}
				v_1_1_0 := v_1_1.Args[0]
				if v_1_1_0.Op != OpRsh64Ux64 {
					continue
				}
				_ = v_1_1_0.Args[1]
				mul := v_1_1_0.Args[0]
				if mul.Op != OpMul64 {
					continue
				}
				_ = mul.Args[1]
				mul_0 := mul.Args[0]
				mul_1 := mul.Args[1]
				for _i2 := 0; _i2 <= 1; _i2, mul_0, mul_1 = _i2+1, mul_1, mul_0 {
					if mul_0.Op != OpConst64 {
						continue
					}

	v_0 := v.Args[0]
	b := v.Block
	// match: (MUL (NEG x) y)
	// result: (MNEG x y)
	for {
		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
			if v_0.Op != OpARM64NEG {
				continue
			}
			x := v_0.Args[0]
			y := v_1
			v.reset(OpARM64MNEG)
			v.AddArg2(x, y)
			return true
		}
		break
	}
	// match: (MUL x (MOVDconst [-1]))
	// result: (NEG x)
	for {

		{"LBRACK", Const, 0},
		{"LEQ", Const, 0},
		{"LOR", Const, 0},
		{"LPAREN", Const, 0},
		{"LSS", Const, 0},
		{"Lookup", Func, 0},
		{"LowestPrec", Const, 0},
		{"MAP", Const, 0},
		{"MUL", Const, 0},
		{"MUL_ASSIGN", Const, 0},
		{"NEQ", Const, 0},
		{"NOT", Const, 0},
		{"NewFileSet", Func, 0},
		{"NoPos", Const, 0},
		{"OR", Const, 0},
		{"OR_ASSIGN", Const, 0},

  }];

  let arguments = (ins
    TF_FpOrI32OrI64Tensor:$input,

    I64Attr:$group_size,
    I64Attr:$group_key,
    I64Attr:$instance_key,
    TF_AnyStrAttrOf<["Min", "Max", "Mul", "Add"]>:$merge_op,
    TF_AnyStrAttrOf<["Id", "Div"]>:$final_op,
    I64ArrayAttr:$subdiv_offsets,
    DefaultValuedOptionalAttr<I64ArrayAttr, "{}">:$wait_for,

    *   Unary GPU kernels: Abs, Atanh, Acos, Acosh, Asin, Asinh, Atan, Cos, Cosh, Sin, Sinh, Tan, Tanh.
    *   Binary GPU kernels: AddV2, Sub, Div, DivNoNan, Mul, MulNoNan, FloorDiv, Equal, NotEqual, Greater, GreaterEqual, LessEqual, Less.

* `tf.lite`
    * Add experimental supports conversion of models that may be larger than 2GB before buffer deduplication