func.return %0 : tensor<f32>
}

// CHECK-LABEL: func @xla_all_reduce_mul
func.func @xla_all_reduce_mul(%input: tensor<f32>) -> tensor<f32> {
  %group_assignment = "tf.Const"() { value = dense<[[0],[1]]> : tensor<2x1xi32> } : () -> tensor<2x1xi32>
  // CHECK: "mhlo.all_reduce"
  // CHECK: mhlo.mul

  %cst_perm = arith.constant dense<[1, 0]> : tensor<2xi32>

  // CHECK: %[[CST:.*]] = arith.constant dense<{{\[\[}}0, 2], {{\[}}1, 3]]> : tensor<2x2xi32>
  // CHECK: return %[[CST]]
  %0 = "tfl.transpose"(%cst, %cst_perm) : (tensor<2x2xi32>, tensor<2xi32>) -> tensor<2x2xi32>
  func.return %0 : tensor<2x2xi32>
}

// CHECK-LABEL: @transpose_2d_splat
func.func @transpose_2d_splat() -> tensor<3x2xi32> {

    %0 = stablehlo.convert %arg0 : (tensor<1x2xf32>) -> tensor<1x2xi32>
    return %0 : tensor<1x2xi32>
  }
// CHECK: @uniform_quantize_0
  func.func private @uniform_quantize_1(%arg0: tensor<1x3xf32>, %arg1: tensor<1x1xf32>, %arg2: tensor<1x1xi8>) -> tensor<1x3xi8> {
    %0 = stablehlo.convert %arg0 : (tensor<1x3xf32>) -> tensor<1x3xi8>

    %9 = "tf.Const"() {value = dense<-1> : tensor<1xi32>} : () -> tensor<1xi32>
    %10 = "tf.Const"() {value = dense<[[0, 0], [3, 3], [3, 3], [0, 0]]> : tensor<4x2xi32>} : () -> tensor<4x2xi32>
    %11 = "tf.Pad"(%arg0, %10) : (tensor<2x224x224x3xf32>, tensor<4x2xi32>) -> tensor<2x230x230x3xf32>
    %12 = "tf.Cast"(%arg1) {Truncate = false} : (tensor<2x1xf32>) -> tensor<2x1xi64>

  // CHECK:  %5 = "tfl.arg_max"(%arg0, %cst) : (tensor<1x32x1xf32>, tensor<1xi32>) -> tensor<1x1xi32>
  // CHECK:  return %4, %5 : tensor<1x1xf32>, tensor<1x1xi32>
}

// CHECK-LABEL:   func @convert_pytorch_argmax
func.func @convert_pytorch_argmax(%arg0: tensor<1x9xi32>) -> tensor<1xi32> {
  %0 = mhlo.constant dense<0> : tensor<i32>

  // CHECK-LABEL: @xla_svd
  func.func @xla_svd(%arg0: tensor<1x1xf32>) -> (tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>) {
    // CHECK-NOT: XlaSvd
    %s, %u, %v = "tf.XlaSvd"(%arg0) {max_iter = 1, epsilon = 1.0E-09 : f32, precision_config = ""} : (tensor<1x1xf32>) -> (tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>)
    func.return %s, %u, %v : tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>
  }

  func.func @identity(%arg0: f32) -> f32 {

}
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:

    %3 = "tfl.reshape"(%2, %cst_0) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<1x1x1x2xf32>, tensor<1xi32>) -> tensor<2xf32>
    %4 = "tfl.reshape"(%3, %cst_1) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<2xf32>, tensor<2xi32>) -> tensor<2x1xf32>
    return %4 : tensor<2x1xf32>
  }
```

#### Compute Costs Pass
In the compute cost pass, we will essentially compute the cost of each op within

  %cst = "tf.Const"() {value = dense<0> : tensor<1x2xi32>} : () -> tensor<1x2xi32>
  %cst_0 = "tf.Const"() {value = dense<1> : tensor<i32>} : () -> tensor<i32>
  %cst_1 = "tf.Const"() {value = dense<2> : tensor<1xi32>} : () -> tensor<1xi32>
  %cst_2 = "tf.Const"() {value = dense<4> : tensor<1x2xi32>} : () -> tensor<1x2xi32>
  %0 = "tf.SpaceToBatchND"(%arg0, %cst_1, %cst_2) : (tensor<1x128x3xf32>, tensor<1xi32>, tensor<1x2xi32>) -> tensor<2x68x3xf32>

func.func @add_with_activation_transpose_rank_two(%arg0: tensor<1x2xf32>) -> tensor<2x1xf32> {
  %0 = stablehlo.constant dense<2.000000e+00> : tensor<2x1xf32>
  %1 = stablehlo.transpose %arg0, dims = [1, 0] : (tensor<1x2xf32>) -> tensor<2x1xf32>
  %2 = stablehlo.add %1, %0 : tensor<2x1xf32>
  return %2 : tensor<2x1xf32>
}
// CHECK: %[[TRANSPOSE_0:.+]] = stablehlo.transpose