func.func @batchNormInference_2D_inner_features(
    %x: tensor<4x256xf32>, %scale: tensor<256xf32>, %offset: tensor<256xf32>,
    %mean: tensor<256xf32>, %variance: tensor<256xf32>)
    -> (tensor<4x256xf32>) {
  // CHECK-DAG: %[[EPS_BCAST:.+]] = mhlo.constant dense<1.001000e-05> : tensor<256xf32>
  // CHECK-DAG: %[[VARIANCE_EPS:.+]] = mhlo.add %[[VARIANCE]], %[[EPS_BCAST]] : tensor<256xf32>

func.func @unfuse_batch_norm(
    %x: tensor<4x256xf32>, %scale: tensor<256xf32>, %offset: tensor<256xf32>,
    %mean: tensor<256xf32>, %variance: tensor<256xf32>)
    -> (tensor<4x256xf32>) {
  // CHECK-DAG: %[[EPS_BCAST:.+]] = mhlo.constant dense<1.001000e-05> : tensor<256xf32>
  // CHECK-DAG: %[[VARIANCE_EPS:.+]] = mhlo.add %[[VARIANCE]], %[[EPS_BCAST]] : tensor<256xf32>

  %2 = "tfl.batch_matmul"(%1, %arg2) {adj_x = true, adj_y = false, asymmetric_quantize_inputs = false} : (tensor<4x8xf32>, tensor<4x256xf32>) -> tensor<8x256xf32>
  func.return %2 : tensor<8x256xf32>
  // CHECK: return %[[RES1]] : tensor<8x256xf32>
}

// CHECK-LABEL: Batchmatmul2Fullyconnected
// CHECK-NOT: "tfl.batch_matmul"

    %12 = "tfl.logistic"(%11) : (tensor<4x2xf32>) -> tensor<4x2xf32>
    %13 = tfl.mul %arg4, %12 {fused_activation_function = "NONE"} : tensor<4x2xf32>
    %14 = "tfl.relu"(%10#1) : (tensor<4x2xf32>) -> tensor<4x2xf32>
    %15 = "tfl.logistic"(%10#0) : (tensor<4x2xf32>) -> tensor<4x2xf32>
    %16 = tfl.mul %15, %14 {fused_activation_function = "NONE"} : tensor<4x2xf32>

       {
         data_format = "NCHW",
         epsilon = 1.001000e-05 : f32
       } : (tensor<?x256x56x56xf32>, tensor<256xf32>, tensor<256xf32>, tensor<256xf32>, tensor<256xf32>)
        -> (tensor<?x256x56x56xf32>, tensor<256xf32>, tensor<256xf32>, tensor<256xf32>, tensor<256xf32>, tensor<*xf32>)

  // CHECK: %[[BATCH_NORM1:[_a-z0-9]*]], {{.*}} = "tf.FusedBatchNormV3"
  // CHECK-SAME: %[[CONV1]]

// CHECK-SAME:                  %[[VAL_0:.*]]: tensor<4x6xf32>) -> tensor<2x2x6xf32> {
// CHECK:           %[[VAL_1:.*]] = arith.constant dense<[2, 2, 6]> : tensor<3xi64>
// CHECK:           %[[VAL_2:.*]] = "tf.Reshape"(%[[VAL_0]], %[[VAL_1]]) : (tensor<4x6xf32>, tensor<3xi64>) -> tensor<2x2x6xf32>
// CHECK:           return %[[VAL_2]] : tensor<2x2x6xf32>
// CHECK:         }
func.func @reshape(%arg0: tensor<4x6xf32>) -> tensor<2x2x6xf32> {

}

func.func @einsum_no_match(%arg0: tensor<4x5x6xf32>, %arg1: tensor<5xf32>) -> tensor<4xf32> {
  %0 = "tf.Einsum"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", equation = "ijk,j->i"}: (tensor<4x5x6xf32>, tensor<5xf32>) -> tensor<4xf32>
  func.return %0 : tensor<4xf32>
// CHECK-LABEL: einsum_no_match
// CHECK: %[[v0:.*]] = "tf.Einsum"(%arg0, %arg1) <{equation = "ijk,j->i"}> {T = "tfdtype$DT_FLOAT"} : (tensor<4x5x6xf32>, tensor<5xf32>) -> tensor<4xf32>

// CHECK-NEXT:      %18 = "tfl.batch_matmul"(%8, %17) <{adj_x = false, adj_y = false, asymmetric_quantize_inputs = false}> : (tensor<4x4x?xf32>, tensor<4x?x256xf32>) -> tensor<4x4x256xf32>
// CHECK-NEXT:      %19 = mhlo.reshape %18 : (tensor<4x4x256xf32>) -> tensor<4x4x256xf32>
// CHECK-NEXT:      return %19 : tensor<4x4x256xf32>
}

func.func @eliminateExtraSelectLhs(%arg0: tensor<4x2x1xf32>, %arg1: tensor<4x2x1xi1>) -> (tensor<4x2x1xf32>) {
  %cst0 = arith.constant dense<1.0> : tensor<2x2xf32>
  %cst1 = arith.constant dense<2.0> : tensor<2xf32>
  %cst_zeros = arith.constant dense<0.0> : tensor<4x2x1xf32>

  %0 = "tfl.select_v2"(%arg1, %cst_zeros, %arg0) : (tensor<4x2x1xi1>, tensor<4x2x1xf32>, tensor<4x2x1xf32>) -> tensor<4x2x1xf32>

func.func @main(tensor<4x5xbf16>) -> tensor<4x5xbf16> {
^bb0(%arg0: tensor<4x5xbf16>):
  // CHECK-LABEL: @main
  // CHECK:  (tensor<4x5xbf16>) -> tensor<4x5xf32>
  // CHECK-NEXT:  (tensor<4x5xf32>) -> tensor<4x5xbf16>
  %0 = "tfl.cast" (%arg0) : (tensor<4x5xbf16>) -> tensor<4x5xf32> loc("cast1")
  %1 = "tfl.cast" (%0) : (tensor<4x5xf32>) -> tensor<4x5xbf16> loc("cast2")
  func.return %1 : tensor<4x5xbf16>