Function fetch_cache_line_width

less_slow.cpp:2782–2809

* @brief Fetches the cache line size using OS-specific APIs. * Supports Linux, macOS, and Windows. * * It's easier to use the @b `std::hardware_destructive_interference_size` * in C++ 17 and newer, if the `__cpp_lib_hardware_interference_size` feature * macro is defined. But this will be incorrect, if the compilation platform * is different from the runtime platform. * * A

 *  between Intel and AMD!
 */
std::size_t fetch_cache_line_width() {

#if defined(__linux__)
    // On Linux, we can read the cache line size and the L2 cache size from the
    // "sysfs" virtual filesystem. It can provide the properties of each individual
    // CPU core.
    std::string file_contents = read_file_contents("/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size");
    std::size_t cache_line_size = std::stoull(file_contents);
    return cache_line_size;

#elif defined(__APPLE__)
    // On macOS, we can use the `sysctlbyname` function to read the
    // `hw.cachelinesize` and `hw.l2cachesize` values into unsigned integers.
    // You can achieve the same by using the `sysctl -a` command-line utility.
    size_t size;
    size_t len = sizeof(size);
    if (sysctlbyname("hw.cachelinesize", &size, &len, nullptr, 0) == 0) return size;

#elif defined(_WIN32)
    SYSTEM_LOGICAL_PROCESSOR_INFORMATION buffer[256];
    DWORD len = sizeof(buffer);
    if (GetLogicalProcessorInformation(buffer, &len))
        for (size_t i = 0; i < len / sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION); ++i)
            if (buffer[i].Relationship == RelationCache && buffer[i].Cache.Level == 1) return buffer[i].Cache.LineSize;
#endif

    return 0;
}

/**
 *  @brief  A minimalistic pointer with non-unit stride/step.