[clang] [X86][bytecode] Allow SSE/AVX BLENDVPD/PD intrinsics to be used in constexpr (PR #157126)

[Timm Baeder via cfe-commits]

================
@@ -3421,6 +3421,17 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
           return F;
         });
 
+  case clang::X86::BI__builtin_ia32_blendvpd:
+  case clang::X86::BI__builtin_ia32_blendvpd256:
+  case clang::X86::BI__builtin_ia32_blendvps:
+  case clang::X86::BI__builtin_ia32_blendvps256:
+    return interp__builtin_elementwise_triop_fp(
+        S, OpPC, Call,
+        [](const APFloat &F, const APFloat &T, const APFloat &C,
+           llvm::RoundingMode) {
+          return C.bitcastToAPInt().isNegative() ? T : F;
----------------
tbaederr wrote:

Is the bitcast here really needed?

https://github.com/llvm/llvm-project/pull/157126