}

// CHECK-LABEL: @splat_scalar_no_change
func.func @splat_scalar_no_change() -> (tensor<f32>, tensor<i32>) {
  // CHECK-NOT: mhlo.broadcast_in_dim
  %cst0 = mhlo.constant dense<0.000000e+00> : tensor<f32>
  %cst1 = mhlo.constant dense<0> : tensor<i32>
  func.return %cst0, %cst1 : tensor<f32>, tensor<i32>

  %cst0 = mhlo.constant dense<[1.0, 2.0, 3.0, 4.0]> : tensor<4xf32>
  %cst1 = mhlo.constant dense<[[[[1.0, 2.0, 3.0, 4.0], [5.0, 6.0, 7.0, 8.0]]]]> : tensor<1x1x2x4xf32>
  %0 = "mhlo.broadcast_in_dim"(%cst0) <{broadcast_dimensions = dense<3> : tensor<1xi64>}> : (tensor<4xf32>) -> tensor<1x1x2x4xf32>
  %1 = mhlo.multiply %0, %cst1 : tensor<1x1x2x4xf32>
  // CHECK:      return %[[RES]] : tensor<1x1x2x4xf32>

  // CHECK:      %[[CST0:.+]] = arith.constant dense<0.000000e+00>
  // CHECK-NEXT: %[[CST1:.+]] = arith.constant dense<0.000000e+00>
  // CHECK-NEXT: "vhlo.add_v1"(%[[CST0]], %[[CST1]]) : (tensor<f32>, tensor<f32>) -> tensor<f32>
  %cst_0 = arith.constant dense<0.000000e+00> : tensor<f32>
  %cst_1 = arith.constant dense<0.000000e+00> : tensor<f32>
  %0 = stablehlo.add %cst_0, %cst_1 : tensor<f32>
  return %0 : tensor<f32>

  module attributes {tf.versions = {bad_consumers = [], min_consumer = 0 : i32, producer = 268 : i32}} {
    func.func @main() -> () {
      %cst0 = "tf.Const"(){ value = dense<0> : tensor<3x5xi1>} : () -> tensor<3x5xi1>
      %0 = "tf.Where"(%cst0) : (tensor<3x5xi1>) -> tensor<?x2xi64>
      func.return
    }
  })";

  CellReader<int64_t> dynamism_counter(kDynamismOpCounterStreamzName);

  // CHECK-DAG: %[[cst0:.*]] = stablehlo.constant dense<0xFF800000> : tensor<f32
  // CHECK-DAG: %[[cst1:.*]] = stablehlo.constant dense<0x7FC00000> : tensor<f32>
  // CHECK-DAG: %[[cst2:.*]] = stablehlo.constant dense<6.000000e+00> : tensor<f32>
  // CHECK-DAG: %[[cst3:.*]] = stablehlo.constant dense<0.000000e+00> : tensor<f32>
  // CHECK-NOT: %[[q0:.*]] = "quantfork.qcast"(%[[cst0]])

func.func @fuseScalarAddIntoConv2dHalf(%arg0: tensor<256x32x32x3xf16>, %arg1: tensor<16x3x3x3xf16>) -> tensor<256x8x7x16xf16> {
  %cst = arith.constant dense<1.5> : tensor<f16>
  %cst_0 = arith.constant dense<[1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0]> : tensor<16xf16>

  %cst = arith.constant dense<1.000000e+00> : tensor<256x256xf32>
  %cst_0 = arith.constant dense<0> : tensor<i32>
  %cst_1 = arith.constant dense<-1> : tensor<i32>

    %cst = "tf.Const"() {device = "", value = dense<[7.11401462, 7.05456924]> : tensor<2xf32>} : () -> tensor<2xf32>
    // CHECK: %[[q_weight_per_channel:.*]] = "quantfork.qcast"
    // CHECK-SAME: -> tensor<2x3x3x2x!quant.uniform<i8<-127:127>:f32:3, {0.075123051020104109,0.072960192762960605}>>
    // CHECK: %[[dq_weight:.*]] = "quantfork.dcast"(%[[q_weight_per_channel]])

// CHECK-NOT: tf.ReadVariableOp
  %0 = "tf.ReadVariableOp"(%arg0) : (tensor<*x!tf_type.resource<tensor<1xf32>>>) -> tensor<1xf32>
  %cst = "tf.Const"() <{value = dense<0.000000e+00> : tensor<f32>}> : () -> tensor<f32>
  %cst_0 = "tf.Const"() <{value = dense<1> : tensor<1xi64>}> : () -> tensor<1xi64>
  %fill = "tf.Fill"(%cst_0, %cst) : (tensor<1xi64>, tensor<f32>) -> tensor<1xf32>
  tf_device.replicate([%0, %fill] as %arg_r0: tensor<1xf32>) {n = 2 : i32} {

    moved after _TPUCompileMlir and the first input will use the
    _TPUCompileMlir program output:

    ```mlir
    "tf_device.launch"() ({
     %cst_0 = "tf.Const"() {value = dense<""> : tensor<1x!tf_type.string>} : () -> tensor<1x!tf_type.string>
     "tf.OpA"(%cst_0) { mini_batch_splits = ""} : (tensor<1x!tf_type.string>) -> ()
     tf_device.return
   }) {device = "/job:localhost/replica:0/task:0/device:CPU:0"} : () -> ()