%0:2 = "tfl.bidirectional_sequence_lstm"(%arg0, %arg1, %arg2, %arg3, %arg4, %arg5, %arg6, %arg7, %arg8, %arg9, %arg10, %arg11, %arg12, %arg13, %arg14, %arg15, %arg16, %arg17, %arg18, %arg19, %arg20, %arg21, %arg22, %arg23, %arg24, %arg25, %arg26, %arg27, %arg28, %arg29, %arg30, %arg31, %arg32, %arg33, %arg34, %arg35, %arg36, %arg37, %arg38, %arg39, %arg40, %arg41, %arg42, %arg43, %arg44, %arg45, %arg46, %arg47) {cell_clip = 1.000000e+00 : f32, fused_activation_function...

  // CHECK: "tfl.unidirectional_sequence_lstm"(%arg0, %arg1, %arg2, %arg3, %arg4, %arg5, %arg6, %arg7, %arg8, %arg9, %arg10, %arg11, %arg12, %arg13, %arg14, %arg15, %arg16, %arg17, %arg18, %arg19, %arg20, %arg21, %arg22, %arg23) <{fused_activation_function = "NONE", time_major = false}> : (tensor<600x10x20xf32>, tensor<?x?xf32>,...

  // CHECK: "tfl.unidirectional_sequence_lstm"(%arg0, %arg1, %arg2, %arg3, %arg4, %arg5, %arg6, %arg7, %arg8, %arg9, %arg10, %arg11, %arg12, %arg13, %arg14, %arg15, %arg16, %arg17, %arg18, %arg19, %arg20, %arg21, %arg22, %arg23) <{asymmetric_quantize_inputs = false, cell_clip = 1.000000e+01 : f32, diagonal_recurrent_tensors...

  ^bb0(%arg1: tensor<i32>, %arg2: tensor<i32>, %arg3: tensor<i32>, %arg4: tensor<!tf_type.resource>, %arg5: tensor<f32>):
    %9 = "tf.Const"() {value = dense<10> : tensor<i32>} : () -> tensor<i32>
    %10 = "tf.Less"(%arg3, %9) {device = ""} : (tensor<i32>, tensor<i32>) -> tensor<i1>
    "tfl.yield"(%10) : (tensor<i1>) -> ()
  },  {

    reducer(%arg1: tensor<i32>, %arg3: tensor<i32>) (%arg2: tensor<i32>, %arg4: tensor<i32>)  {
    %6 = mhlo.compare  GE, %arg1, %arg3 : (tensor<i32>, tensor<i32>) -> tensor<i1>
    %7 = mhlo.select %6, %arg1, %arg3 : tensor<i1>, tensor<i32>
    %8 = mhlo.compare  EQ, %arg1, %arg3 : (tensor<i32>, tensor<i32>) -> tensor<i1>
    %9 = mhlo.minimum %arg2, %arg4 : tensor<i32>
    %10 = mhlo.select %6, %arg2, %arg4 : tensor<i1>, tensor<i32>

		// round 1
		{{range $i, $s := dup 4 .Shift1 -}}
			{{printf "arg0 = arg1 + bits.RotateLeft32((((arg2^arg3)&arg1)^arg3)+arg0+x%x+%#08x, %d)" (idx 1 $i) (index $.Table1 $i) $s | relabel}}
			{{rotate -}}
		{{end}}

		// round 2
		{{range $i, $s := dup 4 .Shift2 -}}
			{{printf "arg0 = arg1 + bits.RotateLeft32((((arg1^arg2)&arg3)^arg2)+arg0+x%x+%#08x, %d)" (idx 2 $i) (index $.Table2 $i) $s | relabel}}
			{{rotate -}}
		{{end}}

  // CHECK-SAME: %[[ARG0:.*]]: tensor<1x2xf32>
  // CHECK-SAME: %[[ARG1:.*]]: tensor<2x3xf32>
  // CHECK-SAME: %[[ARG2:.*]]: tensor<1x3xi1>
  // CHECK-SAME: %[[ARG3:.*]]: tensor<1x3xf32>
  func.func private @composite_and_select(%arg0: tensor<1x2xf32>, %arg1: tensor<2x3xf32>, %arg2: tensor<1x3xi1>, %arg3: tensor<1x3xf32>) -> tensor<1x3xf32> {

// RUN: tf-opt %s -split-input-file -verify-diagnostics -tf-tpu-validate-inputs | FileCheck %s

// CHECK-LABEL: func @num_replicas_replicated
func.func @num_replicas_replicated(%arg0: tensor<i32>, %arg1: tensor<i32>, %arg2: tensor<i32>, %arg3: tensor<i32>) -> (tensor<i32>, tensor<i32>) {
  %0:2 = tf_executor.graph {

module {
func.func @main(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>, %arg2: tensor<1xf32>, %arg3: tensor<1xf32>) -> tensor<2x1xf32> attributes {tf.entry_function = {inputs = "input0,input1,input2,input3", outputs = "output"}} {
  %0 = "tfl.add"(%arg0, %arg1) {fused_activation_function = "RELU6"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>
  %1 = "tfl.mul"(%0, %arg2) {fused_activation_function = "RELU6"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>

  // CHECK: "mhlo.batch_norm_inference"({{.*}}, %arg1, %arg2, %arg3, %arg4) <{epsilon = 1.000000e-03 : f32, feature_index = 3 : i64}> : (tensor<8x8x8x8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>) -> tensor<8x8x8x8xf32>