[clang] [X86][bytecode] Allow SSE/AVX BLEND imm intrinsics to be used in constexpr (PR #157776)

[Simon Pilgrim via cfe-commits]

================
@@ -2831,6 +2831,44 @@ static bool interp__builtin_select(InterpState &S, CodePtr OpPC,
   return true;
 }
 
+static bool interp__builtin_blend(InterpState &S, CodePtr OpPC,
+                                  const CallExpr *Call, unsigned BuiltinID) {
+  PrimType MaskT = *S.getContext().classify(Call->getArg(2));
+  APSInt Mask = popToAPSInt(S.Stk, MaskT);
+  const Pointer &TrueElem = S.Stk.pop<Pointer>();
+  const Pointer &FalseElem = S.Stk.pop<Pointer>();
+  const Pointer &Dst = S.Stk.peek<Pointer>();
+
+  assert(FalseElem.getNumElems() == TrueElem.getNumElems());
+  assert(FalseElem.getNumElems() == Dst.getNumElems());
+  unsigned NumElems = FalseElem.getNumElems();
+  PrimType ElemT = FalseElem.getFieldDesc()->getPrimType();
+  PrimType DstElemT = Dst.getFieldDesc()->getPrimType();
+
+  auto BitIndex = BuiltinID == X86::BI__builtin_ia32_pblendw256
+                      ? [](unsigned I) { return I % 8; }
+                      : [](unsigned I) { return I; };
----------------
RKSimon wrote:

You can use the `I % 8` in all cases

https://github.com/llvm/llvm-project/pull/157776