// 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>

// CHECK-LABEL:   func @batchmatmulv2_lhs_batch
// CHECK:           "mhlo.dynamic_broadcast_in_dim"({{.*}}, {{.*}}) <{broadcast_dimensions = dense<[1, 2]> : tensor<2xi64>}>
// CHECK:           "mhlo.dot_general"({{.*}}, {{.*}}) <{
// CHECK-SAME:        lhs_batching_dimensions = [0]
// CHECK-SAME:        rhs_batching_dimensions = [0]
// CHECK-SAME:        lhs_contracting_dimensions = [2]

  %cst0 = mhlo.constant dense<[1, 2, 3, 4]> : tensor<4xi32>
  // CHECK: %[[BROADCAST:.*]] = "mhlo.broadcast_in_dim"(%[[CONST]]) <{broadcast_dimensions = dense<3> : tensor<1xi64>}> : (tensor<4xi32>) -> tensor<1x1x2x4xi32>
  %0 = "mhlo.broadcast_in_dim"(%cst0) <{broadcast_dimensions = dense<3> : tensor<1xi64>}> : (tensor<4xi32>) -> tensor<1x1x2x4xi32>
  // CHECK: %[[MUL:.*]] = mhlo.multiply %[[BROADCAST]], %[[ARG]] : tensor<1x1x2x4xi32>

    -> (tensor<?x2xi64>, tensor<?x2xi64>, tensor<?x!tf_type.string>, tensor<?xi64>, tensor<2xi64>, tensor<2xi64>, tensor<32xf32>, tensor<32xf32>)

  // CHECK: tfrt.return {{.*}}, [[outputs]]#0
  func.return %outputs#0 : tensor<?x2xi64>
}

// CHECK-LABEL: func @no_native

  // 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>

  %2 = "mhlo.dynamic_broadcast_in_dim"(%cst_0, %1) <{broadcast_dimensions = dense<3> : tensor<1xi64>}> : (tensor<2xf32>, tensor<4xindex>) -> tensor<?x256x256x2xf32>
  %3 = mhlo.multiply %0, %2 : tensor<?x256x256x2xf32>
  %4 = "mhlo.dynamic_broadcast_in_dim"(%cst_1, %1) <{broadcast_dimensions = dense<3> : tensor<1xi64>}> : (tensor<2xf32>, tensor<4xindex>) -> tensor<?x256x256x2xf32>
  %5 = mhlo.add %3, %4 : tensor<?x256x256x2xf32>

  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> {

module {
  func.func @main() -> tensor<1xi64> {
  %0 = "tfl.custom"() {custom_code = "stablehlo.constant", custom_option = #tfl<const_bytes : "0x76616C75650001020109010101062C022401">} : () -> tensor<1xi64>
  func.return %0 : tensor<1xi64>
  }
}

// CHECK:  module {
// CHECK-NEXT:    func @main() -> tensor<1xi64> {
// CHECK-NEXT:    %[[c0:.+]] = stablehlo.constant dense<2> : tensor<1xi64>
// CHECK-NEXT:    return %[[c0]] : tensor<1xi64>

  // CHECK-DAG: %[[SCALE_BCAST:.+]] = "mhlo.broadcast_in_dim"(%[[SCALE]]) <{broadcast_dimensions = dense<1> : tensor<1xi64>}> : (tensor<256xf32>) -> tensor<4x256xf32>
  // CHECK-DAG: %[[OFFSET_BCAST:.+]] = "mhlo.broadcast_in_dim"(%[[OFFSET]]) <{broadcast_dimensions = dense<1> : tensor<1xi64>}> : (tensor<256xf32>) -> tensor<4x256xf32>

// 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"