div64 inline asm and other versions, C++ (vc++)

Language: Layout:

//for  http://rextester.com/l/cpp_online_compiler_visual

//Microsoft (R) C/C++ Optimizing Compiler Version 19.00.23026 for x86

//#include <iostream>
#include <string>
#include <boost/filesystem.hpp>
#include <Windows.h>
//#include <direct.h>

using namespace std;

static string my_exe(void){
    char buf[MAX_PATH];
    DWORD tmp = GetModuleFileNameA( NULL, // self
                                  buf, MAX_PATH);
    return buf;
    //cout << "GMFN: " << tmp << ' ' << buf << endl;
}

int main()
{
    string dircmd = "dir ";
    boost::filesystem::path p( my_exe() );
    //boost::filesystem::path dir = p.parent_path();
    p.remove_filename();
    //p.replace_extension(".asm");

system ( (dircmd + p.string()).c_str() );

auto subdir = p.end();
    subdir--;
    // cout << "subdir: " << subdir->c_str() << endl;
    p /= *subdir;
    p.replace_extension(".asm");
    // cout << "p: " << p.string() << endl;
    system ( (string("type ") + p.string()).c_str() );
}

#include <stdint.h>

#define ABI _vectorcall // compiling with -Gv so vectorcall is the default doesn't work: boost doesn't link

// MSVC
// we could return a struct, but that would complicate thing
int ABI div64(int hi, int lo, int divisor, int *premainder) {
    int quotient, tmp;
    __asm {
        mov   edx, hi;
        mov   eax, lo;
        idiv   divisor
        mov   quotient, eax
        mov   tmp, edx;
        // mov ecx, premainder   // Or this I guess?
        // mov   [ecx], edx
    }
    *premainder = tmp;
    return quotient;
}

int ABI div64_noasm(int lo, int hi, int *premainder, int divisor) {
    int64_t hilo = ( ((uint64_t)hi) << 32) | (uint64_t)lo;
    *premainder = hilo % divisor;
    return hilo / divisor;
}

int ABI div64_noasm_merged(int64_t hilo, int *premainder, int divisor) {
    *premainder = hilo % divisor;
    return hilo / divisor;
}

unsigned ABI div64_noasm_u64_merged(uint64_t hilo, unsigned *premainder, unsigned divisor) {
    *premainder = hilo % divisor;
    return hilo / divisor;
}

Show compiler warnings [ + ] Compiler args [ + ] Show input
Compilation time: 2.71 sec, absolute running time: 1.43 sec, absolute service time: 4,22 sec