[REFACTOR] Delete src/support/libinfo.cc; replace with runtime FFI-registry env query#19477
Open
tqchen wants to merge 3 commits intoapache:mainfrom
Open
[REFACTOR] Delete src/support/libinfo.cc; replace with runtime FFI-registry env query#19477tqchen wants to merge 3 commits intoapache:mainfrom
tqchen wants to merge 3 commits intoapache:mainfrom
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Code Review
This pull request removes the libinfo infrastructure, which previously mirrored CMake configuration flags into the TVM runtime via generated C++ code. The changes include the deletion of LibInfo.cmake, libinfo.cc, and the associated linting scripts. This functionality is replaced by dynamic probing of the FFI registry and runtime modules to determine feature availability and versions. Feedback was provided regarding the performance of the new _cmake_flag_enabled utility, suggesting that its internal mapping dictionaries be moved to the module level to avoid redundant allocations during test execution.
tlopex
approved these changes
Apr 29, 2026
tqchen
force-pushed
the
tvm-phase-out-libinfo
branch
from
c4a3ed8 to
1bc95e6
Compare
…gistry env query support.GetLibInfo exposed ~30 build-time TVM_INFO_* strings — git hash, LLVM/MLIR versions, every USE_* flag. Real callers reduce to "is this feature enabled?" — better answered by runtime device / feature discovery (USE_CUDA=ON does not mean a CUDA device is loadable; runtime probe is the actual signal). Git versioning is already tracked via tvm.__version__. Replace tvm.support.libinfo() with tvm.support.detect_active_modules() which queries the FFI global function registry for ffi.Module.create.<kind> registrations on a minimal key set (cuda, vulkan, opencl). describe() now prints active runtimes instead of CMake build flags. Migrate the 5 in-tree callers: _get_targets() in testing/utils.py now uses cudnn.exists() / tvm.runtime.enabled() for CUDNN and Hexagon; _cmake_flag_enabled() is rewritten as a static mapping from cmake flag names to tvm.runtime.enabled() or FFI-registry probes; clml_sdk_version() uses the existing relax.get_openclml_version FFI global; test_clml_ops.py uses the new helper. Deletes ~323 lines C++, the LibInfo CMake module, the check_cmake_options lint, and the corresponding pre-commit hook. src/support/ becomes header-only.
…on Feature directly
The _cmake_flag_enabled translation table was naming-convention
noise — each "cmake flag X" entry just maps to "probe Y" via
tvm.runtime.enabled() or tvm.get_global_func(). Carry the runtime
check directly on each Feature via run_time_check, matching the
pattern already used by requires_x86_amx.
Removes the cmake_flag attribute from Feature, the
_cmake_flag_enabled translator, and the one out-of-tree caller in
contrib/hexagon/_ci_env_check.py. ~16 Feature instances rewritten
to use run_time_check lambdas calling tvm.runtime.enabled("<kind>")
or tvm.get_global_func() probes directly.
No build-flag introspection remains in the testing utility layer.
…r cmake_flag removal Phase-out of cmake_flag in c201020 moved per-backend availability probes from compile_time_check (which gates compile-only tests in _compile_only_marks()) to run_time_check (which is only evaluated in _run_only_marks(), i.e. compile-and-run tests). This left compile-only tests — @requires_metal(support_required="compile-only"), @requires_vulkan(support_required="compile-only") — without any gate on whether the backend is compiled in. On the CPU CI node (Linux, USE_METAL=OFF), the three metal compile-only tests no longer got skipped. They ran and hit "xcrun: not found" (Apple's Xcode tool) because Metal codegen requires macOS toolchain. Fix: add compile_time_check matching run_time_check for every Feature that previously used cmake_flag. tvm.runtime.enabled("<kind>") and tvm.get_global_func(…) are already the correct compile-time probes — they return True iff the backend was compiled into the TVM shared library, which is exactly what the old cmake_flag check provided. The compile_time_check = run_time_check pairing restores the original behaviour for all six support_required modes.
tqchen
force-pushed
the
tvm-phase-out-libinfo
branch
from
1bc95e6 to
052f46e
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
support.GetLibInfoexposed ~30 build-timeTVM_INFO_*strings (git hash, LLVM/MLIR versions, everyUSE_*flag). Real callers reduce to "is this feature enabled?" — better answered at runtime.USE_CUDA=ONdoes not mean a CUDA device is loadable; runtime discovery is the actual signal. Git versioning is already tracked viatvm.__version__.Changes:
src/support/libinfo.cc,cmake/modules/LibInfo.cmake,tests/lint/check_cmake_options.py, and thecheck-cmake-optionspre-commit hook.tvm.support.libinfo()(no shim — callers migrate to runtime discovery).tvm.support.detect_active_modules()which queries the FFI global function registry forffi.Module.create.<kind>registrations (cuda, vulkan, opencl).describe()now prints active runtimes instead of CMake build flags._get_targets()usescudnn.exists()/tvm.runtime.enabled()for CUDNN and Hexagon;_cmake_flag_enabled()is rewritten as a static map from cmake flag names totvm.runtime.enabled()or FFI-registry probes;clml_sdk_version()uses the existingrelax.get_openclml_versionFFI global;test_clml_ops.pyuses the new helper.After this PR:
src/support/is header-only.