Add internal-use __addressOf instrinsic to replace &

[aidanfnv]

The prefix & operator for taking addresses is being deprecated, but internal Slang code and tests need an equivalent mechanism.

This change adds __addressOf, an internal-use intrinsic declared in core.meta.slang with the same __intrinsic_op(0) behavior as &. It has a new KnownBuiltinDeclName::InternalAddressOf attribute, and the existing immutable-object check in CheckInvokeExprWithCheckedOperands is extended to also match InternalAddressOf, so that error 30079 still fires when passing a read-only value, like &. This change also migrates all internal and test usages of & for address-of to __addressOf.

[coderabbitai]

📝 Walkthrough

Walkthrough

This pull request introduces a new generic intrinsic operation __addressOf that provides an alternative address-taking mechanism alongside the existing address-of operator. The change includes the intrinsic definition, corresponding enum member, compiler checking logic updates, and systematic replacement of address-of usage across test files and a standard library module.

Changes

Cohort / File(s)	Summary
Core Language Infrastructure source/slang/core.meta.slang, source/slang/slang-ast-support-types.h	Adds new __addressOf intrinsic operation declaration with KnownBuiltin(InternalAddressOf) attribute and require(cpp_cuda_spirv_llvm) constraint. Introduces corresponding InternalAddressOf enum member to KnownBuiltinDeclName.
Compiler Type Checking source/slang/slang-check-expr.cpp	Extends constant-pointer diagnostic logic to handle both OperatorAddressOf and InternalAddressOf builtin names as unable to accept constant pointers.
Standard Library source/standard-modules/neural/bindless-storage.slang	Updates BindlessAddress<T>.atomicAdd for vector<T, 2> on CUDA to use __addressOf(handle[baseIndex + index]) instead of address-of operator.
Test Files tests/bugs/gh-3601.slang, tests/bugs/gh-7499.slang, tests/bugs/gh-8659.slang	Replaces address-of operator with __addressOf(...) intrinsic in various pointer-related test scenarios.
CPU Program Tests tests/cpu-program/gfx-smoke.slang, tests/cpu-program/pointer-basics.slang, tests/cpu-program/pointer-deref.slang	Systematically replaces address-of operator usage with __addressOf(...) intrinsic across device creation, pointer initialization, and descriptor passing patterns.
Diagnostic & Feature Tests tests/diagnostics/invalid-constant-pointer-taking.slang, tests/language-feature/bitfield/*.slang, tests/language-feature/pointer/coherent-load-store-physical-storage-buffer.slang, tests/llvm/inline-llvm-ir.slang, tests/spirv/pointer.slang	Replaces address-of operator with __addressOf(...) intrinsic for struct addressing, bitfield representation, coherent loads, inline LLVM IR, and SPIR-V pointer operations.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

Poem

A rabbit hops through pointers bright,
With __addressOf now in sight,
No more just & to take our way,
New intrinsics guide the day! 🐰✨
Address-taking, clean and right.

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 33.33% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely summarizes the main change: introducing a new internal __addressOf intrinsic to replace the deprecated & operator.
Description check	✅ Passed	The description is directly related to the changeset, explaining the motivation for the new intrinsic and detailing the implementation approach.

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

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

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

[Copilot]

Pull request overview

This PR adds a new __addressOf intrinsic to replace the deprecated & (address-of) operator for internal Slang code and tests. The change maintains identical behavior to the & operator, including the immutable-object safety check (error 30079), while providing a clear internal-use alternative as the public & operator is being deprecated.

Changes:

• Added __addressOf intrinsic declaration in core.meta.slang with matching semantics to & operator
• Extended immutable-object validation to also check InternalAddressOf in addition to OperatorAddressOf
• Migrated all internal code and test usages from & to __addressOf across 13 test files and 1 standard module file

Reviewed changes

Copilot reviewed 17 out of 17 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
source/slang/core.meta.slang	Declares new __addressOf intrinsic with __intrinsic_op(0) and KnownBuiltinDeclName::InternalAddressOf attribute
source/slang/slang-ast-support-types.h	Adds InternalAddressOf to KnownBuiltinDeclName enum
source/slang/slang-check-expr.cpp	Extends immutable-object check to validate InternalAddressOf alongside OperatorAddressOf
tests/spirv/pointer.slang	Migrates &p.data to __addressOf(p.data)
tests/llvm/inline-llvm-ir.slang	Migrates &a to __addressOf(a) in memset call
tests/language-feature/pointer/coherent-load-store-physical-storage-buffer.slang	Migrates &secondPtrIn[1] to __addressOf(secondPtrIn[1])
tests/language-feature/bitfield/repr.slang	Migrates &s to __addressOf(s) in cast expression
tests/language-feature/bitfield/repr-mixed.slang	Migrates &m to __addressOf(m) in cast expression
tests/language-feature/bitfield/msvc-repr.slang	Migrates &s to __addressOf(s) in cast expression
tests/diagnostics/invalid-constant-pointer-taking.slang	Migrates test to verify error 30079 still fires with __addressOf on constant buffers
tests/cpu-program/pointer-deref.slang	Migrates &rec to __addressOf(rec)
tests/cpu-program/pointer-basics.slang	Migrates two uses of &value to __addressOf(value)
tests/cpu-program/gfx-smoke.slang	Migrates 9 address-of operations to __addressOf in API calls
tests/bugs/gh-8659.slang	Migrates &ptr[0] to __addressOf(ptr[0])
tests/bugs/gh-7499.slang	Migrates &arr to __addressOf(arr) in function call
tests/bugs/gh-3601.slang	Migrates &p.data to __addressOf(p.data)
source/standard-modules/neural/bindless-storage.slang	Migrates &handle[baseIndex + index] to __addressOf(...) in CUDA atomic operation

[coderabbitai]

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

tests/diagnostics/invalid-constant-pointer-taking.slang (1)

9-17: ⚠️ Potential issue | 🟡 Minor

Clarify whether __addressOf on an RWStructuredBuffer element is intentionally unchecked.

The file-level comment (Line 9) reads: "We do not allow taking a pointer from a StructuredBuffer/RWStructuredBuffer."

After the migration:

• __getAddress(mutable_float_buffer[...]) (Line 17) → error 31160 ✔
• __addressOf(mutable_float_buffer[...]) (Line 14) → no error CHECK

Two interpretations:

1. Intentional escape hatch — __addressOf deliberately bypasses the __getAddress guard for mutable buffers, in which case the comment on Line 9 is now stale and should be narrowed to apply only to __getAddress.
2. Accidental omission — the error 31160 CHECK for the mutable case was dropped during migration, introducing a diagnostic regression.

Please either add the missing // CHECK: ([[# @line+1]]): error ... directive on Line 13, or update the comment on Line 9 to document that __addressOf is permitted on RWStructuredBuffer elements.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@tests/diagnostics/invalid-constant-pointer-taking.slang` around lines 9 - 17,
The test currently asserts an error for __getAddress but not for __addressOf on
elements of mutable_float_buffer; decide whether this is a regression or
intended behavior and update the test accordingly: if __addressOf should also be
disallowed, add a CHECK directive before the line that declares mutablePtr1
asserting the same diagnostic (error 31160) for
__addressOf(mutable_float_buffer[threadId.x]); otherwise, update the file-level
comment that currently says "We do not allow taking a pointer from a
StructuredBuffer/RWStructuredBuffer." to clarify that the restriction applies
only to __getAddress and that __addressOf is permitted, referencing the symbols
__addressOf, __getAddress, and mutable_float_buffer in the updated comment so
the intent is clear.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Outside diff comments:
In `@tests/diagnostics/invalid-constant-pointer-taking.slang`:
- Around line 9-17: The test currently asserts an error for __getAddress but not
for __addressOf on elements of mutable_float_buffer; decide whether this is a
regression or intended behavior and update the test accordingly: if __addressOf
should also be disallowed, add a CHECK directive before the line that declares
mutablePtr1 asserting the same diagnostic (error 31160) for
__addressOf(mutable_float_buffer[threadId.x]); otherwise, update the file-level
comment that currently says "We do not allow taking a pointer from a
StructuredBuffer/RWStructuredBuffer." to clarify that the restriction applies
only to __getAddress and that __addressOf is permitted, referencing the symbols
__addressOf, __getAddress, and mutable_float_buffer in the updated comment so
the intent is clear.

---

ℹ️ Review info

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Knowledge base: Disabled due to Reviews -> Disable Knowledge Base setting

📥 Commits

Reviewing files that changed from the base of the PR and between ed97591 and b654f2e.

📒 Files selected for processing (17)

• source/slang/core.meta.slang
• source/slang/slang-ast-support-types.h
• source/slang/slang-check-expr.cpp
• source/standard-modules/neural/bindless-storage.slang
• tests/bugs/gh-3601.slang
• tests/bugs/gh-7499.slang
• tests/bugs/gh-8659.slang
• tests/cpu-program/gfx-smoke.slang
• tests/cpu-program/pointer-basics.slang
• tests/cpu-program/pointer-deref.slang
• tests/diagnostics/invalid-constant-pointer-taking.slang
• tests/language-feature/bitfield/msvc-repr.slang
• tests/language-feature/bitfield/repr-mixed.slang
• tests/language-feature/bitfield/repr.slang
• tests/language-feature/pointer/coherent-load-store-physical-storage-buffer.slang
• tests/llvm/inline-llvm-ir.slang
• tests/spirv/pointer.slang

[aidanfnv]

Discussed with Yong on Discord that this should not be added and we should instead change __getAddress to not emit an error for CPU or CUDA, so that __getAddress can be used for all of the tests changed here.