%2 = "tf.XlaCallModule"(%arg0, %1, %0) <{Sout = [#tf_type.shape<?x2>], module = "", version = 9 : i64}> {_entry_function = @composite_dot_general_with_bias_and_relu6_dynamic_fn_2, _original_entry_function = "composite_dot_general_with_bias_and_relu6_dynamic_fn_2", _tfl_quant_trait = "fully_quantizable"} : (tensor<?x2xf32>, tensor<2x2xf32>, tensor<2xf32>) -> tensor<?x2xf32>

    %cst = "tf.Const"() {device = "", value = dense<1.000000e+01> : tensor<512x512xf32>} : () -> tensor<512x512xf32>
    %cst_sharded = "tf.XlaSharding"(%cst) {_XlaSharding = "\08\03\1A\03\01\04\02\22\08\00\04\01\05\02\06\03\070\01", device = "", sharding = "\08\03\1A\03\01\04\02\22\08\00\04\01\05\02\06\03\070\01", unspecified_dims = []} : (tensor<512x512xf32>) -> tensor<512x512xf32>

    %6 = "tf.Mul"(%arg2, %arg4) : (tensor<f32>, tensor<512xf32>) -> tensor<512xf32>
    %7 = "tf.Div"(%6, %arg6) : (tensor<512xf32>, tensor<f32>) -> tensor<512xf32>
    %8 = "tf.Cast"(%5) {Truncate = false} : (tensor<1x2240x1120x512xi32>) -> tensor<1x2240x1120x512xf32>
    %9 = "tf.Mul"(%7, %8) : (tensor<512xf32>, tensor<1x2240x1120x512xf32>) -> tensor<1x2240x1120x512xf32>

  }

  // CHECK-LABEL: pack
  func.func @pack(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>) -> tensor<2x1xf32> {
    %0 = "tfl.pack"(%arg0, %arg1) {axis = 0 : i32, values_count = 2 : i32} : (tensor<1xf32>, tensor<1xf32>) -> tensor<2x1xf32>
    func.return %0 : tensor<2x1xf32>
    // CHECK: %[[VAL_0:.*]] = arith.constant dense<[2, 1]> : tensor<2xi32>

func.func @constant_f32() -> tensor<2x2xf32> {
  // CHECK-NOT: f32
  // CHECK{LITERAL}: stablehlo.constant dense<[[1.398440e+00, 0.000000e+00], [3.093750e+00, -2.001950e-01]]> : tensor<2x2xbf16>
  %0 = stablehlo.constant dense<[[1.4, 0.0], [3.1, -0.2]]> : tensor<2x2xf32>
  return %0 : tensor<2x2xf32>
}

// -----

func.func @constant_elided() -> tensor<2x2xf32> {

  %2 = "tf.Reshape"(%0, %cst_0) : (tensor<1x2xf32>, tensor<2xi64>) -> tensor<2x1xf32>
  func.return %1, %2 : tensor<2x1xf32>, tensor<2x1xf32>

// CHECK:  %cst = arith.constant
// CHECK:  %[[FQ:.*]] = "tf.FakeQuantWithMinMaxVars"(%arg0, %arg1, %arg2)
// CHECK:  %[[R1:.*]] = "tf.Reshape"(%[[FQ]], %cst)
// CHECK-SAME: tensor<2x1xf32>
// CHECK:  %[[R2:.*]] = "tf.Reshape"(%[[FQ]], %cst)

  %2 = "tf.Const"() {value = dense<0.0> : tensor<2x2xf32>} : () -> tensor<2x2xf32>
  %3 = "tf.Const"() {value = dense<0.0> : tensor<4x4xf32>} : () -> tensor<4x4xf32>
  "tf.SendTPUEmbeddingGradients"(%2, %3) {_tpu_embedding_layer = "call1", config = "\0A\0B\0C\0D", operandSegmentSizes = array<i32: 2, 0>} : (tensor<2x2xf32>, tensor<4x4xf32>) -> ()

  %2 = "tf.Reshape"(%0, %cst_0) : (tensor<1x2xf32>, tensor<2xi64>) -> tensor<2x1xf32>
  func.return %1, %2 : tensor<2x1xf32>, tensor<2x1xf32>

// CHECK:  %cst = arith.constant
// CHECK:  %[[FQ:.*]] = "tf.FakeQuantWithMinMaxVars"(%arg0, %arg1, %arg2)
// CHECK:  %[[R1:.*]] = "tf.Reshape"(%[[FQ]], %cst)
// CHECK-SAME: tensor<2x1xf32>
// CHECK:  %[[R2:.*]] = "tf.Reshape"(%[[FQ]], %cst)

      %2 = "tf.XlaCallModule"(%arg0, %1, %0) <{Sout = [#tf_type.shape<?x2>], module = "", version = 9 : i64}> {_entry_function = @composite_dot_general_fn_1, _original_entry_function = "composite_dot_general_fn_1", _tfl_quant_trait = "fully_quantizable", _quantization_method = "weight_only_ptq { }"} : (tensor<?x2xf32>, tensor<2x2xf32>, tensor<2xf32>) -> tensor<?x2xf32>
      return %2 : tensor<?x2xf32>

  %11 = "tf.Floor"(%10#0) {device = ""} : (tensor<2x16xf32>) -> tensor<2x16xf32>
  %12 = "tf.Maximum"(%0, %11) {device = ""} : (tensor<f32>, tensor<2x16xf32>) -> tensor<2x16xf32>
  %13 = "tf.Minimum"(%12, %4) {device = ""} : (tensor<2x16xf32>, tensor<f32>) -> tensor<2x16xf32>
  %14 = "tf.Cast"(%13) {Truncate = false, device = ""} : (tensor<2x16xf32>) -> tensor<2x16xi32>