是否可以在编译时初始化数组，以便在运行时不需要时间？

Question

If, for example, I want to create the Sieve of Eratosthenes up to 2 million, it will take a bit of time at run-time. Is there a way to reduce the run-time by using metaprogramming?

Code for the Sieve (initializes std::bitset up to const MILLION ):

void initCiur(std::bitset<MILLION> &c)
{
    c[0] = c[1] = 1;
    for (int i=4; i <= MILLION; i+=2)
        c[i] = 1;
    for (int i=3; i*i <= MILLION; i+=2)
        if (!c[i])
            for (int j=3*i; j <= MILLION; j+=2*i)
                c[j] = 1;
}

Answer 1

Is there a way to reduce the run-time by using metaprogramming?

Of course yes. I understand " metaprogramming " very broadly, as any program generating code (not just clever C++ templates).

On most C++ implementations (eg if using some recent GCC compiler on some GNU/Linux operating system), a C++ translation unit practically is a textual file . Read also the C++11 standard n3337 .

So just configure your build automation tool (eg GNU make or ninja ) to generate at build time (perhaps with a Python or Guile script, or your C++ metaprogram) some C++ file, or some file that would be #include -d elsewhere. Or write your GCC plugin doing so.

BTW, this idea is not new: ANTLR and GNU bison or Qt or SWIG do so (and recent GCC compilers -eg GCC 10 - have a dozen of specialized C++ code generators). And my abandoned project GCC MELT did so. And so do RefPerSys and bismon (for primes numbers useful in hashtables, both are projects I started). Parts of the primes_rps.cc file have been machine generated (see comments inside it).

In some cases, a clever optimizing compiler could optimize constexpr things. Be aware of Rice's theorem which states that it is not always possible to cleverly optimize. See also this draft report.

You could also be interested in code generating libraries such as asmjit or libgccjit , or in generating plugins at runtime (on Linux, you would use dlopen(3) and dlsym(3) , but read first the C++ dlopen mini-howto then how to write shared libraries ).

PS. with other operating systems (eg Windows) and C++ compilers things could be different, but most of the ideas above are reusable. Read the documentation of your C++ compiler and of your operating system . In cross-compiling approaches, you may need more tricks.