//
// Examples:
//   * If `xla_gather_op_output_type` == tensor<*xf32>, then it returns:
//     tensor<*xf32>.
//   * If `xla_gather_op_output_type` == tensor<3x5xi32> and `collapsed_dims` ==
//     {0}, then it returns: tensor<1x3x5xi32>.
//   * If `xla_gather_op_output_type` == tensor<3x5xf32> and `collapsed_dims` ==
//     {1, 3}, then it returns: tensor<3x1x5x1xf32>.
Type GetSliceOpOutputType(Type xla_gather_op_output_type,

// CHECK-NEXT:      %6 = "tfl.pseudo_const"() <{value = dense<[-1, -1, 0]> : tensor<3xi32>}> : () -> tensor<3xi32>
// CHECK-NEXT:      %7 = "tfl.pseudo_const"() <{value = dense<1> : tensor<i32>}> : () -> tensor<i32>
// CHECK-NEXT:      %8 = "tfl.unsorted_segment_prod"(%4, %5, %7) : (tensor<3xi32>, tensor<3xi32>, tensor<i32>) -> tensor<1xi32>

  func.return %2 : tensor<1x128x32xf32>

// CHECK-DAG:  [[VAL_1:%.*]] = arith.constant dense<0> : tensor<3xi32>
// CHECK-DAG:  [[VAL_2:%.*]] = arith.constant dense<[1, 128, 32]> : tensor<3xi32>
// CHECK:  [[VAL_3:%.*]] = "tfl.slice"(%arg0, [[VAL_1]], [[VAL_2]]) : (tensor<4x128x32xf32>, tensor<3xi32>, tensor<3xi32>) -> tensor<1x128x32xf32>
}

// -----

// CHECK-LABEL: @WhileCanonicalizeBug

  %2 = "tfl.dequantize"(%arg0) : (tensor<*x!quant.uniform<u8:f32, 1.0>>) -> tensor<*xf32>
  %3 = "tfl.dequantize"(%arg1) : (tensor<?x?x4x!quant.uniform<u8:f32, 1.0>>) -> tensor<?x?x4xf32>
  %5 = "tfl.shape"(%2) : (tensor<*xf32>) -> tensor<?xi32>
  %6 = "tfl.shape"(%3) : (tensor<?x?x4xf32>) -> tensor<3xi32>
  func.return %5, %6 : tensor<?xi32>, tensor<3xi32>