We overlooked the performance issues with string-to-float and float-to-string conversions.

[GuoYongFa]

Formatting trajectory output using fmt is over 4x faster than using the C++ standard library prior to C++17. For data reading, by leveraging std::from_chars, we optimized the processing time for approximately 4 million atom reads from 6.95 seconds down to 1.34 seconds. Based on these findings, we need to refactor the Reader module and improve its modular design.

[GuoYongFa]

Reference code snippet

// Get references to atom IDs and types
auto &ids = _md_data._structure_data->_h_atoms_id;
auto &types = _md_data._structure_data->_h_atoms_type;
rbmd::Id atom_id;
rbmd::Id atom_type;

_line_start = &_mapped_memory[_locate];
auto atom_style = DataManager::getInstance().getConfigData()->Get<std::string>(
    "atom_style", "init_configuration", "read_data");

if ("atomic" == atom_style) {
  for (auto num = 0; _locate < _file_size && num < atoms_num; ++_locate) {
    if (_mapped_memory[_locate] == '\n') {
      char const *line_begin = _line_start;
      char const *line_end = &_mapped_memory[_locate];

      if (rbmd::IsLegalLine(
              std::string_view(line_begin, line_end - line_begin))) {
        // Skip leading whitespace
        while (line_begin < line_end && std::isspace(*line_begin)) {
          ++line_begin;
        }
        // Parse atom_id
        auto [ptr1, ec1] = std::from_chars(line_begin, line_end, atom_id);
        if (ec1 != std::errc()) {
          _line_start = &_mapped_memory[_locate + 1];
          continue;
        }
        line_begin = ptr1;

        // Skip whitespace
        while (line_begin < line_end && std::isspace(*line_begin)) {
          ++line_begin;
        }

        // Parse atom_type
        auto [ptr2, ec2] = std::from_chars(line_begin, line_end, atom_type);
        if (ec2 != std::errc()) {
          _line_start = &_mapped_memory[_locate + 1];
          continue;
        }
        line_begin = ptr2;

        // Store atom ID and type (convert to 0-based index)
        auto index = atom_id - 1;
        ids[index] = atom_id - 1;
        types[index] = atom_type - 1;

        // Parse three floating-point coordinates
        double x, y, z;

        // Skip whitespace and parse x
        while (line_begin < line_end && std::isspace(*line_begin)) {
          ++line_begin;
        }
        auto [ptr3, ec3] = std::from_chars(line_begin, line_end, x);
        if (ec3 != std::errc()) {
          _line_start = &_mapped_memory[_locate + 1];
          continue;
        }
        line_begin = ptr3;

        // Skip whitespace and parse y
        while (line_begin < line_end && std::isspace(*line_begin)) {
          ++line_begin;
        }
        auto [ptr4, ec4] = std::from_chars(line_begin, line_end, y);
        if (ec4 != std::errc()) {
          _line_start = &_mapped_memory[_locate + 1];
          continue;
        }
        line_begin = ptr4;

        // Skip whitespace and parse z
        while (line_begin < line_end && std::isspace(*line_begin)) {
          ++line_begin;
        }
        auto [ptr5, ec5] = std::from_chars(line_begin, line_end, z);
        if (ec5 != std::errc()) {
          _line_start = &_mapped_memory[_locate + 1];
          continue;
        }

        // Store coordinates
        _md_data._structure_data->_h_px[index] = x;
        _md_data._structure_data->_h_py[index] = y;
        _md_data._structure_data->_h_pz[index] = z;

        ++num;
      }
      // Move to the next line
      _line_start = &_mapped_memory[_locate];
    }
  }
}