Skip to content

README.md

Latest commit

 

History

History
45 lines (35 loc) · 1.25 KB

README.md

File metadata and controls

45 lines (35 loc) · 1.25 KB

The cryptanalysislib provides a very easy to use and strong Single-Instruction Multiple-Data abstraction layer. This layer is currently covering all amd64, arm-neon, riscv-v ISAs.

Usage:

The following pictures the main concepts:

#include <cryptanalysislib/simd/simd.h>
using S = TxN_t<uint32_t, 8>;

int main() {
    S t1 = S::set1(0), 
      t2 = S::setr(0,1,2,3,4,5,6,7);

    const auto t3 =  t1 + t2;
    return t3[4];
}

TxN_t is a class that wraps N times the type T in a single container. Currently only unsinged/signed with 8,16,32,64 bits, integers are supported This container provides all operations like:

  • Addition, Subtraction, Multiplication, Division
  • Scatter/Table lookups, Permutations
  • Memory stores/loads

Additionally, for the 3 main architectures x86, arm and neon and their vector extensions the following specialisations are provides which wrap:

  • x86: __m128i, __m256i and __m512i

  • arm-neon: uint8x16, ...

  • T=uint8_t: uint8x16_t, uint8x32_t

  • T=uint16_t: uint16x8_t, uint16x16_t

  • T=uint32_t: uint32x4_t, uint32x8_t

  • T=uint64_t: uint64x2_t, uint64x4_t

TODO:

  • riscv backeend
  • uint16x8, uint8x16_t in neon, avx
  • cvtepu8 in neon und noch mehr zero extend