// CHECK-NEXT:    %[[RESHAPED_BMM:.*]] = mhlo.reshape %[[BMM_0]]
// CHECK-NEXT:    return %[[RESHAPED_BMM]] : tensor<3x5x1x4xf32>
}


func.func @convert_dot_general_repeated(%arg0: tensor<1x1x1024xf32>, %arg1: tensor<1024x1024xf32>) -> tensor<1x1x1024xf32> {

    dot_dimension_numbers = #mhlo.dot<
      lhs_contracting_dimensions = [2],
      rhs_contracting_dimensions = [0]
  >} : (tensor<3x1x512xf32>, tensor<512x13xf32>) -> tensor<3x1x13xf32>
  %r = "mhlo.concatenate"(%0, %1, %2) <{dimension = 0 : i64}> : (tensor<1x1x13xf32>, tensor<2x1x13xf32>, tensor<3x1x13xf32>) -> tensor<6x1x13xf32>
  func.return %r : tensor<6x1x13xf32>

    %5 = "quantfork.qcast"(%4) {volatile} : (tensor<3x4x2xf32>) -> tensor<3x4x2x!quant.uniform<i8:f32, 0.13170163023705575:-1>>
    %6 = "quantfork.dcast"(%5) : (tensor<3x4x2x!quant.uniform<i8:f32, 0.13170163023705575:-1>>) -> tensor<3x4x2xf32>
    %7 = "stablehlo.gather"(%6, %arg2) {
    dimension_numbers = #stablehlo.gather<

// CHECK:           %[[VAL_11:.*]] = arith.constant dense<[3, 5, 1, 4]> : tensor<4xi64>
// CHECK:           %[[VAL_12:.*]] = "tf.Reshape"(%[[VAL_10]], %[[VAL_11]]) : (tensor<3x5x4xf32>, tensor<4xi64>) -> tensor<3x5x1x4xf32>
// CHECK:           return %[[VAL_12]] : tensor<3x5x1x4xf32>
// CHECK:         }

  // CHECK: %[[v0:.*]] = "tf.Reshape"(%arg1, %[[cst]]) : (tensor<5x6x2xf32>, tensor<2xi64>) -> tensor<5x12xf32>
  // CHECK: %[[v1:.*]] = "tf.BatchMatMulV2"(%arg0, %[[v0]]) <{adj_x = false, adj_y = false}> : (tensor<3x4x5xf32>, tensor<5x12xf32>) -> tensor<3x4x12xf32>
  // CHECK: %[[v2:.*]] = "tf.Reshape"(%[[v1]], %[[cst_1]]) : (tensor<3x4x12xf32>, tensor<4xi64>) -> tensor<3x4x6x2xf32>
  // CHECK: return %[[v2]] : tensor<3x4x6x2xf32>
}

  %0 = "quantfork.stats"(%arg0) {
    layerStats = dense<[0.0, 1.0]> : tensor<2xf32>
  } : (tensor<?x5x5x2xf32>) -> tensor<?x5x5x2xf32>
  %1 = "tfl.pseudo_const"() {value = dense<1.000000e+00> : tensor<3x5x5x2xf32>} : () -> tensor<3x5x5x2xf32>
  %2 = "tfl.pseudo_const"() {value = dense<0.000000e+00> : tensor<3xf32>} : () -> tensor<3xf32>
  %3 = "tfl.conv_2d"(%0, %1, %2) {
    dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32,

func.func @broadcast_mul_implicit_no_fold(%arg0: tensor<5x7xf32>, %arg1: tensor<5xf32>) -> tensor<3x5x7xf32> {
  %cst = arith.constant dense<[3, 5, 7]> : tensor<3xi32>
  %0 = "tf.BroadcastTo"(%arg1, %cst) : (tensor<5xf32>, tensor<3xi32>) -> tensor<3x5x7xf32>
  %1 = "tf.Mul"(%arg0, %0) : (tensor<5x7xf32>, tensor<3x5x7xf32>) -> tensor<3x5x7xf32>
  func.return %1 : tensor<3x5x7xf32>
  // CHECK: %[[C0:.*]] = arith.constant dense<[3, 5, 7]> : tensor<3xi32>

func.func @mean_4d_no_keepdim(%arg0: tensor<1x48x48x512xf32>) -> tensor<1x512xf32> {
  %cst = arith.constant dense<[1, 2]> : tensor<2xi32>
  %0 = "tfl.mean"(%arg0, %cst) {keep_dims = false} : (tensor<1x48x48x512xf32>, tensor<2xi32>) -> tensor<1x512xf32>
  func.return %0 : tensor<1x512xf32>
}

// CHECK:       func @mean_4d_no_keepdim([[VAL_0:%.*]]: tensor<1x48x48x512xf32>) -> tensor<1x512xf32> {

  %flow_1_out = "tf.TensorArrayWriteV3"(%handle_1, %index, %y, %flow_1) {device = "/job:localhost/replica:0/task:0/device:CPU:0"} : (tensor<2x!tf_type.resource<tensor<?x512xf32>>>, tensor<i32>, tensor<?x512xf32>, tensor<f32>) -> tensor<f32>

  %cst_input = arith.constant dense<1.0> : tensor<2x512xf32>
  %cst_weights = arith.constant dense<2.0> : tensor<1024x512xf32>
  %cst_bias = arith.constant dense<4.0> : tensor<1024xf32>

  %0 = "tfl.fully_connected" (%cst_input, %cst_weights, %cst_bias) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<2x512xf32>, tensor<1024x512xf32>, tensor<1024xf32>) -> tensor<2x1024xf32>