The GitHub Actions job "mainline-only" on tvm-ffi.git/main has failed.
Run started by GitHub user tqchen (triggered by tqchen).

Head commit for run:
68222aaa02129b4e090a28aabbece6a2e4724d38 / Junru Shao <[email protected]>
feat(dtype): expand parsing and introspection (#668)

Summary:
- Add compact integer, floating-point, and low-precision dtype aliases
in the native parser.
- Add C++ dtype classification/abbreviation helpers and Python dtype
predicates/constants.
- Incorporate review feedback by caching Python classification sets,
removing dead returns, and force-inlining simple native predicates.

Architecture:
- Normalize compact dtype spellings before applying the canonical native
parsing rules.
- Keep reusable C++ classification helpers alongside dtype conversion
utilities.
- Express integer classification as direct code comparisons and
force-inline primitive predicates for downstream optimization.
- Store Python integer and floating-point classification codes once at
class scope.

Public Interfaces:
- Add `DTypeIsFloat`, `DTypeIsInt`, `DTypeIsBool`, and `DTypeAbbrev` to
the C++ API.
- Add Python dtype predicates, validity checking, and named float8,
float6, and float4 constants.

UI/UX:
- None.

Behavioral Changes:
- Accept compact `i`, `u`, `f`, `bf`, `fp8`, `fp6`, and `fp4` spellings,
including vector-lane suffixes.
- Reject partial low-precision floating-point variant matches
consistently.
- Avoid allocating type-code sets on every `is_integer` or `is_float`
property access.
- Expose simple force-inlined integer and Boolean checks to downstream
optimizers without changing results.

Docs:
- Add API docstrings and Doxygen comments for the new dtype helpers.
- No additional documentation is required for the review-only
performance cleanups.

Tests:
- `uv run --no-sync pytest -q tests/python/test_dtype.py` — 77 passed.
- Rebuilt `tvm_ffi_tests` and ran `--gtest_filter=*DType*:*DataType*` —
9 passed.
- Ruff, Ruff formatting, clang-format, ty, and all other applicable
pre-commit hooks passed.

Untested Edge Cases:
- Alias parsing for externally registered custom dtype codes.
- GCC and MSVC were not run locally; the C++ changes were compiled with
AppleClang.
- The MSVC-specific `TVM_FFI_INLINE` expansion remains covered by
cross-platform CI.

Commits:
- `a8834f73` feat(dtype): expand parsing and introspection
- `8cef0767` perf(dtype): avoid repeated classification allocations
- `d7a30b4c` perf(dtype): inline primitive type predicates

Report URL: https://github.com/apache/tvm-ffi/actions/runs/29455117083

With regards,
GitHub Actions via GitBox


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to