// CHECK:      %[[SLICE3:.*]] = "tf.Slice"(%[[CONCAT]], %[[CST5]], %[[CST6]]) : (tensor<3x8x4xi32>, tensor<3xi64>, tensor<3xi64>) -> tensor<3x6x4xi32>
  // CHECK:      %[[CONCAT1:.*]] = "tf.ConcatV2"(%[[SLICE2]], %[[SLICE3]], %[[CST7]]) : (tensor<3x2x4xi32>, tensor<3x6x4xi32>, tensor<i32>) -> tensor<3x8x4xi32>
  // CHECK:      return %[[CONCAT1]] : tensor<3x8x4xi32>
}

func.func @batchmatmulv2_basic(%arg0: tensor<1x4x2xf32>, %arg1: tensor<3x2x4xf32>) -> tensor<3x4x4xf32> {
// CHECK-LABEL:   func @batchmatmulv2_basic
// CHECK-SAME:        ([[LHS:%.*]]: tensor<1x4x2xf32>, [[RHS:%.*]]: tensor<3x2x4xf32>) -> tensor<3x4x4xf32>
// CHECK:           [[LHSSHAPE:%.*]] = shape.shape_of [[LHS]] : tensor<1x4x2xf32>
// CHECK:           [[RHSSHAPE:%.*]] = shape.shape_of [[RHS]] : tensor<3x2x4xf32>
// CHECK:           [[CM2:%.*]] = arith.constant -2 : index

  %0 = "tf.BatchMatMulV2"(%arg0, %arg1) {T = f32, adj_x = false, adj_y = false, grad_x = false, grad_y = false, device = ""} : (tensor<1x4x2xf32>, tensor<3x2x4xf32>) -> tensor<3x4x4xf32>
  func.return %0 : tensor<3x4x4xf32>
}

module attributes {tf.versions = {producer = 888 : i32}} {
func.func @testReshapeShapeInference(%arg0: tensor<3x4xi32>) -> tensor<*xi32> {
  %cst = arith.constant dense<[1, 6, 2]> : tensor<3xi32>
  // CHECK: "tfl.reshape"(%arg0, %cst) : (tensor<3x4xi32>, tensor<3xi32>) -> tensor<1x6x2xi32>
  %0 = "tfl.reshape"(%arg0, %cst) : (tensor<3x4xi32>, tensor<3xi32>) -> tensor<*xi32>
  func.return %0 : tensor<*xi32>
}
}

// -----

    // CHECK: mhlo.reduce
    // CHECK: mhlo.dot_general
    // CHECK: mhlo.transpose
    %0 = "tf.BatchMatMulV2"(%arg0, %arg1) {T = f32, adj_x = false, adj_y = false, grad_x = false, grad_y = false, device = ""} : (tensor<1x4x2xf32>, tensor<3x2x4xf32>) -> tensor<3x4x4xf32>
    func.return %0 : tensor<3x4x4xf32>
  }

  // CHECK-LABEL: approx_topk

  // CHECK-NEXT: %8 = "mhlo.set_dimension_size"(%arg0, %7) <{dimension = 0 : i64}> : (tensor<3x4xi32>, tensor<i32>) -> tensor<3x4xi32>
  // CHECK-NEXT: %9 = mhlo.constant dense<4> : tensor<i32>
  // CHECK-NEXT: %10 = "mhlo.set_dimension_size"(%8, %9) <{dimension = 1 : i64}> : (tensor<3x4xi32>, tensor<i32>) -> tensor<3x4xi32>
  // CHECK-NEXT: %11 = mhlo.constant dense<0> : tensor<i32>

func.func @valid_unranked_inputs_on_reshape(%arg0: tensor<3x4xi32>, %arg1: tensor<*xi32>) -> tensor<3x4xi32> {
  // CHECK: "tfl.reshape"(%arg0, %arg1)
  %0 = "tfl.reshape"(%arg0, %arg1) : (tensor<3x4xi32>, tensor<*xi32>) -> tensor<3x4xi32>
  func.return %0 : tensor<3x4xi32>
}

// -----

// CHECK-LABEL: valid_one_dynamic_dim_on_reshape
func.func @valid_one_dynamic_dim_on_reshape(%arg0: tensor<3x4xi32>) -> tensor<1x3x4xi32> {

    %14 = stablehlo.subtract %12, %13 : tensor<8x16x16xi32>  // q1 - z1
    %15 = stablehlo.convert %11 : (tensor<8x16x4xi8>) -> tensor<8x16x4xi32>
    %16 = stablehlo.broadcast_in_dim %6, dims = [0, 1, 2] : (tensor<1x1x1xi32>) -> tensor<8x16x4xi32>
    %17 = stablehlo.subtract %15, %16 : tensor<8x16x4xi32>  // q2 - z2
    // Corresponds to einsum expression: b i j, b j d -> b i d

  }) {dimension = 1 : i64, is_stable = true} : (tensor<3x6xf32>, tensor<3x6xi32>) -> (tensor<3x6xf32>, tensor<3x6xi32>)
  func.return %2#0, %2#1 : tensor<3x6xf32>, tensor<3x6xi32>
}

// CHECK-LABEL:   func @convert_sort_to_topk_iotacst_broadcast(
// CHECK-SAME:                                                 %[[ARG_0:.*]]: tensor<3x6xf32>) -> (tensor<3x6xf32>, tensor<3x6xi32>) {

func.func @unary_einsum_reduce_sum_transpose(%arg0: tensor<3x4x5x6xf32>) -> tensor<3x5x4xf32> {
  %0 = "tf.Einsum"(%arg0) {T = "tfdtype$DT_FLOAT", equation = "...gse->...sg"}: (tensor<3x4x5x6xf32>) -> tensor<3x5x4xf32>
  func.return %0 : tensor<3x5x4xf32>
  // CHECK-LABEL: unary_einsum_reduce_sum_transpose
  // CHECK-DAG: %[[cst:.*]] = arith.constant dense<3> : tensor<1xi32>