[SM6.10] LinAlg Validation: Convert

[Ashley Coleman (V-FEXrt)]

Fixes #8509

Implements LinAlg Convert validation rules

---

_{Stack created with GitHub Stacks CLI • Give Feedback 💬}

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[Copilot]

Pull request overview

Implements Shader Model 6.10 validation rules for LinAlg Convert operations.

Changes:

• Validates interpretation constants, vector element types, and output dimensions.
• Adds validation diagnostics and DXIL documentation.
• Adds and updates validation/code-generation tests.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
lib/DxilValidation/DxilValidation.cpp	Implements Convert validation.
lib/DxilValidation/DxilValidationUtils.cpp	Identifies native component types.
lib/DxilValidation/DxilValidationUtils.h	Declares the native-type helper.
utils/hct/hctdb.py	Defines new validation messages.
docs/DXIL.rst	Documents the new rules.
tools/clang/test/LitDXILValidation/LinAlgMatrix/linalgmatrix-convert.ll	Tests validation failures.
tools/clang/test/CodeGenDXIL/hlsl/linalg/builtins/convert/nominal.hlsl	Updates nominal inputs to valid interpretations.

---

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

[Damyan Pepper (damyanp)]

LGTM, although some of the copilot comments look like they need addressing.

[Copilot]

🔵 Needs a closer look

The vector-length diagnostic incorrectly references a matrix, and the user-visible validation change needs preview release-note coverage.

Review details

Suppressed comments (2)

lib/DxilValidation/DxilValidation.cpp:1620

• This reuses the matrix-dimension diagnostic for LinAlgConvert, so an invalid conversion reports that its result must match an “input matrix M dimension” even though this operation has no matrix (as the new test demonstrates). Please add a Convert-specific vector-length rule, e.g. one that says the output vector size must match the expected converted size, and update the generated documentation/test expectation accordingly.

        CI, ValidationRule::InstrLinAlgMatrixDimVectorMismatch,
        {"Return", std::to_string(RetVecTy->getNumElements()),
         std::to_string(ExpectedOutVecSize)});

lib/DxilValidation/DxilValidation.cpp:1549

• This adds user-visible validation behavior for experimental Shader Model 6.10, but the PR does not add an entry under docs/ReleaseNotes.md → Upcoming Preview Release. Please add a release note for the new LinAlg Convert validation, or point to the related stacked PR that will provide the shared release-note coverage.

  ValidateLinAlgOpParameters(CI, ValCtx);

• Files reviewed: 7/7 changed files
• Comments generated: 0 new
• Review effort level: Balanced

[Copilot]

🟡 Changes recommended

The diagnostics contain typos, and the user-visible SM6.10 validation behavior needs release-note coverage.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Review details

Suppressed comments (1)

lib/DxilValidation/DxilValidation.cpp:1551

• This adds user-visible validation behavior for experimental Shader Model 6.10, but the PR does not add an entry under docs/ReleaseNotes.md → “Upcoming Preview Release.” Please add an entry for the Convert validation rules, or point to the shared release-note coverage in the stacked effort.

  ValidateLinAlgOpParameters(CI, ValCtx);

• Files reviewed: 7/7 changed files
• Comments generated: 4
• Review effort level: Balanced

[Copilot]

🔵 Needs a closer look

Release-note coverage is missing, and several newly added messages contain typos.

Review details

Suppressed comments (5)

utils/hct/hctdb.py:8756

• The diagnostic is grammatically malformed: “must must size” should be “must match size.” Please update the generated documentation and test expectation as well.

            "Return vector size '%0' must must size '%1' derived from input vector size and type.",

docs/DXIL.rst:3234

• Keep this generated diagnostic text consistent with the corrected validation rule: “must must size” should be “must match size.”

INSTR.LINALGMATRIXVECELEMCOUNTMISMATCH                Return vector size '%0' must must size '%1' derived from input vector size and type.

tools/clang/test/LitDXILValidation/LinAlgMatrix/linalgmatrix-convert.ll:50

• The expected diagnostic repeats the malformed “must must size” wording. Update it to match the corrected validation message.

  ; CHECK-NEXT: Function: main: error: Return vector size '32' must must size '2' derived from input vector size and type.

lib/DxilValidation/DxilValidation.cpp:1581

• Correct the typo “minimze” to “minimize.”

  // ComponentTypes. Stop early to minimze noise/avoid being unhelpful

lib/DxilValidation/DxilValidation.cpp:1552

• This introduces user-visible validation behavior for experimental Shader Model 6.10, so please add an entry under ### Upcoming Preview Release in docs/ReleaseNotes.md, or point to the shared release-note coverage planned for this stacked effort.

  DxilInst_LinAlgConvert Op(CI);

• Files reviewed: 7/7 changed files
• Comments generated: 0 new
• Review effort level: Balanced

[Joshua Batista (bob80905)]

+1 on the Copilot comment on the double-must typo

[Copilot]

🔵 Needs a closer look

Release-note coverage must be added or confirmed; round-up validation also lacks direct test coverage.

Review details

Suppressed comments (2)

lib/DxilValidation/DxilValidation.cpp:1551

• This introduces user-visible validation behavior for experimental SM 6.10, but docs/ReleaseNotes.md has no Convert-specific entry. Please add one under Upcoming Preview Release, or confirm that the existing #8588 entry is intended as shared coverage for this stacked validation effort.

static void ValidateLinAlgConvert(CallInst *CI, ValidationContext &ValCtx) {
  ValidateLinAlgOpParameters(CI, ValCtx);

tools/clang/test/LitDXILValidation/LinAlgMatrix/linalgmatrix-convert.ll:52

• The vector-size test only covers an exactly divisible conversion (8 F32 elements to 2 packed F8 scalars), so the newly implemented round-up path is not exercised. Add a validator case for an odd source size—such as <3 x half> F16 to <1 x i32> F8—and position the valid case where an unexpected diagnostic cannot be skipped by the first CHECK; an invalid result-size counterpart would also verify rejection.

  ; CHECK-NEXT: Function: main: error: Return vector size '32' must match size '2' derived from input vector size and type.
  ; CHECK-NEXT: note: at {{.*}} @dx.op.linAlgConvert.v32i32.v8f32
  %13 = call <32 x i32> @dx.op.linAlgConvert.v32i32.v8f32(i32 -2147483618, <8 x float> <float 9.000000e+00, float 8.000000e+00, float 7.000000e+00, float 6.000000e+00, float 5.000000e+00, float 4.000000e+00, float 3.000000e+00, float 2.000000e+00>, i32 9, i32 21)  ; LinAlgConvert(inputVector,inputInterpretation,outputInterpretation)

• Files reviewed: 7/7 changed files
• Comments generated: 0 new
• Review effort level: Balanced