[FEA] ANSI SQL Operator JIT Support (3) : Refactor operator library

[lamarrr]

Description

Follows up #22514.
Implements refactoring and bug fixes for the operator library:

• renames the bit_* operators to bitwise_* for consistency with AST operator names
• renames cast_to_b8 to cast_to_bool8 and other cast_to_* functions for consistency with existing AST operator names
• renames ansi_* arithmetic to use _overflow and ansi_precision_check to precision_check for consistency with CUDF's operators
• updates binary operators to use double template types (to make it usable with AST's double dispatch mechanism)
• removes invalid .overflow checks in the ANSI operators
• fixes error returns in ANSI operators (return std::unexpected)
• removes cuda::std::convertible_to concept checks, which are not available in CUDA 12.2
• properly constrains if_else operator to avoid ambiguous argument deduction
• moves the error code enum to a public header
• adds to_string function for the error code enum --- to be used in exceptions
• refactors the transform kernel
• removes nullify_if as it is not necessary for a first PoC for ANSI support
• Fixes call to detail::decimal_cast in cast_to_decimal_* functions.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

📝 Walkthrough

Walkthrough

Refactors operator helpers (bitwise, casts, ANSI arithmetic, logic, null handling), migrates internal error enum to cudf::errc, updates AST invoker mappings to renamed helpers, and inlines transform kernel operation dispatch.

Changes

Operator Library and Transform Kernel Refactoring

Layer / File(s)	Summary
Error handling infrastructure migration cpp/include/cudf/errc.hpp, cpp/include/cudf/detail/operators/ansi_arithmetic.cuh, cpp/include/cudf/detail/operators/operators.cuh	Introduces public cudf::errc enum with SUCCESS, OVERFLOW, DIVISION_BY_ZERO codes and to_string() helper; removes internal cudf/detail/operators/error.hpp; updates operator headers to include the public error type.
Bitwise operator renaming cpp/include/cudf/detail/operators/bitwise.cuh, cpp/include/cudf/ast/detail/operator_functor.cuh	Renames bit_and, bit_invert, bit_or, bit_xor, bit_shift_left, bit_shift_right to bitwise_and, bitwise_invert, bitwise_or, bitwise_xor, bitwise_shift_left, bitwise_shift_right and updates AST invoker mappings to call the new names.
Cast operator renaming and constraint updates cpp/include/cudf/detail/operators/casts.cuh, cpp/include/cudf/ast/detail/operator_functor.cuh	Renames cast helpers to explicit type names (cast_to_bool8, cast_to_int8-64, cast_to_uint8-64, cast_to_float32-64, cast_to_decimal32-128); replaces convertible_to constraints with static_cast validity requires; generalizes rescale template parameter; updates decimal cast implementations and AST cast mappings.
ANSI arithmetic operator template refactoring cpp/include/cudf/detail/operators/ansi_arithmetic.cuh	Converts ansi_add, ansi_sub, ansi_mul, ansi_div, ansi_mod from single-type to constrained two-parameter templates (A, B with requires(cuda::std::same_as<A, B>)); updates overflow-detection calls to new helper style; changes decimal modulus error propagation to return unexpected{error()} on missing value; generalizes ansi_precision_check to accept any integral precision type.
Utility operator generalization cpp/include/cudf/detail/operators/logic.cuh, cpp/include/cudf/detail/operators/null_handling.cuh	Updates if_else to accept any cuda::std::integral predicate type instead of bool; refactors coalesce to use two-parameter template form with type-equality constraints; removes nullify_if helper.
AST operator functor dispatch updates cpp/include/cudf/ast/detail/operator_functor.cuh	Updates AST operator invoker mappings to dispatch bitwise operations to renamed bitwise_* helpers and cast operations to renamed cast_to_int64/uint64/float64 helpers.
Transform kernel operation dispatch inlining cpp/src/transform/jit/kernel.cu	Removes execute_transform_op device helper and inlines operation dispatch logic; builds a lambda that conditionally applies GENERIC_TRANSFORM_OP with or without user data using cuda::std::apply; renames loop variable from element_idx to row.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related issues

• [FEA][STORY] Support ANSI Operators using JIT #22598 — Related to ANSI arithmetic and operator helper renames addressed by this PR.

Possibly related PRs

• rapidsai/cudf#22514: Updates the same ANSI operator-library building blocks and helper APIs adjusted in this PR.

Suggested labels

3 - Ready for Review

Suggested reviewers

• bdice
• devavret
• mhaseeb123
• wence-
• vyasr

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 50.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately reflects the main objective of the PR, which is refactoring the operator library to support ANSI SQL operator JIT. It captures the core change without excessive detail.
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.
Description check	✅ Passed	The pull request description clearly relates to the changeset and covers all major modifications including operator renames, template updates, error handling fixes, and kernel refactoring.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[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 `@cpp/src/transform/jit/kernel.cu`:
- Around line 49-50: The loop body in the operation lambda calls __ballot_sync
with a hard-coded full mask while threads that meet row >= row_size may have
exited the loop, causing undefined behavior; replace the hard-coded mask usage
in the null-aware path (where __ballot_sync is invoked) with an active-thread
mask (e.g., use __activemask() or compute a mask from the per-lane predicate) so
only participating lanes are included, and ensure any subsequent warp-wide sync
uses the same mask; update references around the operation lambda and the
null-aware ballot call to use that computed mask instead of 0xFFFF'FFFFU.

🪄 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: Enterprise

Run ID: 266f14c1-7ea0-4d98-8c68-ab2ca7f21bf0

📥 Commits

Reviewing files that changed from the base of the PR and between 023a6f5 and c738c00.

📒 Files selected for processing (10)

• cpp/include/cudf/ast/detail/operator_functor.cuh
• cpp/include/cudf/detail/operators/ansi_arithmetic.cuh
• cpp/include/cudf/detail/operators/bitwise.cuh
• cpp/include/cudf/detail/operators/casts.cuh
• cpp/include/cudf/detail/operators/error.hpp
• cpp/include/cudf/detail/operators/logic.cuh
• cpp/include/cudf/detail/operators/null_handling.cuh
• cpp/include/cudf/detail/operators/operators.cuh
• cpp/include/cudf/errc.hpp
• cpp/src/transform/jit/kernel.cu

💤 Files with no reviewable changes (1)

• cpp/include/cudf/detail/operators/error.hpp

[bdice]

removes cuda::std::convertible_to concept checks, they are not available in CUDA 12.2

We require a minimum of CUDA 12.9 for builds, is 12.9 supported? If so, this shouldn't be an issue.

[lamarrr]

We require a minimum of CUDA 12.9 for builds, is 12.9 supported? If so, this shouldn't be an issue.

For JIT code, the C++ version & standard library headers are tied to the runtime CUDA version, not the build version

[bdice]

For JIT code, the C++ version & standard library headers are tied to the runtime CUDA version, not the build version

Ouch. Was this something that passed CI previously? Do we have adequate CI test coverage of this area?

[lamarrr]

Ouch. Was this something that passed CI previously? Do we have adequate CI test coverage of this area?

The existing operators are used and tested by AST; the new operators are used and tested in #22602.
Some C++ 20 features, like nested concepts on variadic arguments (requires (requires(T...){})) , silently fail on CUDA 12.2

[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 `@cpp/include/cudf/detail/operators/casts.cuh`:
- Around line 234-237: The rescale template currently accepts any signed_integer
Scale but unconditionally narrows new_scale to int32_t when constructing
numeric::scale_type, risking silent truncation; change the template to accept an
int32_t scale (e.g., template <typename R, signed_integer int32_t Scale> or
simply require Scale be int32_t) or, if you must support wider types, add an
explicit range check in rescale (check new_scale fits within int32_t bounds and
handle out-of-range cases by asserting/throwing/logging) before the static_cast
to int32_t when constructing numeric::scale_type for
numeric::decimal<R>::rescaled.

🪄 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: Enterprise

Run ID: df584d1a-c591-4628-9e87-fff92a4ff64d

📥 Commits

Reviewing files that changed from the base of the PR and between ebc30dc and 214205e.

📒 Files selected for processing (1)

• cpp/include/cudf/detail/operators/casts.cuh

[igorpeshansky]

One remaining question/optional comment, otherwise LGTM .

[PointKernel]

Looks good. Docs are outdated but not blocking

[lamarrr]

I believe all concerns have been addressed @PointKernel @igorpeshansky @mhaseeb123

[igorpeshansky]

LGTM

[lamarrr]

/merge