// CHECK: %[[LHS_SPLIT:.*]]:6 = "tf.Split"(%[[SPLITTING_AXIS]], %[[LHS_RESHAPED]]) : (tensor<i32>, tensor<6x4x5xf32>) -> (tensor<1x4x5xf32>, tensor<1x4x5xf32>, tensor<1x4x5xf32>, tensor<1x4x5xf32>, tensor<1x4x5xf32>, tensor<1x4x5xf32>)
  // CHECK: %[[LHS_1:.*]] = "tf.Reshape"(%[[LHS_SPLIT]]#0, %[[MATMUL_LHS_SHAPE]]) : (tensor<1x4x5xf32>, tensor<2xi64>) -> tensor<4x5xf32>

// CHECK: %0 = tfl.add(%arg0, %cst_0) <{fused_activation_function = "NONE"}> : (tensor<2x3x4x5xf32>, tensor<2x3x4x1xf32>) -> tensor<2x3x4x5xf32>
// CHECK: %1 = "tfl.transpose"(%0, %cst) : (tensor<2x3x4x5xf32>, tensor<4xi32>) -> tensor<5x2x3x4xf32>

// -----

// CHECK-LABEL: pushTposeBcastScalarCstInput
func.func @pushTposeBcastScalarCstInput(%arg0: tensor<2x3x4x5xf32>) -> tensor<5x2x3x4xf32> {

    %0 = "tf.Div"(%arg0, %cst_3) {device = ""} : (tensor<2x6x4x5xf32>, tensor<f32>) -> tensor<2x6x4x5xf32>
    %1 = "tf.AddV2"(%0, %cst_1) {device = ""} : (tensor<2x6x4x5xf32>, tensor<f32>) -> tensor<2x6x4x5xf32>
    %2 = "tf.Maximum"(%1, %cst_1) {device = ""} : (tensor<2x6x4x5xf32>, tensor<f32>) -> tensor<2x6x4x5xf32>
    %3 = "tf.Minimum"(%2, %cst_4) {device = ""} : (tensor<2x6x4x5xf32>, tensor<f32>) -> tensor<2x6x4x5xf32>

  return %0 : tensor<1x1x1x5xf32>
}
func.func private @XlaCallModule_aten.avg_pool2d.default.impl_6(%arg0: tensor<1x1x1x8xf32>) -> tensor<1x1x1x5xf32>

// CHECK-LABEL: avg_pool2d_7
// CHECK: %cst = arith.constant dense<[0, 2, 3, 1]> : tensor<4xi32>

// -----

// CHECK-LABEL: transpose_simple_4d
func.func @transpose_simple_4d() -> tensor<5x2x3x4xf32> {
  %0 = stablehlo.constant dense<1.000000e+0> : tensor<2x3x4x5xf32>
  %1 = stablehlo.transpose %0, dims = [3, 0, 1, 2] : (tensor<2x3x4x5xf32>) -> tensor<5x2x3x4xf32>
  return %1 : tensor<5x2x3x4xf32>
}
// CHECK-DAG: %[[CONST_0:.+]] = stablehlo.constant dense<1.000000e+00> : tensor<5x2x3x4xf32>
// CHECK-NOT: transpose

// CHECK: return %[[v0]]
}

func.func @batch_multilhs_einsum(%arg0: tensor<2x1x1x11xf32>, %arg1: tensor<2x11x2xf32>) -> tensor<2x1x1x2xf32> {
  %0 = "tf.Einsum"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", equation = "BiNj,BjS->BiNS"} : (tensor<2x1x1x11xf32>, tensor<2x11x2xf32>) -> tensor<2x1x1x2xf32>
  func.return %0 : tensor<2x1x1x2xf32>
// CHECK-LABEL: batch_multilhs_einsum
// CHECK-DAG: %[[cst:.*]] = arith.constant dense<[2, 1, 11]> : tensor<3xi64>

func.func @ConvertSliceToIdentityI32(%arg0: tensor<2x3x4x5xf32>) -> tensor<2x3x4x5xf32> {
  %begin = arith.constant dense<0> : tensor<4xi32>
  %shape = arith.constant dense<[2,3,4,5]> : tensor<4xi32>
  %0 = "tfl.slice"(%arg0, %begin, %shape) : (tensor<2x3x4x5xf32>, tensor<4xi32>, tensor<4xi32>) -> tensor<2x3x4x5xf32>
  func.return %0 : tensor<2x3x4x5xf32>
  // CHECK: return %arg0
}

  return %2 : tensor<?x2x2x1xi8>
}

// -----

// CHECK-LABEL: func @convolution_add_add_f32
func.func @convolution_add_add_f32(
    %lhs: tensor<?x3x2x1xf32>, %rhs: tensor<2x1x1x1xf32>,
    %zp_offset: tensor<?x2x2x1xf32>, %bias: tensor<1xf32>
  ) -> tensor<?x2x2x1xf32> {
  // CHECK-DAG: %[[conv:.*]] = mhlo.convolution
  // CHECK-DAG: %[[combined:.*]] = chlo.broadcast_add %[[conv:.*]], %[[zp_offset:.*]]

  %2 = "tf.Conv2D"(%1, %arg1) {padding = "VALID", strides = [1, 1, 1, 1]} : (tensor<2x1x68x3xf32>, tensor<1x5x3x8xf32>) -> tensor<2x1x64x8xf32>
  %3 = "tf.Squeeze"(%2) {squeeze_dims = [-3]} : (tensor<2x1x64x8xf32>) -> tensor<2x64x8xf32>
  %4 = "tf.BatchToSpaceND"(%3, %cst_1, %cst) : (tensor<2x64x8xf32>, tensor<1xi32>, tensor<1x2xi32>) -> tensor<1x128x8xf32>
  func.return %4 : tensor<1x128x8xf32>

  %cst_1 = arith.constant dense<6.000000e+00> : tensor<2x1x3x3xf32>
  %0 = "tfl.quantize"(%cst_1) {qtype = tensor<2x1x3x3x!quant.uniform<i8<-127:127>:f32:0, {6.587140e-03,1.888450e-02}>>} : (tensor<2x1x3x3xf32>) -> tensor<2x1x3x3x!quant.uniform<i8<-127:127>:f32:0, {6.587140e-03,1.888450e-02}>>
  %1 = "tfl.dequantize"(%0) : (tensor<2x1x3x3x!quant.uniform<i8<-127:127>:f32:0, {6.587140e-03,1.888450e-02}>>) -> tensor<2x1x3x3xf32>