indices_are_sorted = false
  } : (tensor<3x4x2xf32>, tensor<2x3x2xi64>) -> tensor<2x3x2x2xf32>
    %8 = "quantfork.qcast"(%7) {volatile} : (tensor<2x3x2x2xf32>) -> tensor<2x3x2x2x!quant.uniform<i8:f32, 0.13170163023705575:-1>>
    %9 = "quantfork.dcast"(%8) : (tensor<2x3x2x2x!quant.uniform<i8:f32, 0.13170163023705575:-1>>) -> tensor<2x3x2x2xf32>
    return %9 : tensor<2x3x2x2xf32>
  }

  // CHECK: quantized_dot_general_fn_1

      index_vector_dim = 2>,
    slice_sizes = array<i64: 1, 1, 2, 2>,
    indices_are_sorted = false
  } : (tensor<3x4x2x2xf32>, tensor<2x3x2xi64>) -> tensor<2x3x2x2xf32>
  return %0 : tensor<2x3x2x2xf32>
}

// CHECK-LABEL: func @gather_float
// CHECK: stablehlo.gather
// CHECK-NOT: tfl.gather_nd
// CHECK-NOT: tfl.gather

// -----

    return %0 : tensor<1x3x4x2xf32>
  }
}

// CHECK-LABEL: quantize_conv_fn
// CHECK-SAME: %[[ARG0:.+]]: tensor<1x3x4x3xf32>
// CHECK: %[[CST:.+]] = stablehlo.constant dense<3.000000e-01> : tensor<2x3x3x2xf32>
// CHECK: %[[Q:.+]] = "quantfork.qcast"(%[[CST]]) : (tensor<2x3x3x2xf32>) -> tensor<2x3x3x2x!quant.uniform<i8:f32, 6.000000e-03:-128>>

  %output_1, %min_2, %max_3, %histogram_4 = "tf.CustomAggregator"(%0) <{calibration_method = 5 : i32, id = "keeping_id", num_bins = 32 : i32, max_percentile = 0.000000e+00 : f32, min_percentile = 0.000000e+00 : f32}> : (tensor<1x2x2x2xf32>) -> (tensor<1x2x2x2xf32>, tensor<f32>, tensor<f32>, tensor<512xi64>)

  %cst = "tf.Const"() {value = dense<0.000000e+00> : tensor<2xf32>} : () -> tensor<2xf32>
  %0 = "tf.Conv2D"(%arg0, %arg1) {data_format = "NHWC", device = "", dilations = [1, 1, 2, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<*xf32>

  module {
    func.func @main(%arg0: tensor<2x2x2xf32>, %arg1: tensor<2x2x2xf32>) -> tensor<2x2x2xf32> attributes {_from_xla_call_module} {
      %0 = stablehlo.dot_general %arg0, %arg1,
        batching_dims = [0] x [0],
        contracting_dims = [2] x [1],
        precision = [DEFAULT, DEFAULT]
      : (tensor<2x2x2xf32>, tensor<2x2x2xf32>) -> tensor<2x2x2xf32>
        return %0 : tensor<2x2x2xf32>
    }
  }
)mlir";

func.func @ConvertSliceToIdentityI32(%arg0: tensor<2x3x4x5xf32>) -> tensor<2x3x4x5xf32> {
  %begin = arith.constant dense<0> : tensor<4xi32>
  %shape = arith.constant dense<[2,3,4,5]> : tensor<4xi32>
  %0 = "tfl.slice"(%arg0, %begin, %shape) : (tensor<2x3x4x5xf32>, tensor<4xi32>, tensor<4xi32>) -> tensor<2x3x4x5xf32>
  func.return %0 : tensor<2x3x4x5xf32>
  // CHECK: return %arg0
}

  %cst = arith.constant dense<[1, 3, 3, 2]> : tensor<4xi32>
  %cst_2 = arith.constant dense<[2, 3, 0, 1]> : tensor<4xi32>
  %2 = "tf.Transpose"(%arg, %cst_2) : (tensor<2x1x3x3xf32>, tensor<4xi32>) -> tensor<3x3x2x1xf32>
  %3 = "tf.Reshape"(%2, %cst) : (tensor<3x3x2x1xf32>, tensor<4xi32>) -> tensor<1x3x3x2xf32>
  return %3: tensor<1x3x3x2xf32>

  return %0 : tensor<1x1x2x2xf32>
}
func.func private @XlaCallModule_aten.avg_pool2d.default.impl_4(%arg0: tensor<1x1x3x3xf32>) -> tensor<1x1x2x2xf32>

// CHECK-LABEL: avg_pool2d_5
// CHECK: %cst = arith.constant dense<[0, 2, 3, 1]> : tensor<4xi32>

  indices_are_sorted = false
} : (tensor<3x4x2xi32>, tensor<2x3x2xi64>) -> tensor<2x3x2x2xi32>
  func.return %2: tensor<2x3x2x2xi32>
}
// CHECK: %[[OPERAND:.*]] = stablehlo.constant
// CHECK: %[[INDICES:.*]] = stablehlo.constant
// CHECK: %[[XLA_CALL_MODULE:.*]] = "tf.XlaCallModule"(%[[OPERAND]], %[[INDICES]])
// CHECK: return %[[XLA_CALL_MODULE:.*]] : tensor<2x3x2x2xi32>
// CHECK: }

// CHECK-LABEL: private @composite_gather_fn_1