my SSE code slower than SISD! Why?

[orbano]

Hello!

I would like to ask somebody to tell me, why is my SSE code for vector and matrix operations is slower. I must do something wrong, but dont know what. My vector addition for example:

data type:

typedef union
{
__m128 data;
float elements[4];
} vector4d;

with SSE (using intrinsics):

inline vector4dadd(vector4d a, vector4d b)
{
vector4d c;
c.data = _mm_add_ps(a.data, b.data);
return c;
};

without SSE:

inline vector4d add_sisd(vector4d a, vector4d b)
{
return set(a.elements[0]+b.elements[0],a.elements[1]+b.elements[1],a.elements[2]+b.elements[2],0);
};

(set() simply returnsa vector4d with theparameter values)

[TimP]

Without more information, we're simply guessing. For example, you might be performing scalar operations on the components before or after. Remember also that the compiler would unroll at least by 8 in order to use mm_add_ps in pairs, when performing automatic vectorization. We don't even know which compiler you're using.

[orbano]

No other operations are performed. i just put each function into a loop and measured the time the computation takes.

First i tried in debug mode, without any optimization. The sisd's time was the 50% of the simd version

With all optimizations on, it was 30% (3 times faster!!!!)

I dont know the compiler version. I have Visual C++ 6.0 with service pack 5 and Processor Pack (for SP5), with default compiler settings.