[Chris B (llvm-beanz)]

This PR implements the hlsl namespace and moves all built-in functions into the hlsl namespace.

This feature was adopted by TC57 in proposal 0009: https://hlsl-tc57.github.io/tc57/proposal/0009/

This PR has significantly grown in complexity during development. Initially it just put declarations into the namespace and did some minor fixes to name lookup. As the test cases developed that proved insufficient because of how the names of functions are lazily added to the AST (which is critical for DXC's compile-time performance).

The full change has a more invasive connection into name lookup, and leverages another part of the ExternalSemaSource.

The first change is marking our namespaces with the ExternallyVisibleStorage, which is a flag on the DeclContext to mark that it has "some external storage" (think serialized AST file, or our code that injects AST nodes) that contains declarations which should be visible in the context. This is similar to the ExternalLexicalStorage flag which we already use, but it extends from just lexical declarations to visible declarations.

When name lookup inspects a declaration context there is a callback, FindExternalVisibleDeclsByName, which gets called if the DeclContext has externally stored visible declarations. This provides a hook to insert visible names during name lookup. This complements our existing name lookup hook in LookupUnqualified by also supporting qualified name lookup, and triggering earlier in the lookup process.

In FindExternalVisibleDeclsByName, if the specified name is an intrinsic we insert a function declaration that serves as a marker for the builtin intrinsic. This function declaration takes no arguments, returns void, and has an attribute applied to mark it as an intrinsic marker. This gives us a declaration to slide into normal name lookup, but also avoids us needing to add all the overloads.

When name lookup returns such a declaration we then know that we have an intrinsic function and we can perform the same overload resolution against it that we would have previously.

The one outstanding bit of work is to move the HLSL resource objects and raytracing objects into the hlsl namespace.

Partial implementation of #8918

[Copilot]

Copilot review overview

🟡 Changes recommended

The AST test’s negative check does not currently verify the absence of an implicit using-directive throughout the output.

Review effort: Balanced
Findings: 1

Open (1)

• CHECK-NOT does not cover the entire AST dump · New

What changed in this PR

Implements TC57 proposal 0009 by moving HLSL 202x built-in functions into the implicit hlsl namespace while preserving earlier language behavior.

Changes:

• Adds version-aware namespace lookup and intrinsic resolution.
• Adds semantic and DXIL coverage for qualification and using-directives.
• Updates affected tests and release notes.

File	Description
tools/​clang/​lib/​Sema/​SemaHLSL.cpp	Creates and resolves the hlsl namespace.
tools/​clang/​lib/​Sema/​SemaExpr.cpp	Enables qualified intrinsic lookup.
tools/​clang/​lib/​Sema/​SemaOverload.cpp	Supports overload resolution in hlsl.
docs/​ReleaseNotes.md	Documents the language change.
tools/​clang/​test/​SemaHLSL/​v202x/​conforming-literals/​asfloat16.hlsl	Qualifies asfloat16.
tools/​clang/​test/​SemaHLSL/​hlsl-using-namespace-tu-scope-202x.hlsl	Tests translation-unit using-directives.
tools/​clang/​test/​SemaHLSL/​hlsl-using-namespace-namespace-scope-202x.hlsl	Tests namespace-scoped using-directives.
tools/​clang/​test/​SemaHLSL/​hlsl-using-namespace-function-scope-202x.hlsl	Tests block-scoped using-directives.
tools/​clang/​test/​SemaHLSL/​hlsl-namespace-qualified-intrinsics.hlsl	Tests qualified and unqualified lookup.
tools/​clang/​test/​SemaHLSL/​hlsl-namespace-pre-202x.hlsl	Verifies legacy AST behavior.
tools/​clang/​test/​SemaHLSL/​hlsl-namespace-not-available-pre-202x.hlsl	Verifies legacy namespace behavior.
tools/​clang/​test/​SemaHLSL/​hlsl-namespace-no-implicit-using-202x.hlsl	Tests absence of implicit imports.
tools/​clang/​test/​SemaHLSL/​hlsl-namespace-202x.hlsl	Checks the HLSL 202x AST.
tools/​clang/​test/​HLSLFileCheckLit/​hlsl/​auto/​auto-variable-deduction.hlsl	Adapts an existing 202x test.
tools/​clang/​test/​HLSLFileCheck/​hlsl/​control_flow/​attributes/​unroll/​full_hint_202x.hlsl	Qualifies dot.
tools/​clang/​test/​HLSLFileCheck/​hlsl/​control_flow/​attributes/​unroll/​count_runtime_hint_202x.hlsl	Qualifies runtime-loop dot calls.
tools/​clang/​test/​HLSLFileCheck/​hlsl/​control_flow/​attributes/​unroll/​count_hint_202x.hlsl	Qualifies counted-loop dot.
tools/​clang/​test/​CodeGenDXIL/​hlsl/​intrinsics/​hlsl-using-namespace-call.hlsl	Tests unqualified DXIL lowering via using.
tools/​clang/​test/​CodeGenDXIL/​hlsl/​intrinsics/​hlsl-namespace-qualified-call.hlsl	Tests qualified DXIL lowering.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🔵 Needs a closer look

Intrinsic lookup and constant-folding regressions remain unresolved.

No open issues 🎉

3 previously missed issues

In code that hasn't changed since last review

Preserve namespace visibility for dependent HLSL intrinsic calls

tools/​clang/​lib/​Sema/​SemaHLSL.cpp:5576

Dependent unqualified intrinsic calls lose access to hlsl during template instantiation. For example, template<typename T> T f(T x) { using namespace hlsl; return sin(x); } cannot resolve sin when first instantiated with float: TreeTransform::RebuildCallExpr passes a null Scope, leaving SearchHlslViaUsing false. A translation-unit using-directive has the same problem. Preserve definition-time namespace visibility for dependent calls, including block-scoped directives, and add regression tests for both scopes.

Restore constant folding for HLSL namespace intrinsics

tools/​clang/​lib/​Sema/​SemaHLSL.cpp:5682

Making m_hlslNSDecl non-null drops the existing constant-folding path for core intrinsics. AddHLSLIntrinsicFunction attaches HLSLBuiltinCallAttr only when NS is null, while CGExpr.cpp:3649 requires that attribute to reach TryEvalIntrinsic. Thus constant calls such as hlsl::sin(1.0f) bypass codegen-time folding. After creating an intrinsic declaration, attach the attribute when m_hlslNSDecl && T.NS == m_hlslNSDecl, preserving existing dx/vk handling, and add a constant-argument codegen regression.

Support HLSL intrinsics in using-declarations

tools/​clang/​lib/​Sema/​SemaHLSL.cpp:5689

In a fresh HLSL 202x translation unit, using hlsl::sin; fails with “no member named 'sin'”. Intrinsic declarations are synthesized by AddOverloadedCallCandidates, but BuildUsingDeclaration performs ordinary qualified lookup; marking external lexical storage does not expose those names. Support intrinsic-name lookup for using-declarations while preserving lazy overload generation, and add a regression that imports and calls an intrinsic before any qualified call.

1 resolved since last review

• CHECK-NOT does not cover the entire AST dump

_{Review effort: Balanced}

[Copilot]

🟡 Changes recommended

Unresolved lookup defects affect template instantiation, selective imports, and qualified-call namespace isolation.

3 open issues

• Suppress ADL for qualified HLSL calls · New
• Preserve HLSL namespace visibility for deferred template calls · New
• Enable lazy intrinsic lookup for using-declarations · New

_{Review effort: Balanced}

[Copilot]

🟡 Changes recommended

Intrinsic families found only through argument-dependent lookup bypass synthesis, rejecting valid unqualified calls.

1 open issue

• Synthesize intrinsic families discovered through ADL · New

3 resolved since last review

• Enable lazy intrinsic lookup for using-declarations
• Preserve HLSL namespace visibility for deferred template calls
• Suppress ADL for qualified HLSL calls

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

The interaction between lazy intrinsic synthesis, namespace lookup, and overload resolution needs final human validation before approval.

1 open issue

• Synthesize intrinsic families discovered through ADL

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

The cross-cutting changes to lazy name lookup and overload resolution warrant final human review supported by compiler regression-test results.

No open issues 🎉

1 resolved since last review

• Synthesize intrinsic families discovered through ADL

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

The cross-cutting changes to lazy name lookup and overload resolution warrant compiler-maintainer review and regression-test validation.

No open issues 🎉

_{Review effort: Balanced}

[Copilot]

🟡 Changes recommended

Inconsistent argument-dependent lookup checks can abort assertion-enabled compilers on valid calls through nested dx or vk namespaces.

1 open issue

• Restrict vk/dx ADL to root namespaces · New

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

The changes span lazy name lookup, template instantiation, and overload resolution, requiring compiler-domain human review supported by regression results.

No open issues 🎉

1 resolved since last review

• Restrict vk/dx ADL to root namespaces

_{Review effort: Balanced}

[Ashley Coleman (V-FEXrt)]

has some tricky stuff but overall reasonable for the required functionality. Tests do a great job covering all the cases so thanks for those:)

[Helena Kotas (hekota)]

LGTM - pending Sarah's feedback in SemaHLSL.cpp.

[Copilot]

🔵 Needs a closer look

Human assessment is needed because an intrinsic-overload edge case leaves no candidates for an argument-count mismatch, producing a misleading missing-member diagnostic.

No open issues 🎉

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

Human review is needed to assess the impact of duplicated argument-dependent lookup work on ordinary HLSL 202x user-function calls.

No open issues 🎉

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

Interacting changes to core lookup, ADL, and template overload resolution require compiler-maintainer review and cross-version DXIL/SPIR-V regression validation beyond this source-only review.

No open issues 🎉

_{Review effort: Balanced}