merge main into amd-staging

clang/include/clang/Basic/BuiltinsX86.td

@@ -156,8 +156,6 @@ let Features = "sse", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in
   def rcpss : X86Builtin<"_Vector<4, float>(_Vector<4, float>)">;
   def rsqrtps : X86Builtin<"_Vector<4, float>(_Vector<4, float>)">;
   def rsqrtss : X86Builtin<"_Vector<4, float>(_Vector<4, float>)">;
-  def sqrtps : X86Builtin<"_Vector<4, float>(_Vector<4, float>)">;
-  def sqrtss : X86Builtin<"_Vector<4, float>(_Vector<4, float>)">;
 }
 
 let Features = "sse2", Attributes = [NoThrow, RequiredVectorWidth<128>] in {
@@ -170,8 +168,6 @@ let Features = "sse2", Attributes = [NoThrow] in {
 
 let Features = "sse2", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
   def psadbw128 : X86Builtin<"_Vector<2, long long int>(_Vector<16, char>, _Vector<16, char>)">;
-  def sqrtpd : X86Builtin<"_Vector<2, double>(_Vector<2, double>)">;
-  def sqrtsd : X86Builtin<"_Vector<2, double>(_Vector<2, double>)">;
   def cvtpd2dq : X86Builtin<"_Vector<2, long long int>(_Vector<2, double>)">;
   def cvtpd2ps : X86Builtin<"_Vector<4, float>(_Vector<2, double>)">;
   def cvttpd2dq : X86Builtin<"_Vector<4, int>(_Vector<2, double>)">;
@@ -513,8 +509,6 @@ let Features = "avx", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWid
 }
 
 let Features = "avx", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
-  def sqrtpd256 : X86Builtin<"_Vector<4, double>(_Vector<4, double>)">;
-  def sqrtps256 : X86Builtin<"_Vector<8, float>(_Vector<8, float>)">;
   def rsqrtps256 : X86Builtin<"_Vector<8, float>(_Vector<8, float>)">;
   def rcpps256 : X86Builtin<"_Vector<8, float>(_Vector<8, float>)">;
   def roundpd256 : X86Builtin<"_Vector<4, double>(_Vector<4, double>, _Constant int)">;
@@ -716,11 +710,13 @@ let Features = "avx2", Attributes = [NoThrow, RequiredVectorWidth<128>] in {
   def gatherq_d : X86Builtin<"_Vector<4, int>(_Vector<4, int>, int const *, _Vector<2, long long int>, _Vector<4, int>, _Constant char)">;
 }
 
-let Features = "f16c", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
+let Features = "f16c",
+    Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in {
   def vcvtps2ph : X86Builtin<"_Vector<8, short>(_Vector<4, float>, _Constant int)">;
 }
 
-let Features = "f16c", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
+let Features = "f16c",
+    Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in {
   def vcvtps2ph256 : X86Builtin<"_Vector<8, short>(_Vector<8, float>, _Constant int)">;
 }
 
@@ -3310,15 +3306,15 @@ let Features = "avx512f", Attributes = [NoThrow, Const, RequiredVectorWidth<128>
   def cvtusi2ss32 : X86Builtin<"_Vector<4, float>(_Vector<4, float>, unsigned int, _Constant int)">;
 }
 
-let Features = "avx512vbmi", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in {
+let Features = "avx512vbmi", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in {
   def vpmultishiftqb512 : X86Builtin<"_Vector<64, char>(_Vector<64, char>, _Vector<64, char>)">;
 }
 
-let Features = "avx512vbmi,avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
+let Features = "avx512vbmi,avx512vl", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in {
   def vpmultishiftqb128 : X86Builtin<"_Vector<16, char>(_Vector<16, char>, _Vector<16, char>)">;
 }
 
-let Features = "avx512vbmi,avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
+let Features = "avx512vbmi,avx512vl", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in {
   def vpmultishiftqb256 : X86Builtin<"_Vector<32, char>(_Vector<32, char>, _Vector<32, char>)">;
 }
 
@@ -3358,10 +3354,6 @@ let Features = "avx512bf16", Attributes = [NoThrow, Const, RequiredVectorWidth<5
   def dpbf16ps_512 : X86Builtin<"_Vector<16, float>(_Vector<16, float>, _Vector<32, __bf16>, _Vector<32, __bf16>)">;
 }
 
-let Features = "avx512bf16", Attributes = [NoThrow, Const] in {
-  def cvtsbf162ss_32 : X86Builtin<"float(__bf16)">;
-}
-
 let Features = "avx512vp2intersect", Attributes = [NoThrow, RequiredVectorWidth<512>] in {
   def vp2intersect_q_512 : X86Builtin<"void(_Vector<8, long long int>, _Vector<8, long long int>, unsigned char *, unsigned char *)">;
 }
@@ -3539,14 +3531,6 @@ let Features = "avx512fp16", Attributes = [NoThrow, Const, RequiredVectorWidth<1
   def reducesh_mask : X86Builtin<"_Vector<8, _Float16>(_Vector<8, _Float16>, _Vector<8, _Float16>, _Vector<8, _Float16>, unsigned char, _Constant int, _Constant int)">;
 }
 
-let Features = "avx512fp16,avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
-  def sqrtph : X86Builtin<"_Vector<8, _Float16>(_Vector<8, _Float16>)">;
-}
-
-let Features = "avx512fp16,avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
-  def sqrtph256 : X86Builtin<"_Vector<16, _Float16>(_Vector<16, _Float16>)">;
-}
-
 let Features = "avx512fp16", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in {
   def sqrtph512 : X86Builtin<"_Vector<32, _Float16>(_Vector<32, _Float16>, _Constant int)">;
 }
@@ -5065,15 +5049,3 @@ let Features = "avx10.2", Attributes = [NoThrow, Const, RequiredVectorWidth<256>
 let Features = "avx10.2", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in {
   def vgetmantbf16512_mask : X86Builtin<"_Vector<32, __bf16>(_Vector<32, __bf16>, _Constant int, _Vector<32, __bf16>, unsigned int)">;
 }
-
-let Features = "avx10.2", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
-  def vsqrtbf16 : X86Builtin<"_Vector<8, __bf16>(_Vector<8, __bf16>)">;
-}
-
-let Features = "avx10.2", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
-  def vsqrtbf16256 : X86Builtin<"_Vector<16, __bf16>(_Vector<16, __bf16>)">;
-}
-
-let Features = "avx10.2", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in {
-  def vsqrtbf16512 : X86Builtin<"_Vector<32, __bf16>(_Vector<32, __bf16>)">;
-}

clang/lib/AST/ByteCode/Interp.cpp

@@ -1435,8 +1435,12 @@ static bool getField(InterpState &S, CodePtr OpPC, const Pointer &Ptr,
     return false;
 
   if (Ptr.isIntegralPointer()) {
-    S.Stk.push<Pointer>(Ptr.asIntPointer().atOffset(S.getASTContext(), Off));
-    return true;
+    if (std::optional<IntPointer> IntPtr =
+            Ptr.asIntPointer().atOffset(S.getASTContext(), Off)) {
+      S.Stk.push<Pointer>(std::move(*IntPtr));
+      return true;
+    }
+    return false;
   }
 
   if (!Ptr.isBlockPointer()) {
@@ -2081,15 +2085,15 @@ bool InvalidShuffleVectorIndex(InterpState &S, CodePtr OpPC, uint32_t Index) {
 
 bool CheckPointerToIntegralCast(InterpState &S, CodePtr OpPC,
                                 const Pointer &Ptr, unsigned BitWidth) {
+  const SourceInfo &E = S.Current->getSource(OpPC);
+  S.CCEDiag(E, diag::note_constexpr_invalid_cast)
+      << 2 << S.getLangOpts().CPlusPlus << S.Current->getRange(OpPC);
+
   if (Ptr.isDummy())
     return false;
   if (Ptr.isFunctionPointer())
     return true;
 
-  const SourceInfo &E = S.Current->getSource(OpPC);
-  S.CCEDiag(E, diag::note_constexpr_invalid_cast)
-      << 2 << S.getLangOpts().CPlusPlus << S.Current->getRange(OpPC);
-
   if (Ptr.isBlockPointer() && !Ptr.isZero()) {
     // Only allow based lvalue casts if they are lossless.
     if (S.getASTContext().getTargetInfo().getPointerWidth(LangAS::Default) !=

clang/lib/AST/ByteCode/Interp.h

@@ -2646,10 +2646,6 @@ template <PrimType Name, class T = typename PrimConv<Name>::T>
 bool CastPointerIntegral(InterpState &S, CodePtr OpPC) {
   const Pointer &Ptr = S.Stk.pop<Pointer>();
 
-  S.CCEDiag(S.Current->getSource(OpPC), diag::note_constexpr_invalid_cast)
-      << diag::ConstexprInvalidCastKind::ThisConversionOrReinterpret
-      << S.getLangOpts().CPlusPlus << S.Current->getRange(OpPC);
-
   if (!CheckPointerToIntegralCast(S, OpPC, Ptr, T::bitWidth()))
     return Invalid(S, OpPC);
 

clang/lib/AST/ByteCode/InterpBuiltin.cpp

@@ -3527,6 +3527,147 @@ static bool interp__builtin_ia32_shufbitqmb_mask(InterpState &S, CodePtr OpPC,
   }
 
   pushInteger(S, RetMask, Call->getType());
+  return true;
+}
+
+static bool interp__builtin_ia32_vcvtps2ph(InterpState &S, CodePtr OpPC,
+                                           const CallExpr *Call) {
+  // Arguments are: vector of floats, rounding immediate
+  assert(Call->getNumArgs() == 2);
+
+  APSInt Imm = popToAPSInt(S, Call->getArg(1));
+  const Pointer &Src = S.Stk.pop<Pointer>();
+  const Pointer &Dst = S.Stk.peek<Pointer>();
+
+  assert(Src.getFieldDesc()->isPrimitiveArray());
+  assert(Dst.getFieldDesc()->isPrimitiveArray());
+
+  const auto *SrcVTy = Call->getArg(0)->getType()->castAs<VectorType>();
+  unsigned SrcNumElems = SrcVTy->getNumElements();
+  const auto *DstVTy = Call->getType()->castAs<VectorType>();
+  unsigned DstNumElems = DstVTy->getNumElements();
+
+  const llvm::fltSemantics &HalfSem =
+      S.getASTContext().getFloatTypeSemantics(S.getASTContext().HalfTy);
+
+  // imm[2] == 1 means use MXCSR rounding mode.
+  // In that case, we can only evaluate if the conversion is exact.
+  int ImmVal = Imm.getZExtValue();
+  bool UseMXCSR = (ImmVal & 4) != 0;
+  bool IsFPConstrained =
+      Call->getFPFeaturesInEffect(S.getASTContext().getLangOpts())
+          .isFPConstrained();
+
+  llvm::RoundingMode RM;
+  if (!UseMXCSR) {
+    switch (ImmVal & 3) {
+    case 0:
+      RM = llvm::RoundingMode::NearestTiesToEven;
+      break;
+    case 1:
+      RM = llvm::RoundingMode::TowardNegative;
+      break;
+    case 2:
+      RM = llvm::RoundingMode::TowardPositive;
+      break;
+    case 3:
+      RM = llvm::RoundingMode::TowardZero;
+      break;
+    default:
+      llvm_unreachable("Invalid immediate rounding mode");
+    }
+  } else {
+    // For MXCSR, we must check for exactness. We can use any rounding mode
+    // for the trial conversion since the result is the same if it's exact.
+    RM = llvm::RoundingMode::NearestTiesToEven;
+  }
+
+  QualType DstElemQT = Dst.getFieldDesc()->getElemQualType();
+  PrimType DstElemT = *S.getContext().classify(DstElemQT);
+
+  for (unsigned I = 0; I != SrcNumElems; ++I) {
+    Floating SrcVal = Src.elem<Floating>(I);
+    APFloat DstVal = SrcVal.getAPFloat();
+
+    bool LostInfo;
+    APFloat::opStatus St = DstVal.convert(HalfSem, RM, &LostInfo);
+
+    if (UseMXCSR && IsFPConstrained && St != APFloat::opOK) {
+      S.FFDiag(S.Current->getSource(OpPC),
+               diag::note_constexpr_dynamic_rounding);
+      return false;
+    }
+
+    INT_TYPE_SWITCH_NO_BOOL(DstElemT, {
+      // Convert the destination value's bit pattern to an unsigned integer,
+      // then reconstruct the element using the target type's 'from' method.
+      uint64_t RawBits = DstVal.bitcastToAPInt().getZExtValue();
+      Dst.elem<T>(I) = T::from(RawBits);
+    });
+  }
+
+  // Zero out remaining elements if the destination has more elements
+  // (e.g., vcvtps2ph converting 4 floats to 8 shorts).
+  if (DstNumElems > SrcNumElems) {
+    for (unsigned I = SrcNumElems; I != DstNumElems; ++I) {
+      INT_TYPE_SWITCH_NO_BOOL(DstElemT, { Dst.elem<T>(I) = T::from(0); });
+    }
+  }
+
+  Dst.initializeAllElements();
+  return true;
+}
+
+static bool interp__builtin_ia32_multishiftqb(InterpState &S, CodePtr OpPC,
+                                              const CallExpr *Call) {
+  assert(Call->getNumArgs() == 2);
+
+  QualType ATy = Call->getArg(0)->getType();
+  QualType BTy = Call->getArg(1)->getType();
+  if (!ATy->isVectorType() || !BTy->isVectorType()) {
+    return false;
+  }
+
+  const Pointer &BPtr = S.Stk.pop<Pointer>();
+  const Pointer &APtr = S.Stk.pop<Pointer>();
+  const auto *AVecT = ATy->castAs<VectorType>();
+  assert(AVecT->getNumElements() ==
+         BTy->castAs<VectorType>()->getNumElements());
+
+  PrimType ElemT = *S.getContext().classify(AVecT->getElementType());
+
+  unsigned NumBytesInQWord = 8;
+  unsigned NumBitsInByte = 8;
+  unsigned NumBytes = AVecT->getNumElements();
+  unsigned NumQWords = NumBytes / NumBytesInQWord;
+  const Pointer &Dst = S.Stk.peek<Pointer>();
+
+  for (unsigned QWordId = 0; QWordId != NumQWords; ++QWordId) {
+    APInt BQWord(64, 0);
+    for (unsigned ByteIdx = 0; ByteIdx != NumBytesInQWord; ++ByteIdx) {
+      unsigned Idx = QWordId * NumBytesInQWord + ByteIdx;
+      INT_TYPE_SWITCH(ElemT, {
+        uint64_t Byte = static_cast<uint64_t>(BPtr.elem<T>(Idx));
+        BQWord.insertBits(APInt(8, Byte & 0xFF), ByteIdx * NumBitsInByte);
+      });
+    }
+
+    for (unsigned ByteIdx = 0; ByteIdx != NumBytesInQWord; ++ByteIdx) {
+      unsigned Idx = QWordId * NumBytesInQWord + ByteIdx;
+      uint64_t Ctrl = 0;
+      INT_TYPE_SWITCH(
+          ElemT, { Ctrl = static_cast<uint64_t>(APtr.elem<T>(Idx)) & 0x3F; });
+
+      APInt Byte(8, 0);
+      for (unsigned BitIdx = 0; BitIdx != NumBitsInByte; ++BitIdx) {
+        Byte.setBitVal(BitIdx, BQWord[(Ctrl + BitIdx) & 0x3F]);
+      }
+      INT_TYPE_SWITCH(ElemT,
+                      { Dst.elem<T>(Idx) = T::from(Byte.getZExtValue()); });
+    }
+  }
+
+  Dst.initializeAllElements();
 
   return true;
 }
@@ -4756,6 +4897,10 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
           return std::make_pair(0, static_cast<int>(LaneOffset + Index));
         });
 
+  case X86::BI__builtin_ia32_vpmultishiftqb128:
+  case X86::BI__builtin_ia32_vpmultishiftqb256:
+  case X86::BI__builtin_ia32_vpmultishiftqb512:
+    return interp__builtin_ia32_multishiftqb(S, OpPC, Call);
   case X86::BI__builtin_ia32_kandqi:
   case X86::BI__builtin_ia32_kandhi:
   case X86::BI__builtin_ia32_kandsi:
@@ -4898,6 +5043,10 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
   case X86::BI__builtin_ia32_insert128i256:
     return interp__builtin_x86_insert_subvector(S, OpPC, Call, BuiltinID);
 
+  case clang::X86::BI__builtin_ia32_vcvtps2ph:
+  case clang::X86::BI__builtin_ia32_vcvtps2ph256:
+    return interp__builtin_ia32_vcvtps2ph(S, OpPC, Call);
+
   case X86::BI__builtin_ia32_vec_ext_v4hi:
   case X86::BI__builtin_ia32_vec_ext_v16qi:
   case X86::BI__builtin_ia32_vec_ext_v8hi:

clang/lib/AST/ByteCode/Pointer.cpp

@@ -895,8 +895,8 @@ std::optional<APValue> Pointer::toRValue(const Context &Ctx,
   return Result;
 }
 
-IntPointer IntPointer::atOffset(const ASTContext &ASTCtx,
-                                unsigned Offset) const {
+std::optional<IntPointer> IntPointer::atOffset(const ASTContext &ASTCtx,
+                                               unsigned Offset) const {
   if (!this->Desc)
     return *this;
   const Record *R = this->Desc->ElemRecord;
@@ -914,6 +914,9 @@ IntPointer IntPointer::atOffset(const ASTContext &ASTCtx,
     return *this;
 
   const FieldDecl *FD = F->Decl;
+  if (FD->getParent()->isInvalidDecl())
+    return std::nullopt;
+
   const ASTRecordLayout &Layout = ASTCtx.getASTRecordLayout(FD->getParent());
   unsigned FieldIndex = FD->getFieldIndex();
   uint64_t FieldOffset =

clang/lib/AST/ByteCode/Pointer.h

@@ -47,7 +47,8 @@ struct IntPointer {
   const Descriptor *Desc;
   uint64_t Value;
 
-  IntPointer atOffset(const ASTContext &ASTCtx, unsigned Offset) const;
+  std::optional<IntPointer> atOffset(const ASTContext &ASTCtx,
+                                     unsigned Offset) const;
   IntPointer baseCast(const ASTContext &ASTCtx, unsigned BaseOffset) const;
 };