[Headers][X86] Allow AVX512 masked arithmetic intrinsics to be used in constexpr

Author: woruyu

clang/include/clang/Basic/BuiltinsX86.td

@@ -4104,7 +4104,7 @@ let Features = "avx512bf16", Attributes = [NoThrow, Const, RequiredVectorWidth<1
   def selectsbf_128 : X86Builtin<"_Vector<8, __bf16>(unsigned char, _Vector<8, __bf16>, _Vector<8, __bf16>)">;
 }
 
-let Features = "avx512f", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
+let Features = "avx512f", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in {
   def selectss_128 : X86Builtin<"_Vector<4, float>(unsigned char, _Vector<4, float>, _Vector<4, float>)">;
   def selectsd_128 : X86Builtin<"_Vector<2, double>(unsigned char, _Vector<2, double>, _Vector<2, double>)">;
 }

clang/lib/AST/ByteCode/InterpBuiltin.cpp

@@ -3598,6 +3598,27 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
       return APInt::getAllOnes(DstBits);
     });
 
+  case clang::X86::BI__builtin_ia32_selectss_128:
+  case clang::X86::BI__builtin_ia32_selectsd_128: {
+    const unsigned N =
+        Call->getArg(1)->getType()->getAs<VectorType>()->getNumElements();
+
+    const Pointer &W = S.Stk.pop<Pointer>();
+    const Pointer &A = S.Stk.pop<Pointer>();
+    APSInt U = popToAPSInt(S, Call->getArg(0));
+    const Pointer &Dst = S.Stk.peek<Pointer>();
+
+    const bool TakeA0 = U.getZExtValue() & 1ULL;
+
+    for (unsigned i = 0; i < N; ++i)
+      Dst.elem<Floating>(i) = W.elem<Floating>(i);
+    if (TakeA0)
+      Dst.elem<Floating>(0) = A.elem<Floating>(0);
+
+    Dst.initializeAllElements();
+    return true;
+  }
+
   case clang::X86::BI__builtin_ia32_vprotbi:
   case clang::X86::BI__builtin_ia32_vprotdi:
   case clang::X86::BI__builtin_ia32_vprotqi:

clang/lib/AST/ExprConstant.cpp

@@ -11705,6 +11705,23 @@ bool VectorExprEvaluator::VisitCallExpr(const CallExpr *E) {
         return Success(APValue(ResultElements.data(), SourceLen), E);
       };
 
+  auto EvalSelectScalar = [&](unsigned Len) -> std::optional<APValue> {
+    APSInt Mask;
+    APValue AVal, WVal;
+    if (!EvaluateInteger(E->getArg(0), Mask, Info) ||
+        !EvaluateAsRValue(Info, E->getArg(1), AVal) ||
+        !EvaluateAsRValue(Info, E->getArg(2), WVal))
+      return std::nullopt;
+
+    const bool TakeA0 = (Mask.getZExtValue() & 1u) != 0;
+    SmallVector<APValue, 4> Res;
+    Res.reserve(Len);
+    Res.push_back(TakeA0 ? AVal.getVectorElt(0) : WVal.getVectorElt(0));
+    for (unsigned i = 1; i < Len; ++i)
+      Res.push_back(WVal.getVectorElt(i));
+    return APValue(Res.data(), Res.size());
+  };
+
   switch (E->getBuiltinCallee()) {
   default:
     return false;
@@ -11933,6 +11950,16 @@ bool VectorExprEvaluator::VisitCallExpr(const CallExpr *E) {
         return APInt((Src).trunc(DstBits));
       return APInt::getAllOnes(DstBits);
     });
+  case clang::X86::BI__builtin_ia32_selectss_128: {
+    if (auto V = EvalSelectScalar(4))
+      return Success(*V, E);
+    return false;
+  }
+  case clang::X86::BI__builtin_ia32_selectsd_128: {
+    if (auto V = EvalSelectScalar(2))
+      return Success(*V, E);
+    return false;
+  }
   case clang::X86::BI__builtin_ia32_pmuldq128:
   case clang::X86::BI__builtin_ia32_pmuldq256:
   case clang::X86::BI__builtin_ia32_pmuldq512: