[BugFix] Reject unsupported fast-math input dtypes

[erhsh]

Summary

• reject integer, boolean, and float8 inputs for the eight explicit fast-math intrinsics
• report an actionable frontend TypeError instead of emitting identity operations or undefined CUDA symbols
• add GPU-independent coverage across every affected intrinsic and dtype family

Root cause

The T.__exp, T.__log, and related wrappers forwarded every input dtype directly to CUDA code generation. CUDA fast-math lowering only has valid paths for standard floating-point type codes: small integers and float8 fell through to an empty function name and became silent identity operations, while 32/64-bit integers emitted undefined CUDA symbols.

The wrappers now use one shared validator matching TVM's floating-point type-code convention. Standard float and bfloat inputs retain their existing frontend behavior; unsupported types fail during intrinsic construction.

Fixes #2597.

Validation

• .venv/bin/python -m pytest testing/python/language/test_tilelang_language_fastmath.py -q (88 passed)
• .venv/bin/python -m pytest testing/python/math/test_math_fast_math.py testing/python/language/test_tilelang_language_dialect.py -q (18 passed, 34 skipped without a local GPU)
• python3 -m pre_commit run --files tilelang/language/math_intrinsics.py testing/python/language/test_tilelang_language_fastmath.py
• git diff --check

Summary

• Added shared frontend dtype validation for all eight fast-math intrinsics (e.g., __exp, __log, __tan, __cos, __sin, __exp10, __log2, __log10) using TVM floating-point type-code conventions.
• Rejects integer, boolean, and float8 input dtypes with actionable frontend TypeErrors (while preserving existing behavior for float/bfloat inputs).
• Updated intrinsic implementations to use the shared validator instead of directly converting inputs.
• Added parametrized GPU-independent tests covering both unsupported dtype rejection (error includes intrinsic name and dtype) and supported float/bfloat acceptance (frontend returns tvm.tirx.Call with matching dtype).

C++ style / lint notes

• No C++/CI/lint-related files or documentation were changed; docs/developer_guide/cpp_style.md and the “C++ API Style Audit (warning only)” CI step are not impacted by this PR.
• All changes are correctness-focused in the Python frontend and corresponding frontend tests (no warning-only style findings introduced or touched).

[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: Pro Plus

Run ID: 065e7ba8-846b-4fa7-8bdf-6ee49ef44cf1

📥 Commits

Reviewing files that changed from the base of the PR and between 092e4cf and ebc1f78.

📒 Files selected for processing (1)

• testing/python/language/test_tilelang_language_fastmath.py

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

• testing/python/language/test_tilelang_language_fastmath.py

---

📝 Walkthrough

Walkthrough

Fastmath intrinsics now validate frontend argument dtypes, accepting float and bfloat inputs while rejecting unsupported types. Parametrized tests cover rejection messages, returned call types, and dtype preservation.

Changes

Fastmath dtype validation

Layer / File(s)	Summary
Frontend fastmath validation tilelang/language/math_intrinsics.py	A shared validator restricts eight fastmath intrinsics to float and bfloat dtypes with intrinsic-specific TypeError messages.
Fastmath dtype tests testing/python/language/test_tilelang_language_fastmath.py	Parametrized tests verify unsupported dtype rejection and floating-point tirx.Call dtype preservation.

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

Possibly related issues

• tile-ai/tilelang issue 2565 — Covers overlapping fastmath intrinsic dtype validation and floating-point coverage concerns.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly matches the main change: rejecting unsupported fast-math input dtypes.
Linked Issues check	✅ Passed	The validator and tests cover all eight fast-math intrinsics and reject integer, bool, and float8 dtypes as required by #2597.
Out of Scope Changes check	✅ Passed	The added acceptance tests and shared helper are directly related to the fast-math dtype-guard fix and stay within scope.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[github-actions]

👋 Hi! Thank you for contributing to the TileLang project.

Please remember to run pre-commit run --all-files in the root directory of the project to ensure your changes are properly linted and formatted. This will help ensure your contribution passes the format check.

We appreciate you taking this step! Our team will review your contribution, and we look forward to your awesome work! 🚀

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@testing/python/language/test_tilelang_language_fastmath.py`:
- Around line 19-36: Update test_fastmath_rejects_unsupported_dtype to include
int64 and uint64 in the parameterized dtype matrix, and strengthen the
pytest.raises match so it asserts the error includes the offending dtype via the
expected “but got {dtype}” text.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: aa5e6786-0008-4677-9c75-dcf6c8440296

📥 Commits

Reviewing files that changed from the base of the PR and between e41fadb and 092e4cf.

📒 Files selected for processing (2)

• testing/python/language/test_tilelang_language_fastmath.py
• tilelang/language/math_intrinsics.py

[LeiWang1999]

LGTM