e32c4d

📥 Commits

Reviewing files that changed from the base of the PR and between 0cc568c and 98546bc.

📒 Files selected for processing (1)

• cpp/cmake/thirdparty/get_nvcomp.cmake

🚧 Files skipped from review as they are similar to previous changes (1)

• cpp/cmake/thirdparty/get_nvcomp.cmake

---

📝 Walkthrough

Summary by CodeRabbit

• New Features

  • Added a CUDF_NVCOMP_LINKAGE configuration option to control nvCOMP linkage mode (AUTO, STATIC, or SHARED).
  • AUTO mode now automatically selects the best available nvCOMP library based on the build configuration and what’s present.

• Bug Fixes

  • Added clear validation and error messages when an unsupported linkage type is requested.
  • Improved compatibility with packaging scenarios where a shared nvCOMP library is already provided.

Walkthrough

cuDF adds a CMake option for selecting nvCOMP linkage and updates target resolution to support validated static, shared, and automatic modes.

Changes

nvCOMP linkage configuration

Layer / File(s)	Summary
Declare nvCOMP linkage option cpp/CMakeLists.txt	Adds the CUDF_NVCOMP_LINKAGE cache option with AUTO, STATIC, and SHARED values.
Resolve nvCOMP target cpp/cmake/thirdparty/get_nvcomp.cmake	Selects and validates the static or shared nvCOMP target according to the configured mode, including automatic fallback behavior and an updated SPDX copyright line.

Estimated code review effort: 2 (Simple) | ~10 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: making nvCOMP linkage configurable.
Description check	✅ Passed	The description matches the changes and explains the new linkage option, behavior, and motivation.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[gerashegalov]

Not needed at the moment per NVIDIA/cudf-spark-jni#4881