Merged main:2f1399c73f52 into amd-gfx:88b84f4b16eb

Local branch amd-gfx 88b84f4 Merged main:16b8c9608f20 into amd-gfx:eb2567585b87
Remote branch main 2f1399c clang/APINotes: fix parenthesization of &&, ||

Author: SC llvm team

clang/include/clang/APINotes/APINotesManager.h

@@ -159,7 +159,7 @@ class APINotesManager {
   ArrayRef<APINotesReader *> getCurrentModuleReaders() const {
     bool HasPublic = CurrentModuleReaders[ReaderKind::Public];
     bool HasPrivate = CurrentModuleReaders[ReaderKind::Private];
-    assert(!HasPrivate || (HasPublic && "private module requires public module"));
+    assert((!HasPrivate || HasPublic) && "private module requires public module");
     if (!HasPrivate && !HasPublic)
       return {};
     return ArrayRef(CurrentModuleReaders).slice(0, HasPrivate ? 2 : 1);

clang/include/clang/Basic/AttrDocs.td

@@ -7315,7 +7315,7 @@ this usage is debugger-dependent.
 }
 
 def CleanupDocs : Documentation {
-  let Category = DocCatType;
+  let Category = DocCatVariable;
   let Content = [{
 This attribute allows a function to be run when a local variable goes out of
 scope. The attribute takes the identifier of a function with a parameter type

clang/include/clang/Sema/Overload.h

@@ -257,10 +257,7 @@ class Sema;
   /// sequence (C++ 13.3.3.1.1). A standard conversion sequence
   /// contains between zero and three conversions. If a particular
   /// conversion is not needed, it will be set to the identity conversion
-  /// (ICK_Identity). Note that the three conversions are
-  /// specified as separate members (rather than in an array) so that
-  /// we can keep the size of a standard conversion sequence to a
-  /// single word.
+  /// (ICK_Identity).
   class StandardConversionSequence {
   public:
     /// First -- The first conversion can be an lvalue-to-rvalue

clang/lib/CodeGen/CGExpr.cpp

@@ -2158,6 +2158,14 @@ RValue CodeGenFunction::EmitLoadOfExtVectorElementLValue(LValue LV) {
   llvm::Value *Vec = Builder.CreateLoad(LV.getExtVectorAddress(),
                                         LV.isVolatileQualified());
 
+  // HLSL allows treating scalars as one-element vectors. Converting the scalar
+  // IR value to a vector here allows the rest of codegen to behave as normal.
+  if (getLangOpts().HLSL && !Vec->getType()->isVectorTy()) {
+    llvm::Type *DstTy = llvm::FixedVectorType::get(Vec->getType(), 1);
+    llvm::Value *Zero = llvm::Constant::getNullValue(CGM.Int64Ty);
+    Vec = Builder.CreateInsertElement(DstTy, Vec, Zero, "cast.splat");
+  }
+
   const llvm::Constant *Elts = LV.getExtVectorElts();
 
   // If the result of the expression is a non-vector type, we must be extracting
@@ -2427,10 +2435,20 @@ void CodeGenFunction::EmitStoreThroughBitfieldLValue(RValue Src, LValue Dst,
 
 void CodeGenFunction::EmitStoreThroughExtVectorComponentLValue(RValue Src,
                                                                LValue Dst) {
+  // HLSL allows storing to scalar values through ExtVector component LValues.
+  // To support this we need to handle the case where the destination address is
+  // a scalar.
+  Address DstAddr = Dst.getExtVectorAddress();
+  if (!DstAddr.getElementType()->isVectorTy()) {
+    assert(!Dst.getType()->isVectorType() &&
+           "this should only occur for non-vector l-values");
+    Builder.CreateStore(Src.getScalarVal(), DstAddr, Dst.isVolatileQualified());
+    return;
+  }
+
   // This access turns into a read/modify/write of the vector.  Load the input
   // value now.
-  llvm::Value *Vec = Builder.CreateLoad(Dst.getExtVectorAddress(),
-                                        Dst.isVolatileQualified());
+  llvm::Value *Vec = Builder.CreateLoad(DstAddr, Dst.isVolatileQualified());
   const llvm::Constant *Elts = Dst.getExtVectorElts();
 
   llvm::Value *SrcVal = Src.getScalarVal();
@@ -2478,7 +2496,8 @@ void CodeGenFunction::EmitStoreThroughExtVectorComponentLValue(RValue Src,
       llvm_unreachable("unexpected shorten vector length");
     }
   } else {
-    // If the Src is a scalar (not a vector) it must be updating one element.
+    // If the Src is a scalar (not a vector), and the target is a vector it must
+    // be updating one element.
     unsigned InIdx = getAccessedFieldNo(0, Elts);
     llvm::Value *Elt = llvm::ConstantInt::get(SizeTy, InIdx);
     Vec = Builder.CreateInsertElement(Vec, SrcVal, Elt);
@@ -4879,7 +4898,6 @@ LValue CodeGenFunction::EmitCastLValue(const CastExpr *E) {
   case CK_IntegralToPointer:
   case CK_PointerToIntegral:
   case CK_PointerToBoolean:
-  case CK_VectorSplat:
   case CK_IntegralCast:
   case CK_BooleanToSignedIntegral:
   case CK_IntegralToBoolean:
@@ -5044,6 +5062,13 @@ LValue CodeGenFunction::EmitCastLValue(const CastExpr *E) {
   }
   case CK_ZeroToOCLOpaqueType:
     llvm_unreachable("NULL to OpenCL opaque type lvalue cast is not valid");
+
+  case CK_VectorSplat: {
+    // LValue results of vector splats are only supported in HLSL.
+    if (!getLangOpts().HLSL)
+      return EmitUnsupportedLValue(E, "unexpected cast lvalue");
+    return EmitLValue(E->getSubExpr());
+  }
   }
 
   llvm_unreachable("Unhandled lvalue cast kind?");

clang/lib/Driver/ToolChains/Flang.cpp

@@ -142,26 +142,6 @@ void Flang::addCodegenOptions(const ArgList &Args,
   if (shouldLoopVersion(Args))
     CmdArgs.push_back("-fversion-loops-for-stride");
 
-  Arg *aliasAnalysis = Args.getLastArg(options::OPT_falias_analysis,
-                                       options::OPT_fno_alias_analysis);
-  // only pass on the argument if it does not match that implied by the
-  // optimization level: so if optimization is requested, only forward
-  // -fno-alias-analysis. If optimization is not requested, only forward
-  // -falias-analysis.
-  Arg *optLevel =
-      Args.getLastArg(options::OPT_Ofast, options::OPT_O, options::OPT_O4);
-  if (aliasAnalysis) {
-    bool faliasAnalysis =
-        aliasAnalysis->getOption().matches(options::OPT_falias_analysis);
-    if (optLevel && !faliasAnalysis) {
-      CmdArgs.push_back("-fno-alias-analysis");
-    } else {
-      if (faliasAnalysis)
-        // requested alias analysis but no optimization enabled
-        CmdArgs.push_back("-falias-analysis");
-    }
-  }
-
   Args.addAllArgs(CmdArgs, {options::OPT_flang_experimental_hlfir,
                             options::OPT_flang_deprecated_no_hlfir,
                             options::OPT_flang_experimental_polymorphism,

clang/lib/Lex/Lexer.cpp

@@ -1990,6 +1990,10 @@ bool Lexer::LexNumericConstant(Token &Result, const char *CurPtr) {
   while (isPreprocessingNumberBody(C)) {
     CurPtr = ConsumeChar(CurPtr, Size, Result);
     PrevCh = C;
+    if (LangOpts.HLSL && C == '.' && (*CurPtr == 'x' || *CurPtr == 'r')) {
+      CurPtr -= Size;
+      break;
+    }
     C = getCharAndSize(CurPtr, Size);
   }
 

clang/lib/Lex/LiteralSupport.cpp

@@ -930,7 +930,11 @@ NumericLiteralParser::NumericLiteralParser(StringRef TokSpelling,
   // and FP constants (specifically, the 'pp-number' regex), and assumes that
   // the byte at "*end" is both valid and not part of the regex.  Because of
   // this, it doesn't have to check for 'overscan' in various places.
-  if (isPreprocessingNumberBody(*ThisTokEnd)) {
+  // Note: For HLSL, the end token is allowed to be '.' which would be in the
+  // 'pp-number' regex. This is required to support vector swizzles on numeric
+  // constants (i.e. 1.xx or 1.5f.rrr).
+  if (isPreprocessingNumberBody(*ThisTokEnd) &&
+      !(LangOpts.HLSL && *ThisTokEnd == '.')) {
     Diags.Report(TokLoc, diag::err_lexing_numeric);
     hadError = true;
     return;

clang/lib/Sema/SemaExprMember.cpp

@@ -1714,6 +1714,16 @@ static ExprResult LookupMemberExpr(Sema &S, LookupResult &R,
                             ObjCImpDecl, HasTemplateArgs, TemplateKWLoc);
   }
 
+  // HLSL supports implicit conversion of scalar types to single element vector
+  // rvalues in member expressions.
+  if (S.getLangOpts().HLSL && BaseType->isScalarType()) {
+    QualType VectorTy = S.Context.getExtVectorType(BaseType, 1);
+    BaseExpr = S.ImpCastExprToType(BaseExpr.get(), VectorTy, CK_VectorSplat,
+                                   BaseExpr.get()->getValueKind());
+    return LookupMemberExpr(S, R, BaseExpr, IsArrow, OpLoc, SS, ObjCImpDecl,
+                            HasTemplateArgs, TemplateKWLoc);
+  }
+
   S.Diag(OpLoc, diag::err_typecheck_member_reference_struct_union)
     << BaseType << BaseExpr.get()->getSourceRange() << MemberLoc;
 

clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl

@@ -0,0 +1,174 @@
+// RUN: %clang_cc1 -std=hlsl2021 -finclude-default-header -x hlsl -triple \
+// RUN:   dxil-pc-shadermodel6.3-library %s -emit-llvm -disable-llvm-passes \
+// RUN:   -o - | FileCheck %s
+
+// CHECK-LABEL: ToTwoInts
+// CHECK: [[splat:%.*]] = insertelement <1 x i32> poison, i32 {{.*}}, i64 0
+// CHECK: [[vec2:%.*]] = shufflevector <1 x i32> [[splat]], <1 x i32> poison, <2 x i32> zeroinitializer
+// CHECK: ret <2 x i32> [[vec2]]
+int2 ToTwoInts(int V){
+  return V.xx;
+}
+
+// CHECK-LABEL: ToFourFloats
+// [[splat:%.*]] = insertelement <1 x float> poison, float {{.*}}, i64 0
+// [[vec4:%.*]] = shufflevector <1 x float> [[splat]], <1 x float> poison, <4 x i32> zeroinitializer
+// ret <4 x float> [[vec4]]
+float4 ToFourFloats(float V){
+  return V.rrrr;
+}
+
+// CHECK-LABEL: FillOne
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x i32>, align 4
+// CHECK: store <1 x i32> <i32 1>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec1:%.*]] = load <1 x i32>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec2:%.*]] = shufflevector <1 x i32> [[vec1]], <1 x i32> poison, <2 x i32> zeroinitializer
+// CHECK: ret <2 x i32> [[vec2]]
+int2 FillOne(){
+  return 1.xx;
+}
+
+// CHECK-LABEL: FillOneUnsigned
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x i32>, align 4
+// CHECK: store <1 x i32> <i32 1>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec1:%.*]] = load <1 x i32>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec3:%.*]] = shufflevector <1 x i32> [[vec1]], <1 x i32> poison, <3 x i32> zeroinitializer
+// CHECK: ret <3 x i32> [[vec3]]
+uint3 FillOneUnsigned(){
+  return 1u.xxx;
+}
+
+// CHECK-LABEL: FillOneUnsignedLong
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x i64>, align 8
+// CHECK: store <1 x i64> <i64 1>, ptr [[vec1Ptr]], align 8
+// CHECK: [[vec1:%.*]] = load <1 x i64>, ptr [[vec1Ptr]], align 8
+// CHECK: [[vec4:%.*]] = shufflevector <1 x i64> [[vec1]], <1 x i64> poison, <4 x i32> zeroinitializer
+// CHECK: ret <4 x i64> [[vec4]]
+vector<uint64_t,4> FillOneUnsignedLong(){
+  return 1ul.xxxx;
+}
+
+// CHECK-LABEL: FillTwoPointFive
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x double>, align 8
+// CHECK: store <1 x double> <double 2.500000e+00>, ptr [[vec1Ptr]], align 8
+// CHECK: [[vec1:%.*]] = load <1 x double>, ptr [[vec1Ptr]], align 8
+// CHECK: [[vec2:%.*]] = shufflevector <1 x double> [[vec1]], <1 x double> poison, <2 x i32> zeroinitializer
+// CHECK: ret <2 x double> [[vec2]]
+double2 FillTwoPointFive(){
+  return 2.5.rr;
+}
+
+// CHECK-LABEL: FillOneHalf
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x double>, align 8
+// CHECK: store <1 x double> <double 5.000000e-01>, ptr [[vec1Ptr]], align 8
+// CHECK: [[vec1:%.*]] = load <1 x double>, ptr [[vec1Ptr]], align 8
+// CHECK: [[vec3:%.*]] = shufflevector <1 x double> [[vec1]], <1 x double> poison, <3 x i32> zeroinitializer
+// CHECK: ret <3 x double> [[vec3]]
+double3 FillOneHalf(){
+  return .5.rrr;
+}
+
+// CHECK-LABEL: FillTwoPointFiveFloat
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x float>, align 4
+// CHECK: store <1 x float> <float 2.500000e+00>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec1:%.*]] = load <1 x float>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec4:%.*]] = shufflevector <1 x float> [[vec1]], <1 x float> poison, <4 x i32> zeroinitializer
+// CHECK: ret <4 x float> [[vec4]]
+float4 FillTwoPointFiveFloat(){
+  return 2.5f.rrrr;
+}
+
+// The initial codegen for this case is correct but a bit odd. The IR optimizer
+// cleans this up very nicely.
+
+// CHECK-LABEL: FillOneHalfFloat
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x float>, align 4
+// CHECK: store <1 x float> <float 5.000000e-01>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec1:%.*]] = load <1 x float>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec1Ret:%.*]] = shufflevector <1 x float> [[vec1]], <1 x float> undef, <1 x i32> zeroinitializer
+// CHECK: ret <1 x float> [[vec1Ret]]
+vector<float, 1> FillOneHalfFloat(){
+  return .5f.r;
+}
+
+// The initial codegen for this case is correct but a bit odd. The IR optimizer
+// cleans this up very nicely.
+
+// CHECK-LABEL: HowManyFloats
+// CHECK: [[VAddr:%.*]] = alloca float, align 4
+// CHECK: [[vec2Ptr:%.*]] = alloca <2 x float>, align 8
+// CHECK: [[VVal:%.*]] = load float, ptr [[VAddr]], align 4
+// CHECK: [[splat:%.*]] = insertelement <1 x float> poison, float [[VVal]], i64 0
+// CHECK: [[vec2:%.*]] = shufflevector <1 x float> [[splat]], <1 x float> poison, <2 x i32> zeroinitializer
+// CHECK: store <2 x float> [[vec2]], ptr [[vec2Ptr]], align 8
+// CHECK: [[vec2:%.*]] = load <2 x float>, ptr [[vec2Ptr]], align 8
+// CHECK: [[vec2Res:%.*]] = shufflevector <2 x float> [[vec2]], <2 x float> poison, <2 x i32> zeroinitializer
+// CHECK: ret <2 x float> [[vec2Res]]
+float2 HowManyFloats(float V) {
+  return V.rr.rr;
+}
+
+// This codegen is gnarly because `1.` is a double, so this creates double
+// vectors that need to be truncated down to floats. The optimizer cleans this
+// up nicely too.
+
+// CHECK-LABEL: AllRighty
+// CHECK: [[XTmp:%.*]] = alloca <1 x double>, align 8
+// CHECK: [[YTmp:%.*]] = alloca <1 x double>, align 8
+// CHECK: [[ZTmp:%.*]] = alloca <1 x double>, align 8
+
+// CHECK: store <1 x double> <double 1.000000e+00>, ptr [[XTmp]], align 8
+// CHECK: [[XVec:%.*]] = load <1 x double>, ptr [[XTmp]], align 8
+// CHECK: [[XVec3:%.*]] = shufflevector <1 x double> [[XVec]], <1 x double> poison, <3 x i32> zeroinitializer
+// CHECK: [[XVal:%.*]] = extractelement <3 x double> [[XVec3]], i32 0
+// CHECK: [[XValF:%.*]] = fptrunc double [[XVal]] to float
+// CHECK: [[Vec3F1:%.*]] = insertelement <3 x float> undef, float [[XValF]], i32 0
+
+// CHECK: store <1 x double> <double 1.000000e+00>, ptr [[YTmp]], align 8
+// CHECK: [[YVec:%.*]] = load <1 x double>, ptr [[YTmp]], align 8
+// CHECK: [[YVec3:%.*]] = shufflevector <1 x double> [[YVec]], <1 x double> poison, <3 x i32> zeroinitializer
+// CHECK: [[YVal:%.*]] = extractelement <3 x double> [[YVec3]], i32 1
+// CHECK: [[YValF:%.*]] = fptrunc double [[YVal]] to float
+// CHECK: [[Vec3F2:%.*]] = insertelement <3 x float> [[Vec3F1]], float [[YValF]], i32 1
+
+// CHECK: store <1 x double> <double 1.000000e+00>, ptr [[ZTmp]], align 8
+// CHECK: [[ZVec:%.*]] = load <1 x double>, ptr [[ZTmp]], align 8
+// CHECK: [[ZVec3:%.*]] = shufflevector <1 x double> [[ZVec]], <1 x double> poison, <3 x i32> zeroinitializer
+// CHECK: [[ZVal:%.*]] = extractelement <3 x double> [[ZVec3]], i32 2
+// CHECK: [[ZValF:%.*]] = fptrunc double [[ZVal]] to float
+// CHECK: [[Vec3F3:%.*]] = insertelement <3 x float> [[Vec3F2]], float [[ZValF]], i32 2
+
+// ret <3 x float> [[Vec3F3]]
+float3 AllRighty() {
+  return 1..rrr;
+}
+
+// CHECK-LABEL: AssignInt
+// CHECK: [[VAddr:%.*]] = alloca i32, align 4
+// CHECK: [[XAddr:%.*]] = alloca i32, align 4
+
+// Load V into a vector, then extract V out and store it to X.
+// CHECK: [[V:%.*]] = load i32, ptr [[VAddr]], align 4
+// CHECK: [[Splat:%.*]] = insertelement <1 x i32> poison, i32 [[V]], i64 0
+// CHECK: [[VExtVal:%.*]] = extractelement <1 x i32> [[Splat]], i32 0
+// CHECK: store i32 [[VExtVal]], ptr [[XAddr]], align 4
+
+// Load V into two separate vectors, then add the extracted X components.
+// CHECK: [[V:%.*]] = load i32, ptr [[VAddr]], align 4
+// CHECK: [[Splat:%.*]] = insertelement <1 x i32> poison, i32 [[V]], i64 0
+// CHECK: [[LHS:%.*]] = extractelement <1 x i32> [[Splat]], i32 0
+
+// CHECK: [[V:%.*]] = load i32, ptr [[VAddr]], align 4
+// CHECK: [[Splat:%.*]] = insertelement <1 x i32> poison, i32 [[V]], i64 0
+// CHECK: [[RHS:%.*]] = extractelement <1 x i32> [[Splat]], i32 0
+
+// CHECK: [[Sum:%.*]] = add nsw i32 [[LHS]], [[RHS]]
+// CHECK: store i32 [[Sum]], ptr [[XAddr]], align 4
+// CHECK: [[X:%.*]] = load i32, ptr [[XAddr]], align 4
+// CHECK: ret i32 [[X]]
+
+int AssignInt(int V){
+  int X = V.x;
+  X.x = V.x + V.x;
+  return X;
+}

clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl

@@ -0,0 +1,31 @@
+// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-library  -x hlsl -finclude-default-header -verify %s
+
+int2 ToTwoInts(int V) {
+  return V.xy; // expected-error{{vector component access exceeds type 'int __attribute__((ext_vector_type(1)))' (vector of 1 'int' value)}}
+}
+
+float2 ToTwoFloats(float V) {
+  return V.rg; // expected-error{{vector component access exceeds type 'float __attribute__((ext_vector_type(1)))' (vector of 1 'float' value)}}
+}
+
+int4 SomeNonsense(int V) {
+  return V.poop; // expected-error{{illegal vector component name 'p'}}
+}
+
+float2 WhatIsHappening(float V) {
+  return V.; // expected-error{{expected unqualified-id}}
+}
+
+// These cases produce no error.
+
+float2 HowManyFloats(float V) {
+  return V.rr.rr;
+}
+
+int64_t4 HooBoy() {
+  return 4l.xxxx;
+}
+
+float3 AllRighty() {
+  return 1..rrr;
+}