[HLSL][Matrix] Add support for Matrix element and trunc Casts (#168915)

farzonl · web-flow · commit dd1b4abfb748 · 2025-12-02T19:02:25.000-05:00
fixes #168737 fixes #168755 This change fixes adds support for Matrix truncations via the ICK_HLSL_Matrix_Truncation enum. That ends up being most of the files changed. It also allows Matrix as an HLSL Elementwise cast as long as the cast does not perform a shape transformation ie 3x2 to 2x3. Tests for the new elementwise and truncation behavior were added. As well as sema tests to make sure we error n the shape transformation cast. I am punting right now on the ConstExpr Matrix support. That will need to be addressed later. Will file a seperate issue for that if reviewers agree it can wait.
diff --git a/clang/include/clang/AST/OperationKinds.def b/clang/include/clang/AST/OperationKinds.def
@@ -364,6 +364,9 @@ CAST_OPERATION(IntToOCLSampler)
 // Truncate a vector type by dropping elements from the end (HLSL only).
 CAST_OPERATION(HLSLVectorTruncation)
 
+// Truncate a matrix type by dropping elements from the end (HLSL only).
+CAST_OPERATION(HLSLMatrixTruncation)
+
 // Non-decaying array RValue cast (HLSL only).
 CAST_OPERATION(HLSLArrayRValue)
 
diff --git a/clang/include/clang/Basic/DiagnosticGroups.td b/clang/include/clang/Basic/DiagnosticGroups.td
@@ -1061,6 +1061,7 @@ def SuperSubClassMismatch : DiagGroup<"super-class-method-mismatch">;
 def OverridingMethodMismatch : DiagGroup<"overriding-method-mismatch">;
 def VariadicMacros : DiagGroup<"variadic-macros">;
 def VectorConversion : DiagGroup<"vector-conversion">;      // clang specific
+def MatrixConversion : DiagGroup<"matrix-conversion">;      // clang specific
 def VexingParse : DiagGroup<"vexing-parse">;
 def VLAUseStaticAssert : DiagGroup<"vla-extension-static-assert">;
 def VLACxxExtension : DiagGroup<"vla-cxx-extension", [VLAUseStaticAssert]>;
@@ -1335,6 +1336,8 @@ def : DiagGroup<"int-conversions",
                 [IntConversion]>; // -Wint-conversions = -Wint-conversion
 def : DiagGroup<"vector-conversions",
                 [VectorConversion]>; // -Wvector-conversions = -Wvector-conversion
+def : DiagGroup<"matrix-conversions",
+                [MatrixConversion]>; // -Wmatrix-conversions = -Wmatrix-conversion
 def : DiagGroup<"unused-local-typedefs", [UnusedLocalTypedef]>;
                 // -Wunused-local-typedefs = -Wunused-local-typedef
 
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -4356,6 +4356,9 @@ def warn_param_typestate_mismatch : Warning<
 def warn_unknown_sanitizer_ignored : Warning<
   "unknown sanitizer '%0' ignored">, InGroup<UnknownSanitizers>;
 
+def warn_impcast_matrix_scalar : Warning<
+  "implicit conversion turns matrix to scalar: %0 to %1">,
+  InGroup<MatrixConversion>;
 def warn_impcast_vector_scalar : Warning<
   "implicit conversion turns vector to scalar: %0 to %1">,
   InGroup<Conversion>, DefaultIgnore;
@@ -13282,7 +13285,10 @@ def err_hlsl_builtin_scalar_vector_mismatch
           "vector type with matching scalar element type%diff{: $ vs $|}2,3">;
 
 def warn_hlsl_impcast_vector_truncation : Warning<
-  "implicit conversion truncates vector: %0 to %1">, InGroup<Conversion>;
+  "implicit conversion truncates vector: %0 to %1">, InGroup<VectorConversion>;
+
+def warn_hlsl_impcast_matrix_truncation : Warning<
+  "implicit conversion truncates matrix: %0 to %1">, InGroup<MatrixConversion>;
 
 def warn_hlsl_availability : Warning<
   "%0 is only available %select{|in %4 environment }3on %1 %2 or newer">,
diff --git a/clang/include/clang/Sema/Overload.h b/clang/include/clang/Sema/Overload.h
@@ -198,6 +198,9 @@ class Sema;
     /// HLSL vector truncation.
     ICK_HLSL_Vector_Truncation,
 
+    /// HLSL Matrix truncation.
+    ICK_HLSL_Matrix_Truncation,
+
     /// HLSL non-decaying array rvalue cast.
     ICK_HLSL_Array_RValue,
 
diff --git a/clang/lib/AST/Expr.cpp b/clang/lib/AST/Expr.cpp
@@ -1934,6 +1934,7 @@ bool CastExpr::CastConsistency() const {
   case CK_FixedPointToBoolean:
   case CK_HLSLArrayRValue:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
   CheckNoBasePath:
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
@@ -11770,6 +11770,10 @@ bool VectorExprEvaluator::VisitCastExpr(const CastExpr *E) {
       Elements.push_back(Val.getVectorElt(I));
     return Success(Elements, E);
   }
+  case CK_HLSLMatrixTruncation: {
+    // TODO: See #168935. Add matrix truncation support to expr constant.
+    return Error(E);
+  }
   case CK_HLSLAggregateSplatCast: {
     APValue Val;
     QualType ValTy;
@@ -18440,6 +18444,10 @@ bool IntExprEvaluator::VisitCastExpr(const CastExpr *E) {
       return Error(E);
     return Success(Val.getVectorElt(0), E);
   }
+  case CK_HLSLMatrixTruncation: {
+    // TODO: See #168935. Add matrix truncation support to expr constant.
+    return Error(E);
+  }
   case CK_HLSLElementwiseCast: {
     SmallVector<APValue> SrcVals;
     SmallVector<QualType> SrcTypes;
@@ -19033,6 +19041,10 @@ bool FloatExprEvaluator::VisitCastExpr(const CastExpr *E) {
       return Error(E);
     return Success(Val.getVectorElt(0), E);
   }
+  case CK_HLSLMatrixTruncation: {
+    // TODO: See #168935. Add matrix truncation support to expr constant.
+    return Error(E);
+  }
   case CK_HLSLElementwiseCast: {
     SmallVector<APValue> SrcVals;
     SmallVector<QualType> SrcTypes;
@@ -19190,6 +19202,7 @@ bool ComplexExprEvaluator::VisitCastExpr(const CastExpr *E) {
   case CK_IntegralToFixedPoint:
   case CK_MatrixCast:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
     llvm_unreachable("invalid cast kind for complex value");
diff --git a/clang/lib/CIR/CodeGen/CIRGenExpr.cpp b/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
@@ -188,6 +188,7 @@ Address CIRGenFunction::emitPointerWithAlignment(const Expr *expr,
     case CK_HLSLArrayRValue:
     case CK_HLSLElementwiseCast:
     case CK_HLSLVectorTruncation:
+    case CK_HLSLMatrixTruncation:
     case CK_IntToOCLSampler:
     case CK_IntegralCast:
     case CK_IntegralComplexCast:
@@ -1323,6 +1324,7 @@ LValue CIRGenFunction::emitCastLValue(const CastExpr *e) {
   case CK_IntegralToFixedPoint:
   case CK_MatrixCast:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLArrayRValue:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
diff --git a/clang/lib/CIR/CodeGen/CIRGenExprComplex.cpp b/clang/lib/CIR/CodeGen/CIRGenExprComplex.cpp
@@ -534,6 +534,7 @@ mlir::Value ComplexExprEmitter::emitCast(CastKind ck, Expr *op,
   case CK_IntegralToFixedPoint:
   case CK_MatrixCast:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLArrayRValue:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
diff --git a/clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp b/clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp
@@ -1012,6 +1012,7 @@ class ConstExprEmitter
     case CK_MatrixCast:
     case CK_HLSLArrayRValue:
     case CK_HLSLVectorTruncation:
+    case CK_HLSLMatrixTruncation:
     case CK_HLSLElementwiseCast:
     case CK_HLSLAggregateSplatCast:
       return {};
diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
@@ -5798,6 +5798,7 @@ LValue CodeGenFunction::EmitCastLValue(const CastExpr *E) {
   case CK_IntegralToFixedPoint:
   case CK_MatrixCast:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLArrayRValue:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
diff --git a/clang/lib/CodeGen/CGExprAgg.cpp b/clang/lib/CodeGen/CGExprAgg.cpp
@@ -1036,7 +1036,7 @@ void AggExprEmitter::VisitCastExpr(CastExpr *E) {
   case CK_ZeroToOCLOpaqueType:
   case CK_MatrixCast:
   case CK_HLSLVectorTruncation:
-
+  case CK_HLSLMatrixTruncation:
   case CK_IntToOCLSampler:
   case CK_FloatingToFixedPoint:
   case CK_FixedPointToFloating:
@@ -1550,6 +1550,7 @@ static bool castPreservesZero(const CastExpr *CE) {
   case CK_NonAtomicToAtomic:
   case CK_AtomicToNonAtomic:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
     return true;
diff --git a/clang/lib/CodeGen/CGExprComplex.cpp b/clang/lib/CodeGen/CGExprComplex.cpp
@@ -621,6 +621,7 @@ ComplexPairTy ComplexExprEmitter::EmitCast(CastKind CK, Expr *Op,
   case CK_IntegralToFixedPoint:
   case CK_MatrixCast:
   case CK_HLSLVectorTruncation:
+  case CK_HLSLMatrixTruncation:
   case CK_HLSLArrayRValue:
   case CK_HLSLElementwiseCast:
   case CK_HLSLAggregateSplatCast:
diff --git a/clang/lib/CodeGen/CGExprConstant.cpp b/clang/lib/CodeGen/CGExprConstant.cpp
@@ -1333,6 +1333,7 @@ class ConstExprEmitter
     case CK_ZeroToOCLOpaqueType:
     case CK_MatrixCast:
     case CK_HLSLVectorTruncation:
+    case CK_HLSLMatrixTruncation:
     case CK_HLSLArrayRValue:
     case CK_HLSLElementwiseCast:
     case CK_HLSLAggregateSplatCast:
diff --git a/clang/lib/CodeGen/CGExprScalar.cpp b/clang/lib/CodeGen/CGExprScalar.cpp
@@ -2422,9 +2422,31 @@ static Value *EmitHLSLElementwiseCast(CodeGenFunction &CGF, LValue SrcVal,
     }
     return V;
   }
+  if (auto *MatTy = DestTy->getAs<ConstantMatrixType>()) {
+    assert(LoadList.size() >= MatTy->getNumElementsFlattened() &&
+           "Flattened type on RHS must have the same number or more elements "
+           "than vector on LHS.");
+
+    llvm::Value *V =
+        CGF.Builder.CreateLoad(CGF.CreateIRTemp(DestTy, "flatcast.tmp"));
+    // V is an allocated temporary to build the truncated matrix into.
+    for (unsigned I = 0, E = MatTy->getNumElementsFlattened(); I < E; I++) {
+      unsigned ColMajorIndex =
+          (I % MatTy->getNumRows()) * MatTy->getNumColumns() +
+          (I / MatTy->getNumRows());
+      RValue RVal = CGF.EmitLoadOfLValue(LoadList[ColMajorIndex], Loc);
+      assert(RVal.isScalar() &&
+             "All flattened source values should be scalars.");
+      llvm::Value *Cast = CGF.EmitScalarConversion(
+          RVal.getScalarVal(), LoadList[ColMajorIndex].getType(),
+          MatTy->getElementType(), Loc);
+      V = CGF.Builder.CreateInsertElement(V, Cast, I);
+    }
+    return V;
+  }
   // if its a builtin just do an extract element or load.
   assert(DestTy->isBuiltinType() &&
-         "Destination type must be a vector or builtin type.");
+         "Destination type must be a vector, matrix, or builtin type.");
   RValue RVal = CGF.EmitLoadOfLValue(LoadList[0], Loc);
   assert(RVal.isScalar() && "All flattened source values should be scalars.");
   return CGF.EmitScalarConversion(RVal.getScalarVal(), LoadList[0].getType(),
@@ -2954,15 +2976,47 @@ Value *ScalarExprEmitter::VisitCastExpr(CastExpr *CE) {
     llvm::Value *Zero = llvm::Constant::getNullValue(CGF.SizeTy);
     return Builder.CreateExtractElement(Vec, Zero, "cast.vtrunc");
   }
+  case CK_HLSLMatrixTruncation: {
+    assert((DestTy->isMatrixType() || DestTy->isBuiltinType()) &&
+           "Destination type must be a matrix or builtin type.");
+    Value *Mat = Visit(E);
+    if (auto *MatTy = DestTy->getAs<ConstantMatrixType>()) {
+      SmallVector<int> Mask;
+      unsigned NumCols = MatTy->getNumColumns();
+      unsigned NumRows = MatTy->getNumRows();
+      unsigned ColOffset = NumCols;
+      if (auto *SrcMatTy = E->getType()->getAs<ConstantMatrixType>())
+        ColOffset = SrcMatTy->getNumColumns();
+      for (unsigned R = 0; R < NumRows; R++) {
+        for (unsigned C = 0; C < NumCols; C++) {
+          unsigned I = R * ColOffset + C;
+          Mask.push_back(I);
+        }
+      }
+
+      return Builder.CreateShuffleVector(Mat, Mask, "trunc");
+    }
+    llvm::Value *Zero = llvm::Constant::getNullValue(CGF.SizeTy);
+    return Builder.CreateExtractElement(Mat, Zero, "cast.mtrunc");
+  }
   case CK_HLSLElementwiseCast: {
     RValue RV = CGF.EmitAnyExpr(E);
     SourceLocation Loc = CE->getExprLoc();
 
-    assert(RV.isAggregate() && "Not a valid HLSL Elementwise Cast.");
-    // RHS is an aggregate
-    LValue SrcVal = CGF.MakeAddrLValue(RV.getAggregateAddress(), E->getType());
+    Address SrcAddr = Address::invalid();
+
+    if (RV.isAggregate()) {
+      SrcAddr = RV.getAggregateAddress();
+    } else {
+      SrcAddr = CGF.CreateMemTemp(E->getType(), "hlsl.ewcast.src");
+      LValue TmpLV = CGF.MakeAddrLValue(SrcAddr, E->getType());
+      CGF.EmitStoreThroughLValue(RV, TmpLV);
+    }
+
+    LValue SrcVal = CGF.MakeAddrLValue(SrcAddr, E->getType());
     return EmitHLSLElementwiseCast(CGF, SrcVal, DestTy, Loc);
   }
+
   } // end of switch
 
   llvm_unreachable("unknown scalar cast");
diff --git a/clang/lib/Edit/RewriteObjCFoundationAPI.cpp b/clang/lib/Edit/RewriteObjCFoundationAPI.cpp
@@ -1085,6 +1085,7 @@ static bool rewriteToNumericBoxedExpression(const ObjCMessageExpr *Msg,
       llvm_unreachable("OpenCL-specific cast in Objective-C?");
 
     case CK_HLSLVectorTruncation:
+    case CK_HLSLMatrixTruncation:
     case CK_HLSLElementwiseCast:
     case CK_HLSLAggregateSplatCast:
       llvm_unreachable("HLSL-specific cast in Objective-C?");
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
@@ -37,6 +37,7 @@
 #include "clang/AST/TemplateBase.h"
 #include "clang/AST/TemplateName.h"
 #include "clang/AST/Type.h"
+#include "clang/AST/TypeBase.h"
 #include "clang/AST/TypeLoc.h"
 #include "clang/AST/UnresolvedSet.h"
 #include "clang/Basic/AddressSpaces.h"
@@ -12570,9 +12571,10 @@ void Sema::CheckImplicitConversion(Expr *E, QualType T, SourceLocation CC,
       if (SourceMgr.isInSystemMacro(CC))
         return;
       return DiagnoseImpCast(*this, E, T, CC, diag::warn_impcast_vector_scalar);
-    } else if (getLangOpts().HLSL &&
-               Target->castAs<VectorType>()->getNumElements() <
-                   Source->castAs<VectorType>()->getNumElements()) {
+    }
+    if (getLangOpts().HLSL &&
+        Target->castAs<VectorType>()->getNumElements() <
+            Source->castAs<VectorType>()->getNumElements()) {
       // Diagnose vector truncation but don't return. We may also want to
       // diagnose an element conversion.
       DiagnoseImpCast(*this, E, T, CC,
@@ -12588,9 +12590,22 @@ void Sema::CheckImplicitConversion(Expr *E, QualType T, SourceLocation CC,
     Source = cast<VectorType>(Source)->getElementType().getTypePtr();
     Target = cast<VectorType>(Target)->getElementType().getTypePtr();
   }
-  if (auto VecTy = dyn_cast<VectorType>(Target))
+  if (const auto *VecTy = dyn_cast<VectorType>(Target))
     Target = VecTy->getElementType().getTypePtr();
 
+  if (isa<ConstantMatrixType>(Source)) {
+    if (Target->isScalarType())
+      return DiagnoseImpCast(*this, E, T, CC, diag::warn_impcast_matrix_scalar);
+
+    if (getLangOpts().HLSL &&
+        Target->castAs<ConstantMatrixType>()->getNumElementsFlattened() <
+            Source->castAs<ConstantMatrixType>()->getNumElementsFlattened()) {
+      // Diagnose Matrix truncation but don't return. We may also want to
+      // diagnose an element conversion.
+      DiagnoseImpCast(*this, E, T, CC,
+                      diag::warn_hlsl_impcast_matrix_truncation);
+    }
+  }
   // Strip complex types.
   if (isa<ComplexType>(Source)) {
     if (!isa<ComplexType>(Target)) {
diff --git a/clang/lib/Sema/SemaExprCXX.cpp b/clang/lib/Sema/SemaExprCXX.cpp
@@ -5196,19 +5196,18 @@ Sema::PerformImplicitConversion(Expr *From, QualType ToType,
   case ICK_Incompatible_Pointer_Conversion:
   case ICK_HLSL_Array_RValue:
   case ICK_HLSL_Vector_Truncation:
+  case ICK_HLSL_Matrix_Truncation:
   case ICK_HLSL_Vector_Splat:
     llvm_unreachable("Improper second standard conversion");
   }
 
   if (SCS.Dimension != ICK_Identity) {
     // If SCS.Element is not ICK_Identity the To and From types must be HLSL
     // vectors or matrices.
-
-    // TODO: Support HLSL matrices.
-    assert((!From->getType()->isMatrixType() && !ToType->isMatrixType()) &&
-           "Dimension conversion for matrix types is not implemented yet.");
-    assert((ToType->isVectorType() || ToType->isBuiltinType()) &&
-           "Dimension conversion output must be vector or scalar type.");
+    assert(
+        (ToType->isVectorType() || ToType->isConstantMatrixType() ||
+         ToType->isBuiltinType()) &&
+        "Dimension conversion output must be vector, matrix, or scalar type.");
     switch (SCS.Dimension) {
     case ICK_HLSL_Vector_Splat: {
       // Vector splat from any arithmetic type to a vector.
@@ -5234,6 +5233,17 @@ Sema::PerformImplicitConversion(Expr *From, QualType ToType,
 
       break;
     }
+    case ICK_HLSL_Matrix_Truncation: {
+      auto *FromMat = From->getType()->castAs<ConstantMatrixType>();
+      QualType TruncTy = FromMat->getElementType();
+      if (auto *ToMat = ToType->getAs<ConstantMatrixType>())
+        TruncTy = Context.getConstantMatrixType(TruncTy, ToMat->getNumRows(),
+                                                ToMat->getNumColumns());
+      From = ImpCastExprToType(From, TruncTy, CK_HLSLMatrixTruncation,
+                               From->getValueKind())
+                 .get();
+      break;
+    }
     case ICK_Identity:
     default:
       llvm_unreachable("Improper element standard conversion");
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
@@ -3823,7 +3823,6 @@ bool SemaHLSL::CanPerformAggregateSplatCast(Expr *Src, QualType DestTy) {
 }
 
 // Can we perform an HLSL Elementwise cast?
-// TODO: update this code when matrices are added; see issue #88060
 bool SemaHLSL::CanPerformElementwiseCast(Expr *Src, QualType DestTy) {
 
   // Don't handle casts where LHS and RHS are any combination of scalar/vector
@@ -3836,6 +3835,10 @@ bool SemaHLSL::CanPerformElementwiseCast(Expr *Src, QualType DestTy) {
       (DestTy->isScalarType() || DestTy->isVectorType()))
     return false;
 
+  if (SrcTy->isConstantMatrixType() &&
+      (DestTy->isScalarType() || DestTy->isConstantMatrixType()))
+    return false;
+
   llvm::SmallVector<QualType> DestTypes;
   BuildFlattenedTypeList(DestTy, DestTypes);
   llvm::SmallVector<QualType> SrcTypes;
diff --git a/clang/lib/Sema/SemaOverload.cpp b/clang/lib/Sema/SemaOverload.cpp
diff --git a/clang/lib/StaticAnalyzer/Core/ExprEngineC.cpp b/clang/lib/StaticAnalyzer/Core/ExprEngineC.cpp
diff --git a/clang/test/CodeGenHLSL/BasicFeatures/MatrixElementTypeCast.hlsl b/clang/test/CodeGenHLSL/BasicFeatures/MatrixElementTypeCast.hlsl
diff --git a/clang/test/CodeGenHLSL/BasicFeatures/MatrixExplicitTruncation.hlsl b/clang/test/CodeGenHLSL/BasicFeatures/MatrixExplicitTruncation.hlsl
diff --git a/clang/test/CodeGenHLSL/BasicFeatures/MatrixImplicitTruncation.hlsl b/clang/test/CodeGenHLSL/BasicFeatures/MatrixImplicitTruncation.hlsl
diff --git a/clang/test/Driver/autocomplete.c b/clang/test/Driver/autocomplete.c
diff --git a/clang/test/SemaHLSL/MatrixElementOverloadResolution.hlsl b/clang/test/SemaHLSL/MatrixElementOverloadResolution.hlsl
diff --git a/clang/test/SemaHLSL/Types/BuiltinMatrix/MatrixCastErrors.hlsl b/clang/test/SemaHLSL/Types/BuiltinMatrix/MatrixCastErrors.hlsl
diff --git a/clang/test/SemaHLSL/Types/BuiltinMatrix/MatrixImplicitTruncCastWarnings.hlsl b/clang/test/SemaHLSL/Types/BuiltinMatrix/MatrixImplicitTruncCastWarnings.hlsl
