When the environment variable `TF_DETERMINISTIC_OPS` is set to `"true"`
        or `"1"` (when op-determinism is expected), an attempt to run the
        following ops on a GPU will throw `tf.errors.UnimplementedError` (with
        an understandable message) when `data` is a floating-point type,
        including complex types (if supported): `tf.math.segment_prod`,

                    char* buffer, TF_Status* status) {
  auto posix_file = static_cast<PosixFile*>(file->plugin_file);
  char* dst = buffer;
  int64_t read = 0;

  while (n > 0) {
    // Some platforms, notably macs, throw `EINVAL` if `pread` is asked to read
    // more than fits in a 32-bit integer.
    size_t requested_read_length;
    if (n > INT32_MAX)
      requested_read_length = INT32_MAX;
    else
      requested_read_length = n;