func.func @tile_by_reshape(%arg0: tensor<4x8xf32>) -> tensor<28x24xf32> {
  // CHECK: %[[BROADCASTED:.*]] = "mhlo.broadcast_in_dim"(%arg0) <{broadcast_dimensions = dense<[1, 3]> : tensor<2xi64>}> : (tensor<4x8xf32>) -> tensor<7x4x3x8xf32>
  // CHECK: %[[RESULT:.*]] = mhlo.reshape %[[BROADCASTED]] : (tensor<7x4x3x8xf32>) -> tensor<28x24xf32>
  // CHECK: return %[[RESULT]] : tensor<28x24xf32>

    : ( tensor<1x28x28xf32>,
        tensor<20x28xf32>, tensor<20x28xf32>, tensor<20x28xf32>, tensor<20x28xf32>,
        tensor<20x20xf32>, tensor<20x20xf32>, tensor<20x20xf32>, tensor<20x20xf32>,
        none, none, none,
        tensor<20xf32>, tensor<20xf32>, tensor<20xf32>, tensor<20xf32>,
        none, none,

    return %0 : tensor<2x2xf32>
  }
  func.func private @composite_matmul_fn_1(%arg0: tensor<2x2xf32>, %arg1: tensor<2x2xf32>) -> tensor<2x2xf32> attributes {tf_quant.composite_function} {
    %0 = "tf.MatMul"(%arg0, %arg1) {attr_map = "0:transpose_a,1:transpose_b", device = "", transpose_a = false, transpose_b = false} : (tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>
    return %0 : tensor<2x2xf32>
  }

    %0 = "tf.MatMul"(%arg0, %arg1) {attr_map = "0:transpose_a,1:transpose_b", device = "", transpose_a = false, transpose_b = false} : (tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>
    return %0 : tensor<2x2xf32>
  }

    %0 = "tf.Div"(%arg1, %cst_0) : (tensor<2x2xf32>, tensor<f32>) -> tensor<2x2xf32>
    %1 = "tf.AddV2"(%0, %cst_1) : (tensor<2x2xf32>, tensor<f32>) -> tensor<2x2xf32>
    %2 = "tf.Floor"(%1) : (tensor<2x2xf32>) -> tensor<2x2xf32>
    %3 = "tf.ClipByValue"(%2, %cst_5, %cst_6) : (tensor<2x2xf32>, tensor<f32>, tensor<f32>) -> tensor<2x2xf32>
    %4 = "tf.Cast"(%3) {Truncate = false} : (tensor<2x2xf32>) -> tensor<2x2xi8>

  %0 = "tf.Div"(%arg0, %cst) : (tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>
  func.return %0 : tensor<2x2xf32>

  // CHECK-LABEL: RemoveTrivialDiv
  // CHECK-NEXT: return %arg0 : tensor<2x2xf32>
}

func.func @RemoveTrivialRealDiv(%arg0: tensor<2x2xf32>, %arg1: tensor<2x2xf32>) -> tensor<2x2xf32> {
  %cst = arith.constant dense<1.0> : tensor<2x2xf32>

  %4 = "tfl.concatenation"(%1, %3) {axis = -1 : i32, fused_activation_function = "NONE"} : (tensor<2x1xf32>, tensor<2x3xf32>) -> tensor<2x4xf32>
  %5 = "tfl.add"(%4, %arg2) {fused_activation_function = "NONE"} : (tensor<2x4xf32>, tensor<2x4xf32>) -> tensor<2x4xf32>
  func.return %5: tensor<2x4xf32>

// MixedPrecision-NEXT: %[[q:.*]] = "tfl.quantize"(%arg0)
// MixedPrecision-NEXT: %[[dq:.*]] = "tfl.dequantize"(%[[q]])

  // CHECK:  %cst = arith.constant dense<2> : tensor<1xi32>
  // CHECK:  %3 = "tfl.reduce_max"(%arg0, %cst) <{keep_dims = false}> : (tensor<2x2x4xf32>, tensor<1xi32>) -> tensor<2x2xf32>
  // CHECK:  %4 = "tfl.arg_max"(%arg0, %cst) : (tensor<2x2x4xf32>, tensor<1xi32>) -> tensor<2x2xi32>
  // CHECK:  return %3, %4 : tensor<2x2xf32>, tensor<2x2xi32>
}

// CHECK-LABEL:   func @convert_argmax_constant_non_z_axis

}
func.func @_func(%arg0: tensor<2x4xf32>, %arg1: tensor<4x2xf32>) -> tensor<2x2xf32> {
  %0 = "tf.MatMul"(%arg0, %arg1) {_XlaSharding = "\08\03\1A\02\02\01\22\02\00\01"} : (tensor<2x4xf32>, tensor<4x2xf32>) -> tensor<2x2xf32>
  %1 = "tf.Identity"(%0) : (tensor<2x2xf32>) -> tensor<2x2xf32>
  return %1 : tensor<2x2xf32>
}

// -----
// The following op sharding is used in the following test case:

func.func @ReorderAddWithConstant(%arg0: tensor<2x2xf32>) -> tensor<2x2xf32> {
  %cst = arith.constant dense<1.0> : tensor<2x2xf32>
  %cst_1 = arith.constant dense<2.0> : tensor<2x2xf32>
  %0 = "tfl.add"(%arg0, %cst) {fused_activation_function = "NONE"} : (tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>
  %1 = "tfl.add"(%0, %cst_1) {fused_activation_function = "NONE"} : (tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>