#include <stdio.h>

typedef struct vector
{
    float s1, s2, s3, s4;
} vector;

int main()
{

    vector a, b, c, d, result1, result2, resultCheck;

    // zmienic wartosci!!!
    // TODO!!

    a.s1 = 11;
    a.s2 = 12;
    a.s3 = 13;
    a.s4 = 14;
    b.s1 = 1;
    b.s2 = 2;
    b.s3 = 3;
    b.s4 = 4;
    c.s1 = 17;
    c.s2 = -8;
    c.s3 = -11;
    c.s4 = -2.4;
    d.s1 = 6.6;
    d.s2 = -1;
    d.s3 = 0;
    d.s4 = 0;

   // do sprawdzania
    // asm(
    //     "movups %1, %%xmm0\n\t"
    //     "movups %2, %%xmm1\n\t"
    //     "shufps $0x21, %%xmm0, %%xmm0\n\t"
    //     "movups %%xmm0, %0\n"
    //     : "=g"(resultCheck) //%0
    //     : "g"(a), "g"(b), "g"(c), "g"(d), "g"(d)
    //     : //%1       %2      %3          %4      %5
    // );

    // printf("Wektor %f %f %f %f",resultCheck.s1, resultCheck.s2, resultCheck.s3, resultCheck.s4);

    asm(
        // wektor A i B
        // 1 Nawias lewe mnozenie
        "movups %1, %%xmm0\n\t"
        "movups %2, %%xmm1\n\t"
        // 00 01 10 11
        "shufps $0x21, %%xmm0, %%xmm0\n\t"
        // 01 00 11 10
        "shufps $0x4E, %%xmm1, %%xmm1\n\t"
        "mulps %%xmm1, %%xmm0\n\t"
        "movaps %%xmm0, %%xmm4\n\t" //wynik do [4]
        // ----
        // 1 Nawias prawe mnozenie
        "movups %1, %%xmm0\n\t"
        "movups %2, %%xmm1\n\t"
        // 01 00 11 10
        "shufps $0x4E, %%xmm0, %%xmm0\n\t"
        // 00 01 10 11
        "shufps $0x21, %%xmm1, %%xmm1\n\t"
        "mulps %%xmm1, %%xmm0\n\t"
        // Odejmowanie Lewe - Prawe
        "subps %%xmm0, %%xmm4\n\t" // xmm4 - xmm0, wynik do [4]

        // wektor C i D
        // 1 Nawias lewe mnozenie
        "movups %3, %%xmm0\n\t"
        "movups %4, %%xmm1\n\t"
        // 10 01 01 00
        "shufps $0x94, %%xmm0, %%xmm0\n\t"
        // 11 11 10 11
        "shufps $0xFB, %%xmm1, %%xmm1\n\t"
        "mulps %%xmm1, %%xmm0\n\t"
        "movaps %%xmm0, %%xmm3\n\t" //wynik do [3]
        // 1 Nawias prawe mnozenie
        "movups %3, %%xmm0\n\t"
        "movups %4, %%xmm1\n\t"
        // 11 11 10 11
        "shufps $0xFB, %%xmm0, %%xmm0\n\t"
        // 10 01 01 00
        "shufps $0x94, %%xmm1, %%xmm1\n\t"
        "mulps %%xmm1, %%xmm0\n\t"
        // Odejmowanie Lewe -
        "subps %%xmm0, %%xmm3" // xmm4 - xmm0, wynik do [4]


        "movups %%xmm0, %0\n"
        : "=g"(resultCheck) //%0
        : "g"(a), "g"(b), "g"(c), "g"(d), "g"(d)
        : //%1      %2      %3          %4      %5
    );

    return 0;
}