pqc/crypto_kem/kyber1024-90s/avx2/aes256ctr.c
John M. Schanck 688ff2fe50 Round 3 Kyber
2020-11-26 21:42:35 -05:00

143 lines
4.6 KiB
C

#include "aes256ctr.h"
#include <immintrin.h>
#include <stddef.h>
#include <stdint.h>
/* Based heavily on public-domain code by Romain Dolbeau
* Different handling of nonce+counter than original version using
* separated 64-bit nonce and internal 64-bit counter, starting from zero
* Public Domain */
static inline void aesni_encrypt4(uint8_t out[64], __m128i *n, const __m128i rkeys[16]) {
__m128i f, f0, f1, f2, f3;
const __m128i idx = _mm_set_epi8(8, 9, 10, 11, 12, 13, 14, 15, 7, 6, 5, 4, 3, 2, 1, 0);
/* Load current counter value */
f = _mm_load_si128(n);
/* Increase counter in 4 consecutive blocks */
f0 = _mm_shuffle_epi8(_mm_add_epi64(f, _mm_set_epi64x(0, 0)), idx);
f1 = _mm_shuffle_epi8(_mm_add_epi64(f, _mm_set_epi64x(1, 0)), idx);
f2 = _mm_shuffle_epi8(_mm_add_epi64(f, _mm_set_epi64x(2, 0)), idx);
f3 = _mm_shuffle_epi8(_mm_add_epi64(f, _mm_set_epi64x(3, 0)), idx);
/* Write counter for next iteration, increased by 4 */
_mm_store_si128(n, _mm_add_epi64(f, _mm_set_epi64x(4, 0)));
/* Actual AES encryption, 4x interleaved */
f = _mm_load_si128(&rkeys[0]);
f0 = _mm_xor_si128(f0, f);
f1 = _mm_xor_si128(f1, f);
f2 = _mm_xor_si128(f2, f);
f3 = _mm_xor_si128(f3, f);
for (int i = 1; i < 14; i++) {
f = _mm_load_si128(&rkeys[i]);
f0 = _mm_aesenc_si128(f0, f);
f1 = _mm_aesenc_si128(f1, f);
f2 = _mm_aesenc_si128(f2, f);
f3 = _mm_aesenc_si128(f3, f);
}
f = _mm_load_si128(&rkeys[14]);
f0 = _mm_aesenclast_si128(f0, f);
f1 = _mm_aesenclast_si128(f1, f);
f2 = _mm_aesenclast_si128(f2, f);
f3 = _mm_aesenclast_si128(f3, f);
/* Write results */
_mm_storeu_si128((__m128i *)(out + 0), f0);
_mm_storeu_si128((__m128i *)(out + 16), f1);
_mm_storeu_si128((__m128i *)(out + 32), f2);
_mm_storeu_si128((__m128i *)(out + 48), f3);
}
void PQCLEAN_KYBER102490S_AVX2_aes256ctr_init(aes256ctr_ctx *state, const uint8_t key[32], uint64_t nonce) {
__m128i key0, key1, temp0, temp1, temp2, temp4;
int idx = 0;
key0 = _mm_loadu_si128((__m128i *)(key + 0));
key1 = _mm_loadu_si128((__m128i *)(key + 16));
state->n = _mm_loadl_epi64((__m128i *)&nonce);
state->rkeys[idx++] = key0;
temp0 = key0;
temp2 = key1;
temp4 = _mm_setzero_si128();
#define BLOCK1(IMM) \
temp1 = _mm_aeskeygenassist_si128(temp2, IMM); \
state->rkeys[idx++] = temp2; \
temp4 = (__m128i)_mm_shuffle_ps((__m128)temp4, (__m128)temp0, 0x10); \
temp0 = _mm_xor_si128(temp0, temp4); \
temp4 = (__m128i)_mm_shuffle_ps((__m128)temp4, (__m128)temp0, 0x8c); \
temp0 = _mm_xor_si128(temp0, temp4); \
temp1 = (__m128i)_mm_shuffle_ps((__m128)temp1, (__m128)temp1, 0xff); \
temp0 = _mm_xor_si128(temp0, temp1)
#define BLOCK2(IMM) \
temp1 = _mm_aeskeygenassist_si128(temp0, IMM); \
state->rkeys[idx++] = temp0; \
temp4 = (__m128i)_mm_shuffle_ps((__m128)temp4, (__m128)temp2, 0x10); \
temp2 = _mm_xor_si128(temp2, temp4); \
temp4 = (__m128i)_mm_shuffle_ps((__m128)temp4, (__m128)temp2, 0x8c); \
temp2 = _mm_xor_si128(temp2, temp4); \
temp1 = (__m128i)_mm_shuffle_ps((__m128)temp1, (__m128)temp1, 0xaa); \
temp2 = _mm_xor_si128(temp2, temp1)
BLOCK1(0x01);
BLOCK2(0x01);
BLOCK1(0x02);
BLOCK2(0x02);
BLOCK1(0x04);
BLOCK2(0x04);
BLOCK1(0x08);
BLOCK2(0x08);
BLOCK1(0x10);
BLOCK2(0x10);
BLOCK1(0x20);
BLOCK2(0x20);
BLOCK1(0x40);
state->rkeys[idx++] = temp0;
}
void PQCLEAN_KYBER102490S_AVX2_aes256ctr_squeezeblocks(uint8_t *out,
size_t nblocks,
aes256ctr_ctx *state) {
size_t i;
for (i = 0; i < nblocks; i++) {
aesni_encrypt4(out, &state->n, state->rkeys);
out += 64;
}
}
void PQCLEAN_KYBER102490S_AVX2_aes256ctr_prf(uint8_t *out,
size_t outlen,
const uint8_t key[32],
uint64_t nonce) {
unsigned int i;
uint8_t buf[64];
aes256ctr_ctx state;
PQCLEAN_KYBER102490S_AVX2_aes256ctr_init(&state, key, nonce);
while (outlen >= 64) {
aesni_encrypt4(out, &state.n, state.rkeys);
outlen -= 64;
out += 64;
}
if (outlen) {
aesni_encrypt4(buf, &state.n, state.rkeys);
for (i = 0; i < outlen; i++) {
out[i] = buf[i];
}
}
}