hadd confusing about mipp HOT 6 OPEN

Hi @flori-uni,

In MIPP we chose that horizontal operations return a scalar (hadd, hmul, etc.). I'm not sure I want to modify this default behavior documented here: https://github.com/aff3ct/MIPP#reductions-horizontal-functions-.
But if you need a vector instead of a scalar, you can write mipp::Reduction<T,mipp::add<T>>::apply(v); (apply instead of sapply) in your code.

from mipp.

Well I plan to use the functionality for calculation on complex numbers. In detail for calculating real²+imaginary².

The real and imaginary part are in the same memory stored alternating. So I have re,im,re,im...

Maybe I'll implement some separated methods for my use-case. Later I probably need also the real and imaginary parts separate...

from mipp.

mipp::Reduction<T,mipp::add>::apply(v); actually does not result in the right data. I think i need to implement it on my own 😢

from mipp.

I'm not sure to understand, what is the result you expect? Could you give me an example?

from mipp.

Using AVX2 I'm able to do the following:

for (auto idx = 0u; idx < max - 3; idx += 4)
{
         auto item = _mm256_load_ps(&data[idx]);
         item = _mm256_mul_ps(item, item );
         auto mag = _mm256_hadd_ps(item, zeros);
         sum = _mm256_add_ps(sum, _mm256_sqrt_ps(mag));
}
auto totalSum = sum[0] + sum[1] + sum[4] + sum[5];       

So it calculates sqrt(x*x+y*y) for each complex pair. Afterwards i can calculate the mean of the sum and some other stuff

from mipp.

    [1,  2,  3,  4,  5,  6,  7,  8]
*   [1,  2,  3,  4,  5,  6,  7,  8]
    [1,  4,  9, 16, 25, 36, 49, 64]
hadd[5,  0, 25,  0, 61,  0,113, 0,]
sqrt[.............................]

from mipp.