non_feature_dims.push_back(i);
      }
      auto reduce_dims = GetI64ElementsAttr(non_feature_dims, &rewriter);
      auto scalar_broadcast_dims = rewriter.getDenseI64ArrayAttr({});

      // scratch1 = rsqrt(var + epsilon)
      RankedTensorType scalar_float =
          tensorflow::GetTypeFromTFTensorShape({}, kernel_type);
      auto epsilon = rewriter.create<ConstantOp>(

	v_0 := v.Args[0]
	// match: (FSQRT (FMOVDconst [x]))
	// cond: x >= 0
	// result: (FMOVDconst [math.Sqrt(x)])
	for {
		if v_0.Op != OpPPC64FMOVDconst {
			break
		}
		x := auxIntToFloat64(v_0.AuxInt)
		if !(x >= 0) {
			break
		}
		v.reset(OpPPC64FMOVDconst)
		v.AuxInt = float64ToAuxInt(math.Sqrt(x))
		return true
	}
	return false
}

	alias("internal/runtime/atomic", "Oruintptr", "internal/runtime/atomic", "Or64", sys.ArchARM64)

	/******** math ********/
	addF("math", "sqrt",
		func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
			return s.newValue1(ssa.OpSqrt, types.Types[types.TFLOAT64], args[0])
		},