%0 = arith.constant dense<[[1.0, 2.0]]> : tensor<1x2xf32>
  %1 = "tfl.batch_matmul"(%arg0, %0) {adj_x = false, adj_y = true, asymmetric_quantize_inputs = false} : (tensor<4x128x2xf32>, tensor<1x2xf32>) -> tensor<4x128x1xf32>
  func.return %1 : tensor<4x128x1xf32>
  // CHECK: %[[CONST_WEIGHT:.*]] = arith.constant
  // CHECK-SAME: [1.000000e+00, 2.000000e+00]
  // CHECK-SAME: tensor<1x2xf32>

// -----

// Tests that int constants are not folded.

// CHECK-LABEL: transpose_int
func.func @transpose_int() -> tensor<3x2xi32> {
  %0 = stablehlo.constant dense<0> : tensor<2x3xi32>
  %1 = stablehlo.transpose %0, dims = [1, 0] : (tensor<2x3xi32>) -> tensor<3x2xi32>
  return %1 : tensor<3x2xi32>
}
// CHECK: transpose

// -----

// Tests that transposing an argument cannot be folded.

// CHECK-LABEL: transpose_arg

// 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>
  }
}

// -----

      time_major = false} : (
        tensor<1x2x3xf32>,
        tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>,
        tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>,
        none, none, none,
        tensor<3xf32>, tensor<3xf32>, tensor<3xf32>, tensor<3xf32>,
        none, none,
        tensor<1x3xf32>, tensor<1x3xf32>,
        none, none, none, none) -> tensor<1x2x3xf32>

}

// 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>>
  %cst = "tfl.pseudo_qconst"() {qtype = tensor<1x2x!quant.uniform<u8:f32, 1.0>>, value = dense<-76> : tensor<1x2xi8>} : () -> 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-NEXT:     %[[GROUP:.*]] = "tf.Const"()
  // CHECK-SAME:       [0, 1, 2, 3]
  // CHECK-NEXT:     %[[REDUCED:.*]] = "tf.XlaAllReduce"(%[[ID]], %[[GROUP]]) <{mode = "CrossReplica", reduce_op = "Add"}> : (tensor<f32>, tensor<1x4xi32>) -> tensor<f32>
  // CHECK-NEXT:     "tf.OpA"(%[[REDUCED]]) : (tensor<f32>) -> ()
  tf_device.replicate {n = 4 : i32} {
    "tf_device.cluster"() ({
      "tf.OpA"(%arg0) : (tensor<f32>) -> ()
      tf_device.return

  return %2 : tensor<1x3xf32>
}
func.func private @composite_dot_general_fn_1(%arg0: tensor<1x1024xf32>, %arg1: tensor<1024x3xf32>) -> tensor<1x3xf32> attributes {_from_xla_call_module} {
  %0 = stablehlo.dot_general %arg0, %arg1, contracting_dims = [1] x [0] : (tensor<1x1024xf32>, tensor<1024x3xf32>) -> tensor<1x3xf32>
  return %0 : tensor<1x3xf32>
}
// CHECK: func.func @main