#definerotate_left(x,n)与uint32_trotate_left(uint32_tx,uint32_tn)

Question

在我的 sha256 哈希类中，有以下函数。

    uint32_t shift_right ...

    uint32_t rotate_right(uint32_t x, uint32_t n)
    {
        return ((x >> n) | (x << ((sizeof(x) << 3) - n)));
    }

    uint32_t rotate_left ...
    uint32_t choose ...
    uint32_t majority ...

    void unpack32(size_t x, uint8_t* str)
    {
        *((str)+3) = (uint8_t)((x));
        *((str)+2) = (uint8_t)((x) >> 8);
        *((str)+1) = (uint8_t)((x) >> 16);
        *((str)+0) = (uint8_t)((x) >> 24);
    }

    void pack32(const uint8_t* str, uint32_t* x)
    {
        *(x) = ((uint32_t) * ((str)+3))
            | ((uint32_t) * ((str)+2) << 8)
            | ((uint32_t) * ((str)+1) << 16)
            | ((uint32_t) * ((str)+0) << 24);
    }

    uint32_t f1(uint32_t x)
    {
        return (rotate_right(x, 2) ^ rotate_right(x, 13) ^ rotate_right(x, 32));
    }

    uint32_t f2(uint32_t x)
    {
        return (rotate_right(x, 6) ^ rotate_right(x, 11) ^ rotate_right(x, 25));
    }

    uint32_t f3(uint32_t x)
    {
        return (rotate_right(x, 7) ^ rotate_right(x, 18) ^ shift_right(x, 3));
    }

    uint32_t f4(uint32_t x)
    {
        return (rotate_right(x, 17) ^ rotate_right(x, 19) ^ shift_right(x, 10));
    }

我的问题是：如果我使用下面的宏来代替这些函数，会不会更快一点？如果速度很快，为什么？

#define rotate_right(x, n) ((x >> n) | (x << ((sizeof(x) << 3) - n)))

Answer 1

使用以下代码：

#include <iostream>
#include <cstdint>

#define macro_rotate_right(x, n) ((x >> n) | (x << ((sizeof(x) << 3) - n)))

uint32_t rotate_right(uint32_t x, uint32_t n)
{
    return ((x >> n) | (x << ((sizeof(x) << 3) - n)));
}

uint32_t test1() {
    uint32_t input1;
    std::cin>>input1;

    uint32_t input2;
    std::cin>>input2;

    return macro_rotate_right(input1,input2);
}

uint32_t test2() {
    uint32_t input1;
    std::cin>>input1;

    uint32_t input2;
    std::cin>>input2;

    return rotate_right(input1,input2);
}

int main() {
    test1();
    test2();
    return 0;

}

Godbolt

-std=c++20 -O2

在两次测试中都显示了相同的组件：

rotate_right(unsigned int, unsigned int):
        mov     eax, edi
        mov     ecx, esi
        ror     eax, cl
        ret
test1():
        sub     rsp, 24
        mov     edi, OFFSET FLAT:_ZSt3cin
        lea     rsi, [rsp+8]
        call    std::basic_istream<char, std::char_traits<char> >& std::basic_istream<char, std::char_traits<char> >::_M_extract<unsigned int>(unsigned int&)
        lea     rsi, [rsp+12]
        mov     edi, OFFSET FLAT:_ZSt3cin
        call    std::basic_istream<char, std::char_traits<char> >& std::basic_istream<char, std::char_traits<char> >::_M_extract<unsigned int>(unsigned int&)
        mov     ecx, DWORD PTR [rsp+12]
        mov     eax, DWORD PTR [rsp+8]
        add     rsp, 24
        ror     eax, cl
        ret
test2():
        sub     rsp, 24
        mov     edi, OFFSET FLAT:_ZSt3cin
        lea     rsi, [rsp+8]
        call    std::basic_istream<char, std::char_traits<char> >& std::basic_istream<char, std::char_traits<char> >::_M_extract<unsigned int>(unsigned int&)
        lea     rsi, [rsp+12]
        mov     edi, OFFSET FLAT:_ZSt3cin
        call    std::basic_istream<char, std::char_traits<char> >& std::basic_istream<char, std::char_traits<char> >::_M_extract<unsigned int>(unsigned int&)
        mov     ecx, DWORD PTR [rsp+12]
        mov     eax, DWORD PTR [rsp+8]
        add     rsp, 24
        ror     eax, cl
        ret
main:
        sub     rsp, 8
        call    test1()
        call    test2()
        xor     eax, eax
        add     rsp, 8
        ret

因此它们的性能是相同的。事实上，该值是内联的。