%1 = "tf.PartitionedCall"(%arg0, %cst_0) {_tfl_quant_trait = "fully_quantizable", config = "", config_proto = "", executor_type = "", f = @composite_matmul_fn_1} : (tensor<2x12xf32>, tensor<12x2xf32>) -> tensor<*xf32>
    func.return %1: tensor<*xf32>
  }

// CHECK-SAME: (%[[ARG_0:[a-z0-9]*]]: tensor<!tf_type.resource<tensor<10x3xf32>>>, %[[ARG_1:[a-z0-9]*]]: tensor<!tf_type.resource<tensor<10x3xf32>>>, %[[ARG_2:[a-z0-9]*]]: tensor<!tf_type.resource<tensor<10x3xf32>>>, %[[ARG_3:[a-z0-9]*]]: tensor<!tf_type.resource<tensor<10x3xf32>>>)
!rtype = tensor<!tf_type.resource<tensor<10x3xf32>>>

// CHECK: return

  func.func private @composite_dot_general_fn(%arg0: tensor<1x4xf32>, %arg1: tensor<4x3xf32>) -> tensor<1x3xf32> attributes {_from_xla_call_module} {
      %0 = stablehlo.dot_general %arg0, %arg1, contracting_dims = [1] x [0] : (tensor<1x4xf32>, tensor<4x3xf32>) -> tensor<1x3xf32>
      return %0 : tensor<1x3xf32>
  }
}

// -----

  %1 = "tf.ReadVariableOp"(%0) {device = "/CPU:0"} : (tensor<!tf_type.resource<tensor<1x1xf32>>>) -> tensor<1x1xf32>
  %2:2 = "tf.Split"(%cst, %arg0) {device = "/CPU:0"} : (tensor<i32>,  tensor<1x4xf32>) -> (tensor<1x2xf32>, tensor<1x2xf32>)
  %3:2 = "tf.Split"(%cst, %2#0) {device = "/CPU:0"} : (tensor<i32>,  tensor<1x2xf32>) -> (tensor<1x1xf32>, tensor<1x1xf32>)

  // CHECK-NEXT: %[[UPDATE:.*]] = "tf.XlaDynamicUpdateSlice"(%[[STACK_VAL]], %[[UPDATE_SLICE]], %[[CONCAT_OFFETS]]) : (tensor<10x2xi32>, tensor<1x2xi32>, tensor<2xi32>) -> tensor<10x2xi32>
  // CHECK-NEXT: "tf.AssignVariableOp"(%[[BUFFER]], %[[UPDATE]]) : (tensor<!tf_type.resource<tensor<10x2xi32>>>, tensor<10x2xi32>) -> ()
  // CHECK-NEXT: %[[CONST1:.*]] = "tf.Const"() <{value = dense<1> : tensor<1xi32>}> : () -> tensor<1xi32>

  func.return %1 : tensor<10x10x10xf32>
}

// -----

func.func @pack_CPU(%arg0: tensor<100xf32>, %arg1: tensor<100xf32>) -> tensor<2x100xf32> attributes {tac.device = "CPU", tac.interface_name = "func_2"} {
  // CHECK: tac.cost = 1.000000e+02
  %0 = "tfl.pack"(%arg0, %arg1) {axis = 0 : i32, tac.device = "CPU", values_count = 2 : i32} : (tensor<100xf32>, tensor<100xf32>) -> tensor<2x100xf32>

  func.return %6 : tensor<1x401408xf32>
}

// CHECK-LABEL: quantized_constant
func.func @quantized_constant(%arg0: tensor<1x2xf32>) -> tensor<2x2xf32> {
  %1 = "tfl.quantize"(%arg0) {qtype = tensor<1x2x!quant.uniform<u8:f32, 1.0>>, volatile} : (tensor<1x2xf32>) -> tensor<1x2x!quant.uniform<u8:f32, 1.0>>

// CHECK-DAG: %[[input_18:.*]] = "quantfork.stats"({{.*}}) <{layerStats = dense<[-8.000000e-01, 1.600000e+00]> : tensor<2xf32>}> : (tensor<1x4xf32>) -> tensor<1x4xf32>
// CHECK-DAG: %[[input_19:.*]] = "quantfork.stats"({{.*}}) <{layerStats = dense<[-2.000000e+00, 4.000000e+00]> : tensor<2xf32>}> : (tensor<1x2xf32>) -> tensor<1x2xf32>

// CHECK: "tfl.unidirectional_sequence_lstm"({{.*}}, %[[input_18]], %[[input_19]], %{{[0-9]+}}, %{{[0-9]+}}, %{{[0-9]+}}, %{{[0-9]+}})

// RUN:   -split-input-file | FileCheck %s --check-prefix=STATIC-RANGE-PTQ-TO-COMPUTE-HEAVY

// STATIC-RANGE-PTQ-TO-COMPUTE-HEAVY: @main
func.func @main(%arg0: tensor<1x2xf32>) -> tensor<1x2xf32> {
  %0 = stablehlo.add %arg0, %arg0 : tensor<1x2xf32>
  return %0 : tensor<1x2xf32>
}
// Tests that `composite_add_fn_1` does not quantize when quantizing
// only compute-heavy ops.

// CHECK-LABEL: reduce_window_max_activation_transpose
// CHECK-SAME: (%[[ARG:.+]]: tensor<16x8xf32>) -> tensor<4x8xf32>
func.func @reduce_window_max_activation_transpose_rank2(%arg0: tensor<16x8xf32>) -> tensor<4x8xf32> {
  %0 = stablehlo.constant dense<0xFF800000> : tensor<f32>  // -inf
  %1 = stablehlo.transpose %arg0, dims = [1, 0] : (tensor<16x8xf32>) -> tensor<8x16xf32>
  %2 = "stablehlo.reduce_window"(%1, %0) ({