func.return %0 : tensor<3x4x4xf32>
}

func.func @batchmatmulv2_lhs_batch(%arg0: tensor<3x4x2xf32>, %arg1: tensor<2x4xf32>) -> tensor<3x4x4xf32> {
// CHECK-LABEL:   func @batchmatmulv2_lhs_batch
// CHECK:           "mhlo.dynamic_broadcast_in_dim"({{.*}}, {{.*}}) <{broadcast_dimensions = dense<[1, 2]> : tensor<2xi64>}>
// CHECK:           "mhlo.dot_general"({{.*}}, {{.*}}) <{

  %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: 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:      %[[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>
}

// CHECK-NEXT:    return %2 : tensor<3x2xf32>
}

func.func @transpose_3d(%arg0: tensor<1x2x3xf32>) -> tensor<3x2x1xf32> {
  %0 = "mhlo.transpose"(%arg0) <{permutation = dense<[2, 1, 0]> : tensor<3xi64>}> : (tensor<1x2x3xf32>) -> tensor<3x2x1xf32>
  func.return %0 : tensor<3x2x1xf32>

// CHECK-LABEL:   transpose_3d

func.func @test_batch_matmul_to_einsum(%arg0: tensor<1x2x3xf32>, %arg1: tensor<3x4xf32>) -> tensor<1x2x4xf32> {
  // CHECK-LABEL: test_batch_matmul_to_einsum
  // CHECK: "tf.Einsum"(%arg0, %arg1) <{equation = "...mk,...kn->...mn"}> : (tensor<1x2x3xf32>, tensor<3x4xf32>) -> tensor<1x2x4xf32>
  %0 = "tf.BatchMatMul"(%arg0, %arg1) {adj_x = false, adj_y = false} : (tensor<1x2x3xf32>, tensor<3x4xf32>) -> tensor<1x2x4xf32>
  func.return %0: tensor<1x2x4xf32>
}

func.func @testTranspose(tensor<2x3x4xf32>) -> tensor<3x2x4xf32> {
^bb0(%arg0: tensor<2x3x4xf32>):
  %cst = arith.constant dense<[2, 0, 1]> : tensor<3xi32>
  // expected-error @+1 {{requires y.shape[0] (3) to be equal to x.shape[perm[2]] (4)}}
  %0 = "tf.Transpose"(%arg0, %cst) {T = "tfdtype$DT_FLOAT", Tperm = "tfdtype$DT_INT32"} : (tensor<2x3x4xf32>, tensor<3xi32>) -> tensor<3x2x4xf32>
  func.return %0 : tensor<3x2x4xf32>
}

}

func.func @einsum_basic(%arg0: tensor<3x4x5xf32>, %arg1: tensor<3x5x6xf32>) -> tensor<3x4x6xf32> {
  %0 = "tf.Einsum"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", equation = "ijk,ikm->ijm"}: (tensor<3x4x5xf32>, tensor<3x5x6xf32>) -> tensor<3x4x6xf32>
  func.return %0 : tensor<3x4x6xf32>
  // CHECK-LABEL: einsum_basic
  // CHECK: "tf.BatchMatMulV2"(%arg0, %arg1) <{adj_x = false, adj_y = false}> : (tensor<3x4x5xf32>, tensor<3x5x6xf32>) -> tensor<3x4x6xf32>
}

  %lookup0 = "tfl.embedding_lookup"(%index0, %value0) : (tensor<5xi32>, tensor<3x2xf32>) -> tensor<5x2xf32>
  %lookup1 = "tfl.embedding_lookup"(%index1, %value1) : (tensor<3xi32>, tensor<2x2x2xf32>) -> tensor<3x2x2xf32>
  func.return %lookup0, %lookup1 : tensor<5x2xf32>, tensor<3x2x2xf32>

// CHECK:           %12 = "tf.Reshape"(%0, %11) : (tensor<2x?x3x4xf32>, tensor<4xi32>) -> tensor<2x?x3x4xf32>
// CHECK:           %13 = "tf.BatchMatMulV3"(%6, %12) <{adj_x = false, adj_y = false, grad_x = false, grad_y = false}> : (tensor<2x?x2x3xf32>, tensor<2x?x3x4xf32>) -> tensor<2x?x2x4xf32>
// CHECK:           %14 = "tf.Shape"(%arg0) : (tensor<2x?x2x3xf32>) -> tensor<4xi32>