1
1
mirror of https://github.com/henrydcase/pqc.git synced 2024-11-26 17:31:38 +00:00
pqcrypto/crypto_kem/kyber1024/avx2/verify.c
John M. Schanck 127e9ec326 Round 3 Kyber
2021-03-24 21:02:49 +00:00

74 lines
2.3 KiB
C

#include "verify.h"
#include <immintrin.h>
#include <stdint.h>
#include <stdlib.h>
/*************************************************
* Name: PQCLEAN_KYBER1024_AVX2_verify
*
* Description: Compare two arrays for equality in constant time.
*
* Arguments: const uint8_t *a: pointer to first byte array
* const uint8_t *b: pointer to second byte array
* size_t len: length of the byte arrays
*
* Returns 0 if the byte arrays are equal, 1 otherwise
**************************************************/
int PQCLEAN_KYBER1024_AVX2_verify(const uint8_t *a, const uint8_t *b, size_t len) {
size_t i;
uint64_t r;
__m256i f, g, h;
h = _mm256_setzero_si256();
for (i = 0; i < len / 32; i++) {
f = _mm256_loadu_si256((__m256i *)&a[32 * i]);
g = _mm256_loadu_si256((__m256i *)&b[32 * i]);
f = _mm256_xor_si256(f, g);
h = _mm256_or_si256(h, f);
}
r = 1 - _mm256_testz_si256(h, h);
a += 32 * i;
b += 32 * i;
len -= 32 * i;
for (i = 0; i < len; i++) {
r |= a[i] ^ b[i];
}
r = (-r) >> 63;
return r;
}
/*************************************************
* Name: PQCLEAN_KYBER1024_AVX2_cmov
*
* Description: Copy len bytes from x to r if b is 1;
* don't modify x if b is 0. Requires b to be in {0,1};
* assumes two's complement representation of negative integers.
* Runs in constant time.
*
* Arguments: unsigned char *r: pointer to output byte array
* const unsigned char *x: pointer to input byte array
* size_t len: Amount of bytes to be copied
* unsigned char b: Condition bit; has to be in {0,1}
**************************************************/
void PQCLEAN_KYBER1024_AVX2_cmov(uint8_t *restrict r, const uint8_t *x, size_t len, uint8_t b) {
size_t i;
__m256i xvec, rvec, bvec;
bvec = _mm256_set1_epi64x(-(uint64_t)b);
for (i = 0; i < len / 32; i++) {
rvec = _mm256_loadu_si256((__m256i *)&r[32 * i]);
xvec = _mm256_loadu_si256((__m256i *)&x[32 * i]);
rvec = _mm256_blendv_epi8(rvec, xvec, bvec);
_mm256_storeu_si256((__m256i *)&r[32 * i], rvec);
}
r += 32 * i;
x += 32 * i;
len -= 32 * i;
for (i = 0; i < len; i++) {
r[i] ^= -b & (x[i] ^ r[i]);
}
}