Assembly 32bits sum of elements of two vectors

user_16255922 0
Viewed 81

I have a problem with sum of elements of two vectors type double which are the same size. Code always returns 0.

#include <iostream>
using namespace std;
int main()
{
    int n = 5;
    double* tab = new double[n];
    double* tab3 = new double[n];
    for (size_t i = 0; i < n; i++)
    {
        tab[i] = 1;
        tab3[i] = 1;
    }
    double sum;
    __asm {
        mov eax, n; //vector size
        mov edi, tab; //first vector
        mov esi, tab3; //second vector
        fldz;
    l:
        fadd[edi + 8 * eax - 8];
        fadd[esi + 8 * eax - 8];
        dec eax;
        jnz l;
        fstp sum;
    }
    cout << sum;
}
x86 assembly visual-c++ windows c++
2 Answers

Sadly i am not on windows, so i had to modify the code to use g++ instead of msvc, but i used intel syntax assembly too. During debugging it turned out that fadd instructions had no effect. I fixed it by adding qword ptr before the [edi + 8 * eax - 8] and [esi + 8 * eax - 8] to tell assembler to use pointers to an 8 byte value (since you are using double instead of float):

fadd qword ptr [edi + 8 * eax - 8];
fadd qword ptr [esi + 8 * eax - 8];
user_54574260

So you are looking for asm for this bit of C code, right? https://godbolt.org/z/vbdfEb94s

#include <cstddef>

double add(double *a, double *b, std::size_t len) {
    double sum = 0;
    while (len-- > 0) {
        sum += *a++;
        sum += *b++;
    }
    return sum;
}

I (meaning gcc) come up with this code for 64bit:

add(double*, double*, unsigned long):
        xor     eax, eax
        xorps   xmm0, xmm0
.L3:
        cmp     rdx, rax
        je      .L1
        addsd   xmm0, QWORD PTR [rdi+rax*8]
        addsd   xmm0, QWORD PTR [rsi+rax*8]
        inc     rax
        jmp     .L3
.L1:
        ret

and this for 32bit i386:

add(double*, double*, unsigned int):
        push    ebp
        xor     eax, eax
        fldz
        mov     ebp, esp
        mov     ecx, DWORD PTR [ebp+8]
        mov     edx, DWORD PTR [ebp+12]
.L3:
        cmp     DWORD PTR [ebp+16], eax
        je      .L1
        fadd    QWORD PTR [ecx+eax*8]
        fadd    QWORD PTR [edx+eax*8]
        inc     eax
        jmp     .L3
.L1:
        pop     ebp
        ret
user_42623440
Related