// CHECK: return %[[GATHER]] : tensor<1x300x10xf32>

// -----

module {
  func.func @propagate_xlagather(%arg0: tensor<10x2xi32>) -> tensor<1x300x10xf32> {
    %cst = "tf.Const"() {value = dense<[1, 1, 300]> : tensor<3xi64>} : () -> tensor<3xi64>

  // CHECK-NEXT:  return [[FIND]] : tensor<*xi64>
  func.return %1 : tensor<*xi64>
}

// -----

// Test for case with int64 -> string hashtable.
func.func @hashtable_int64_to_string(%arg0: tensor<i64>) -> tensor<*x!tf_type.string> {
  %cst = arith.constant dense<"f"> : tensor<!tf_type.string>

  // CHECK: %[[CST:.*]] = "tf.Const"() <{value = dense<[0, 2, 3, 1]> : tensor<4xi64>}>
  // CHECK: %[[R0:.*]] = "tf.Transpose"(%[[ARG0]], %[[CST]])
  // CHECK: %[[R1:.*]] = "tf.BiasAdd"(%[[R0]], %[[ARG1]]) <{data_format = "NHWC"}> {device = ""}
  // CHECK: %[[CST_0:.*]] = "tf.Const"() <{value = dense<[0, 3, 1, 2]> : tensor<4xi64>}>
  // CHECK: "tf.Transpose"(%[[R1]], %[[CST_0]])

// CHECK-LABEL: func @transposeConv2D
func.func @transposeConv2D(%input: tensor<1x32x32x3xf32>, %filter: tensor<1x1x3x8xf32>) -> tensor<1x7x7x8xf32> {

  // CHECK: %[[ARG_PERM:.*]] = "tf.Const"() <{value = dense<[0, 3, 1, 2]> : tensor<4xi64>}>
  // CHECK: %[[ARG_TRANSPOSE:[0-9]*]] = "tf.Transpose"(%arg0, %[[ARG_PERM]])

  // CHECK: %[[CONV2D:[0-9]*]] = "tf.Conv2D"(%[[ARG_TRANSPOSE]], %arg1)
  // CHECK-SAME: data_format = "NCHW"

  func.return %2, %3 : tensor<*xi64>, tensor<*xi64>
}

// -----

// CHECK-LABEL: @not_eliminate_multinomial
func.func @not_eliminate_multinomial(%0: tensor<*xf32>, %1: tensor<*xi32>) -> (tensor<*xi64>, tensor<*xi64>) {
  // CHECK-NEXT: tf.Multinomial
  // CHECK-SAME: seed = 0
  // CHECK-NEXT: tf.Multinomial
  // CHECK-SAME: seed = 1
  // CHECK-NEXT: tf.Multinomial

  func.return %cst : tensor<1x750xf32>

  // CHECK-DAG: %0 = mhlo.constant dense<7.680000e+02> : tensor<f32>
  // CHECK: %1 = "mhlo.broadcast_in_dim"(%0) <{broadcast_dimensions = dense<> : tensor<0xi64>}> : (tensor<f32>) -> tensor<1x750xf32>
  // CHECK: return %1 : tensor<1x750xf32>
}

// CHECK-LABEL: @unfold_splat_constant_integer
func.func @unfold_splat_constant_integer() -> tensor<1x750xi32> {

  %cst = arith.constant dense<[2, 100]> : tensor<2xi64>
  // CHECK: tac.cost = 4.000000e+01
  %0 = "tfl.concatenation"(%arg0, %arg1) {axis = 0 : i32, fused_activation_function = "NONE", tac.device = "GPU"} : (tensor<100xf32>, tensor<100xf32>) -> tensor<200xf32>
  // CHECK: tac.cost = 4.040000e+01
  %1 = "tfl.reshape"(%0, %cst) {tac.device = "GPU"} : (tensor<200xf32>, tensor<2xi64>) -> tensor<2x100xf32>
  func.return %1 : tensor<2x100xf32>
}

// CHECK: %[[XLA_CALL_MODULE:.+]] = "tf.XlaCallModule"(%[[CUSTOM_AGGREGATOR_0]], %[[CST]])
// CHECK-SAME: _entry_function = @composite_dot_general_fn_1, _original_entry_function = "composite_dot_general_fn_1"

// RUN: tf-opt %s -tfl-prepare-tf -tfl-legalize-tf='run-tfl-runtime-verification=false'  -tfl-optimize | FileCheck %s

func.func @broadcast_to_bf16(%arg0: tensor<3xbf16>, %arg1: tensor<2xi64>) -> tensor<3x3xbf16> {
  %0 = "tf.BroadcastTo"(%arg0, %arg1) : (tensor<3xbf16>, tensor<2xi64>) -> tensor<3x3xbf16>
  func.return %0: tensor<3x3xbf16>

// CHECK-LABEL: broadcast_to_bf16
// CHECK:  [[CST:%.*]] = arith.constant dense<1.000000e+00> : tensor<3x3xbf16>

  %1:4 = "tf.CustomAggregator"(%arg0) {id = "1", calibration_method = 1 : i32, num_bins = 0 : i32, max_percentile = 0.000000e+00 : f32, min_percentile = 0.000000e+00 : f32} : (tensor<8x8x8x8xf32>) -> (tensor<8x8x8x8xf32>, tensor<f32>, tensor<f32>, tensor<*xi64>)
  func.return %0#0, %1#0 : tensor<8x8x8x8xf32>, tensor<8x8x8x8xf32>
}
// CHECK: func @customAggregator