array from C in Assembly

Question

I'm trying to do some adding with SSE and I'm using to this C with assembly. Why something like this doesn't work?

struct vector {
    float x1, x2, x3, x4;
};

struct vector *dodawanie(const struct vector v1[], const struct vector v2[], int size) {

struct vector vec[size];
int i;
for(i = 0; i < size; i++) {
        asm(
            "MOV %1, %%rax \n"
            "MOV %2, %%rdx \n"

            "MOVUPS (%%rax), %%xmm0 \n"
            "MOVUPS (%%rdx), %%xmm1 \n"
            "ADDPS %%xmm0, %%xmm1 \n"

            "MOVUPS %%xmm1, %0 \n"

            :"=g"(vec[i])       //wyjscie
            :"g"(v1[i]), "g"(v2[i]) //wejscie
            :"%rax", "%rdx"
        );
}
return vec;
}

I got error: Thread 1: EXC_BAD_ACCESS (CODE = EXC_I386_GPFLT)

But when instead of v1[i], v2[i] I put v1, v2 etc. this work correctly but of course only with the first element of array.

What's wrong in my code?

1. take a look at generated asm it may happen that %1 and %2 are being passed by rax/rdx. 2. try to pass %1 and %2 directly by rax and rdx. Look here: gcc.gnu.org/onlinedocs/gcc/… for x86 specific constraints — Michał Walenciak
– Michał Walenciak, Commented May 27, 2016 at 20:59
First I would be inclined to use compiler intrinsics for addps. If you are intent on using inline assembler I'd follow Michal's advice of looking at getting the assembler template to do most of the work. Assuming v1, v2, and vec are all arrays of vectors (__m128) then something like this may work asm( "MOVAPS %[v1], %[out]\n\t" "ADDPS %[v2], %[out]\n\t" :[out]"=&x,m"(vec[i]) :[v1]"mx,x"(v1[i]), [v2]"mx,x"(v2[i]) ); — Michael Petch
– Michael Petch, Commented May 27, 2016 at 23:50
Your code appears to be returning a pointer to a local variable which could cause some unexpected behavior since you are relying on the stack not being trashed after the function returns and before the data is used. — Michael Petch
– Michael Petch, Commented May 28, 2016 at 18:30

Michał Walenciak · Accepted Answer · 2016-05-27 21:04:34Z

3

You are using values from tables ( v1[i], v2[i] ) and you treat them as adresses ("MOVUPS (%%rax), %%xmm0 \n"). Use &v1[i] and &v2[i] respectively.

This is also why form v1 and v2 works as in such case address is being passed.

answered May 27, 2016 at 21:04

Michał Walenciak

4,4094 gold badges38 silver badges66 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

demoo Over a year ago

Thanks a lot, it's obviously and I didn't see this. Thanks to u I save a couple of minutes.

Collectives™ on Stack Overflow

array from C in Assembly

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related