int main() {
__v4sf x;
_mm_loadh_pi(x, 0);
const __v2sf* cx = 0;
__builtin_ia32_loadhps(x, cx);
{
__v4sf a = {1,2,3,4};
__v4sf b = {5,6,7,8};
__v4sf c = {9,10,11,12};
printf4vector(&a);
printf4vector(&b);
printf4vector(&c);
//builtin "ia32_mulps" not supported by clang -- use _mm_mul_ps
__v4sf tmp = _mm_mul_ps(a, b); // a * b
printf4vector(&tmp);
//builtin "ia32_addps" not supported by clang -- use _mm_add_ps
__v4sf e = _mm_add_ps(tmp, c); // e = (a * b) + c
printf4vector(&e);
}
return 0;
}
int main()
{
union f4vector a, b, c;
a.f[0] = 1.1; a.f[1] = 2; a.f[2] = 3; a.f[3] = 4.0;
b.f[0] = 5.0; b.f[1] = 6; b.f[2] = 7; b.f[3] = 8.0;
c.f[0] = 9.1; c.f[1] = 10; c.f[2] = 11; c.f[3] = 12.3;
v4sf tmp = __builtin_ia32_mulps (a.v, b.v);
v4sf foo = __builtin_ia32_addps(tmp, c.v);
std::cout << "calculate foo = (a * b) + c:\n";
std::cout << "a = "; printf4vector(&a.v);
std::cout << "b = "; printf4vector(&b.v);
std::cout << "c = "; printf4vector(&c.v);
std::cout << "foo = "; printf4vector(&foo);
v4si A ={1,2,3,4};
v4si B ={5,6,7,8};
v4si C;
std::cout << "A : "; printi4vector(&A);
std::cout << "B : "; printi4vector(&B);
C = A + B;
std::cout << "C=A+B : "; printi4vector(&C);
C = A - B;
std::cout << "C=A-B : "; printi4vector(&C);
C = B - A;
std::cout << "C=B-A : "; printi4vector(&C);
C = A * B;
std::cout << "C=A*B : "; printi4vector(&C);
C = B / A;
std::cout << "C=B/A : "; printi4vector(&C);
C = A / B;
std::cout << "C=A/B : "; printi4vector(&C);
v4sf fA ={1,2,3,4.5};
v4sf fB ={5,6,7,8.5};
v4sf fC;
std::cout << "fA : "; printf4vector(&fA);
std::cout << "fB : "; printf4vector(&fB);
fC = fA + fB;
std::cout << "fC=fA+fB : "; printf4vector(&fC);
fC = fA - fB;
std::cout << "fC=fA-fB : "; printf4vector(&fC);
fC = fB - fA;
std::cout << "fC=fB-fA : "; printf4vector(&fC);
fC = fA * fB;
std::cout << "fC=fA*fB : "; printf4vector(&fC);
fC = fA * fA;
std::cout << "fC=fA*fA : "; printf4vector(&fC);
fC = fB * fB;
std::cout << "fC=fB*fB : "; printf4vector(&fC);
fC = fA / fB;
std::cout << "fC=fA/fB : "; printf4vector(&fC);
fC = fB / fA;
std::cout << "fC=fB/fA : "; printf4vector(&fC);
/* ... */
extern void vectorcode();
vectorcode(); // seems to execute --- no segfault. How about that?
}
