// CHECK: %[[VAR_VAL:.*]] = "tfl.read_variable"(%[[RESOURCE]]) : (tensor<!tf_type.resource<tensor<1x10xf32>>>) -> tensor<1x10xf32>
  // CHECK: %[[ADD:.*]] = tfl.add %[[VAR_VAL]], %arg0 {fused_activation_function = "NONE"} : tensor<1x10xf32>
  // CHECK: "tfl.assign_variable"(%[[RESOURCE]], %[[ADD]]) : (tensor<!tf_type.resource<tensor<1x10xf32>>>, tensor<1x10xf32>) -> ()

// CHECK-LABEL: QuantizeReadVariable
func.func @QuantizeReadVariable() -> (tensor<1x2x1x3x!quant.uniform<i8:f32, 1.0>>) {
  %1 = "tfl.var_handle"() : () -> tensor<!tf_type.resource>
  %2 = "tfl.read_variable"(%1) : (tensor<!tf_type.resource>) -> tensor<1x2x1x3xf32>
  %3 = "tfl.quantize"(%2) {qtype = tensor<1x2x1x3x!quant.uniform<i8:f32, 1.0>>, volatile} : (tensor<1x2x1x3xf32>) -> tensor<1x2x1x3x!quant.uniform<i8:f32, 1.0>>

// CHECK-SAME: {{.*shared_name = "var_1".*}}

// ReadVariableOps are inserted for each VarHandleOp to read the tensor values.
// CHECK-DAG: %[[READ_VARIABLE_0:.*]] = "tf.ReadVariableOp"(%[[VAR_HANDLE_0]]) : (tensor<!tf_type.resource<tensor<2xf32>>>) -> tensor<2xf32>
// CHECK-DAG: %[[READ_VARIABLE_1:.*]] = "tf.ReadVariableOp"(%[[VAR_HANDLE_1]]) : (tensor<!tf_type.resource<tensor<3xf32>>>) -> tensor<3xf32>

// CHECK: tf_executor.graph
// CHECK: %[[VAR_HANDLE:.*]], {{.*}} = tf_executor.island wraps "tf.VarHandleOp"() <{{{.*shared_name = "var_0".*}}}>
// CHECK: %[[READ_VARIABLE:.*]], {{.*}} = tf_executor.island wraps "tf.ReadVariableOp"(%[[VAR_HANDLE]])
// CHECK-DAG: %[[CST_0:.*]], {{.*}} = tf_executor.island wraps "tf.Const"() <{{{.*value = dense<"var_0"> : tensor<1x!tf_type\.string>.*}}}>

  %0 = "tfl.var_handle"() {container = "", shared_name = "states"} : () -> tensor<!tf_type.resource<tensor<1x2x3xf32>>>
  %1 = "tfl.read_variable"(%0) : (tensor<!tf_type.resource<tensor<1x2x3xf32>>>) -> tensor<1x2x3xf32>

  REAL=134,
  COMPLEX_ABS=135,
  HASHTABLE = 136,
  HASHTABLE_FIND = 137,
  HASHTABLE_IMPORT = 138,
  HASHTABLE_SIZE = 139,
  REDUCE_ALL = 140,
  CONV_3D_TRANSPOSE = 141,
  VAR_HANDLE = 142,
  READ_VARIABLE = 143,
  ASSIGN_VARIABLE = 144,
}
// LINT.ThenChange(nnapi_linter/linter.proto)

// Options for the builtin operators.
union BuiltinOptions {
  Conv2DOptions,
  DepthwiseConv2DOptions,

  REAL=134,
  COMPLEX_ABS=135,
  HASHTABLE = 136,
  HASHTABLE_FIND = 137,
  HASHTABLE_IMPORT = 138,
  HASHTABLE_SIZE = 139,
  REDUCE_ALL = 140,
  CONV_3D_TRANSPOSE = 141,
  VAR_HANDLE = 142,
  READ_VARIABLE = 143,
  ASSIGN_VARIABLE = 144,
  BROADCAST_ARGS = 145,
  RANDOM_STANDARD_NORMAL = 146,
  BUCKETIZE = 147,
  RANDOM_UNIFORM = 148,
  MULTINOMIAL = 149,
  GELU = 150,
  DYNAMIC_UPDATE_SLICE = 151,

// RUN: tf-mlir-translate -mlir-to-graphdef %s -o - | FileCheck %s

// Verify that attribute T is not set as ref type.
// CHECK-LABEL: name: "Ref_Variable"
// CHECK-LABEL: name: "foo"
// CHECK:      attr {
// CHECK-NEXT:   key: "T"
// CHECK-NEXT:   value {
// CHECK-NEXT:     type: DT_INT32{{$}}
// CHECK-NEXT:   }
// CHECK-NEXT: }

func.func @main() {
  tf_executor.graph {

                         "Non scalar type">;

def HasSingleElement: Constraint<CPred<"($0.size() == 1)">>;

def EmptyList: NativeCodeCall<"llvm::SmallVector<mlir::Value>{}">;

// Creates a tf.ReadVariable op that reads a resource `$2` that has the same
// element type as `$1`. The op created will use location of `$0`.
def CreateTFReadVariableOp : NativeCodeCall<
    "$_builder.create<TF::ReadVariableOp>("
    "  $0.getLoc(),"

  %val1 = "tf.Identity"(%val0) {_class = ["loc:@v"]} : (tensor<96x!tf_type.f32ref>) -> tensor<96xf32>
  func.return
}

// -----

// Test case: Two ReadVariable ops.

func.func @f() {
  // CHECK: "tf.VarHandleOp"

  // During lowering to resource variables, this pass will preserve the
  // locations of the ReadVariableOps as Identity ops to keep the original graph