%cst = "tf.Const"() { value = dense<[10, 3, 2]> : tensor<3xi32> } : () -> tensor<3xi32>
  %1 = "tf.ScatterNd"(%arg0, %arg1, %cst) : (tensor<5x1xi32>, tensor<5x3x2xf32>, tensor<3xi32>) -> tensor<10x3x2xf32>
  func.return %1 : tensor<10x3x2xf32>

// CHECK-LABEL:scatterNdVectorIndices
// CHECK: %[[CST:.*]] = arith.constant dense<[10, 3, 2]> : tensor<3xi32>

  %2 = "tf.Cast"(%1) {Truncate = false} : (tensor<1x1x2xbf16>) -> tensor<1x1x2xf32>
  %3 = "tf.IdentityN"(%2) {device = ""} : (tensor<1x1x2xf32>) -> tensor<1x1x2xf32>
  return %3 : tensor<1x1x2xf32>
}

// CHECK: func @cast_bf16_batch_matmul_v2_to_fp32
// CHECK-DAG: %[[cst:.*]] = "tf.Const"() <{value = dense<{{.*}}> : tensor<10x2xf32>}> : () -> tensor<10x2xf32>
// CHECK: %[[batch_matmul:.*]] = "tf.BatchMatMulV2"(%arg0, %[[cst]])

// CHECK-SAME: (%[[ARG0:.*]]: tensor<10x8x9xf32>, %[[ARG1:.*]]: tensor<5xi32>, %[[ARG2:.*]]: tensor<5x8x9xf32>) -> tensor<10x8x9xf32>
func.func @main(%arg0: tensor<10x8x9xf32>, %arg1: tensor<5xi32>, %arg2: tensor<5x8x9xf32>) -> tensor<10x8x9xf32> {
  %elem_shape = "tf.Const"() {value = dense<[8, 9]> : tensor<2xi32>} : () -> tensor<2xi32>
  // CHECK: %[[BUFFER:.*]] = "tf.Identity"(%[[ARG0]]) : (tensor<10x8x9xf32>) -> tensor<10x8x9xf32>

    %0 = stablehlo.dot_general %arg0, %arg1, batching_dims = [0] x [0], contracting_dims = [2] x [1] {mhlo.frontend_attributes = {grad_x = "false", grad_y = "false"}} : (tensor<10x1x1024xf32>, tensor<10x1024x3xf32>) -> tensor<10x1x3xf32>
    return %0 : tensor<10x1x3xf32>
  }
  // CHECK: %[[DOT:.*]] = stablehlo.dot_general

func.func @test_batch_matmulV2_adj_to_einsum(%arg0: tensor<1x3x2xf32>, %arg1: tensor<3x4xf32>) -> tensor<1x2x4xf32> {
  // CHECK: %[[RES_EINSUM:[0-9]*]] = "tf.Einsum"(%arg0, %arg1) <{equation = "...km,...kn->...mn"}> : (tensor<1x3x2xf32>, tensor<3x4xf32>) -> tensor<1x2x4xf32>
  // CHECK: return %[[RES_EINSUM]] : tensor<1x2x4xf32>
  %0 = "tf.BatchMatMulV2"(%arg0, %arg1) {adj_x = true, adj_y = false} : (tensor<1x3x2xf32>, tensor<3x4xf32>) -> tensor<1x2x4xf32>

    %output_0, %min_1, %max_2, %histogram_3 = "tf.CustomAggregator"(%0) <{calibration_method = 1 : i32, id = "keeping_id", max_percentile = 0.000000e+00 : f32, min_percentile = 0.000000e+00 : f32, num_bins = 0 : i32}> : (tensor<10x1x3xf32>) -> (tensor<10x1x3xf32>, tensor<f32>, tensor<f32>, tensor<0xi64>)
    return %output_0 : tensor<10x1x3xf32>
  }
  // CHECK-LABEL: @main

  %1 = "tf.Maximum"(%0, %cst_0) : (tensor<1x3x4x2xf32>, tensor<f32>) -> tensor<1x3x4x2xf32>
  %2 = "tf.Minimum"(%1, %cst_1) : (tensor<1x3x4x2xf32>, tensor<f32>) -> tensor<1x3x4x2xf32>
  func.return %2 : tensor<1x3x4x2xf32>
// CHECK-DAG: %[[CONST_0:.*]] = "tf.Const"() <{value = dense<{{.*}}> : tensor<1x1x3x2xf32>}> : () -> tensor<1x1x3x2xf32>

module {
func.func @tfInplaceUpdate(%arg0: tensor<2x1x2xf32>) -> tensor<2x1x2xf32> {
  %1 = arith.constant dense<1> : tensor<1xi32>
  %2 = arith.constant dense<2.0> : tensor<1x1x2xf32>
  %3 = "tf.InplaceUpdate"(%arg0, %1, %2) {device = ""}
    : (tensor<2x1x2xf32>, tensor<1xi32>, tensor<1x1x2xf32>) -> tensor<2x1x2xf32>
  func.return %3 : tensor<2x1x2xf32>
}
}

//CHECK: module attributes
//CHECK-SAME: keep_stablehlo_constant = "true"

  // CHECK-DAG: %[[CST:.+]] = mhlo.constant dense<[1.000000e-01, 2.000000e-01]> : tensor<2xf32>
  // CHECK-DAG: %[[CST_BCAST:.+]] = "mhlo.broadcast_in_dim"(%[[CST]]) <{broadcast_dimensions = dense<3> : tensor<1xi64>}> : (tensor<2xf32>) -> tensor<1x1x3x2xf32>
  // CHECK-DAG: %[[NEW_FILTER:.+]] =  mhlo.multiply %[[CST_BCAST]], %[[FILTER]] : tensor<1x1x3x2xf32>

  // CHECK: %[[v0:.*]] = "tf.Transpose"(%arg1, %[[cst]]) : (tensor<3x2x1xf32>, tensor<3xi32>) -> tensor<1x3x2xf32>
  // CHECK: %[[v1:.*]] = "tf.BatchMatMulV2"(%arg0, %[[v0]]) <{adj_x = false, adj_y = false}> : (tensor<5x4x3xf32>, tensor<1x3x2xf32>) -> tensor<5x4x2xf32>
  // CHECK: %[[v2:.*]] = "tf.Transpose"(%[[v1]], %[[cst_1]]) : (tensor<5x4x2xf32>, tensor<3xi32>) -> tensor<4x2x5xf32>