[clang] [CIR][AArch64] Upstream addition NEON builtins (PR #202005)

[Andrzej Warzyński via cfe-commits]

================
@@ -664,12 +664,20 @@ static mlir::Value emitCommonNeonBuiltinExpr(
     ops[0] = cgf.getBuilder().createBitcast(loc, ops[0], vTy);
     return emitNeonSplat(cgf.getBuilder(), loc, ops[0], ops[1], numElements);
   }
+  case NEON::BI__builtin_neon_vadd_v:
+  case NEON::BI__builtin_neon_vaddq_v: {
+    unsigned numBytes = (builtinID == NEON::BI__builtin_neon_vaddq_v) ? 16 : 8;
+    cir::VectorType byteTy =
+        cir::VectorType::get(cgf.getBuilder().getUInt8Ty(), numBytes);
+    ops[0] = cgf.getBuilder().createBitcast(ops[0], byteTy);
+    ops[1] = cgf.getBuilder().createBitcast(ops[1], byteTy);
----------------
banach-space wrote:

Given that there are no bit-casts in tests, do you reckon that we could skip this bit-casting? I have similar question for the ARM.cpp implementation :) 

https://github.com/llvm/llvm-project/pull/202005