-rw-r--r-- 2054 libntruprime-20240825/crypto_encode/1013x3/avx/encode.c raw
#include <immintrin.h>
#include "crypto_encode.h"
#include "crypto_uint8.h"
#define uint8 crypto_uint8

#define p 1013
#define loops 8
#define overshoot 3

static const __attribute((aligned(32))) uint8 lobytes_buf[32] = {
  255,0,255,0,255,0,255,0,
  255,0,255,0,255,0,255,0,
  255,0,255,0,255,0,255,0,
  255,0,255,0,255,0,255,0,
};
#define lobytes (*(__m256i *) lobytes_buf)

void crypto_encode(unsigned char *s,const void *v)
{
  const uint8 *f = v;
  int loop;
  const uint8 *nextf = f+128-4*overshoot;
  unsigned char *nexts = s+32-overshoot;

  for (loop = loops;loop > 0;--loop) {
    __m256i f0 = _mm256_loadu_si256((const __m256i *) (f+0));
    __m256i f1 = _mm256_loadu_si256((const __m256i *) (f+32));
    __m256i f2 = _mm256_loadu_si256((const __m256i *) (f+64));
    __m256i f3 = _mm256_loadu_si256((const __m256i *) (f+96));
    f = nextf;
    nextf += 128;

    __m256i a0 = _mm256_packus_epi16(f0&lobytes,f1&lobytes);
      /* 0 2 4 6 8 10 12 14 32 34 36 38 40 42 44 46 */
      /* 16 18 20 22 24 26 28 30 48 50 52 54 56 58 60 62 */
    __m256i a1 = _mm256_packus_epi16(_mm256_srli_epi16(f0,8),_mm256_srli_epi16(f1,8));
      /* 1 3 ... */
    __m256i a2 = _mm256_packus_epi16(f2&lobytes,f3&lobytes);
    __m256i a3 = _mm256_packus_epi16(_mm256_srli_epi16(f2,8),_mm256_srli_epi16(f3,8));

    a0 = _mm256_add_epi8(a0,_mm256_slli_epi16(a1&_mm256_set1_epi8(63),2));
    a2 = _mm256_add_epi8(a2,_mm256_slli_epi16(a3&_mm256_set1_epi8(63),2));

    __m256i b0 = _mm256_packus_epi16(a0&lobytes,a2&lobytes);
      /* 0 4 8 12 32 36 40 44 64 68 72 76 96 100 104 108 */
      /* 16 20 24 28 48 52 56 60 80 84 88 92 112 116 120 124 */
    __m256i b2 = _mm256_packus_epi16(_mm256_srli_epi16(a0,8),_mm256_srli_epi16(a2,8));
      /* 2 6 ... */

    b0 = _mm256_add_epi8(b0,_mm256_slli_epi16(b2&_mm256_set1_epi8(15),4));

    b0 = _mm256_permutevar8x32_epi32(b0,_mm256_set_epi32(7,3,6,2,5,1,4,0));

    b0 = _mm256_add_epi8(b0,_mm256_set1_epi8(85));

    _mm256_storeu_si256((__m256i *) s,b0);
    s = nexts;
    nexts += 32;
  }

  *s++ = *f++ + 1;
}