kste / skinny_avx Goto Github PK

In skinny64/avx2/skinny64.c, there is a definition of tmp:

u256 rc, tmp[3];

but in the line 71 and line 83, there is tmp[3] = x[15]; and x[3] = XOR(tmp[3], x[15]);，which pass the end of array tmp[3].

I think define tmp as u256 tmp[4] is right.

Hello,

My name is Robbe, I'm currently writing a bit-sliced SIMD implementation on ForkSKINNY primitive. I'm using this implementation as a benchmark and have a question.

As far as I understand the implementation, you slice the nibbles of 64 cipher states in a single __m256i SIMD variable, correct? So for example, the u256 x[16] parameter entering the rounds here:

https://github.com/kste/skinny_avx/blob/aed96f5d1560a27a6f1e77807e9709df575e9451/skinny64/avx2/skinny64.c#LL15C1-L15C1

Is an array of 16 256-bit values, where every value in this array contains 64 equally-significant nibbles (thus 4*64 = 256 bits), correct? Then the round operations are performed between multiple such 256-bit slices. If not, how or where is the slice and unslice operation performed?

Thank you so much,

Robbe