native.intel.cfb.cfb512.c Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of bcprov-lts8on Show documentation
Show all versions of bcprov-lts8on Show documentation
The Long Term Stable (LTS) Bouncy Castle Crypto package is a Java implementation of cryptographic algorithms. This jar contains the JCA/JCE provider and low-level API for the BC LTS version 2.73.1 for Java 8 and later.
//
//
#include
#include "immintrin.h"
#include "cfb.h"
static inline void aes256_trailing(__m256i *d0, __m128i *feedback, __m128i *roundKeys, const uint32_t blocks,
const uint32_t max_rounds) {
__m256i tmp0;
tmp0 = _mm256_set_m128i(_mm256_extracti128_si256(*d0, 0), *feedback);
*feedback = _mm256_extracti128_si256(*d0, 1);
tmp0 = _mm256_xor_si256(tmp0, _mm256_broadcastsi128_si256(roundKeys[0]));
int rounds;
for (rounds = 1; rounds < max_rounds; rounds++) {
tmp0 = _mm256_aesenc_epi128(tmp0, _mm256_broadcastsi128_si256(roundKeys[rounds]));
}
tmp0 = _mm256_aesenclast_epi128(tmp0, _mm256_broadcastsi128_si256(roundKeys[rounds]));
*d0 = _mm256_xor_si256(*d0, tmp0);
}
static inline void aes128_trailing(
__m128i *d0,
__m128i *feedback, __m128i *roundKeys, const uint32_t blocks,
const uint32_t max_rounds) {
__m128i tmp0;
tmp0 = _mm_xor_si128(*feedback, roundKeys[0]);
*feedback = *d0;
int rounds;
for (rounds = 1; rounds < max_rounds; rounds++) {
tmp0 = _mm_aesenc_si128(tmp0, roundKeys[rounds]);
}
tmp0 = _mm_aesenclast_si128(tmp0, roundKeys[rounds]);
*d0 = _mm_xor_si128(*d0, tmp0);
}
static inline __m128i set_feedback(const __m512i in_cipher_blocks, const uint32_t num_blocks) {
if (num_blocks == 1) {
return _mm512_castsi512_si128(in_cipher_blocks);
} else if (num_blocks == 2) {
return _mm512_extracti32x4_epi32(in_cipher_blocks, 1);
} else if (num_blocks == 3) {
return _mm512_extracti32x4_epi32(in_cipher_blocks, 2);
}
return _mm512_extracti32x4_epi32(in_cipher_blocks, 3);
}
/**
*
* @param d0 in / out
* @param d1 in / out
* @param d2 in / out
* @param d3 in / out
* @param d4 in / out
* @param feedback the chainblock
* @param roundKeys
* @param blocks blocks must be even number 16 to 2, 1 does odd block processing, any other value does nothing.
* @param max_rounds The maximum number of rounds.
*
*/
static inline void aes512w_cfb_decrypt(
__m512i *d0, __m512i *d1, __m512i *d2, __m512i *d3,
__m128i *feedback, __m128i *roundKeys, const uint32_t blocks,
const uint32_t max_rounds) {
__m512i tmp0, tmp1, tmp2, tmp3;
if (blocks >= 16) {
tmp0 = _mm512_alignr_epi64(*d0, _mm512_broadcast_i32x4(*feedback), 6);
tmp1 = _mm512_alignr_epi64(*d1, *d0, 6);
tmp2 = _mm512_alignr_epi64(*d2, *d1, 6);
tmp3 = _mm512_alignr_epi64(*d3, *d2, 6);
*feedback = _mm512_extracti32x4_epi32(*d3, 3);
tmp0 = _mm512_xor_si512(tmp0, _mm512_broadcast_i32x4(roundKeys[0]));
tmp1 = _mm512_xor_si512(tmp1, _mm512_broadcast_i32x4(roundKeys[0]));
tmp2 = _mm512_xor_si512(tmp2, _mm512_broadcast_i32x4(roundKeys[0]));
tmp3 = _mm512_xor_si512(tmp3, _mm512_broadcast_i32x4(roundKeys[0]));
int rounds;
for (rounds = 1; rounds < max_rounds; rounds++) {
tmp0 = _mm512_aesenc_epi128(tmp0, _mm512_broadcast_i32x4(roundKeys[rounds]));
tmp1 = _mm512_aesenc_epi128(tmp1, _mm512_broadcast_i32x4(roundKeys[rounds]));
tmp2 = _mm512_aesenc_epi128(tmp2, _mm512_broadcast_i32x4(roundKeys[rounds]));
tmp3 = _mm512_aesenc_epi128(tmp3, _mm512_broadcast_i32x4(roundKeys[rounds]));
}
tmp0 = _mm512_aesenclast_epi128(tmp0, _mm512_broadcast_i32x4(roundKeys[rounds]));
tmp1 = _mm512_aesenclast_epi128(tmp1, _mm512_broadcast_i32x4(roundKeys[rounds]));
tmp2 = _mm512_aesenclast_epi128(tmp2, _mm512_broadcast_i32x4(roundKeys[rounds]));
tmp3 = _mm512_aesenclast_epi128(tmp3, _mm512_broadcast_i32x4(roundKeys[rounds]));
*d0 = _mm512_xor_si512(*d0, tmp0);
*d1 = _mm512_xor_si512(*d1, tmp1);
*d2 = _mm512_xor_si512(*d2, tmp2);
*d3 = _mm512_xor_si512(*d3, tmp3);
} else if (blocks >= 12) {
const uint32_t partial_blocks = blocks - 12;
tmp0 = _mm512_alignr_epi64(*d0, _mm512_broadcast_i32x4(*feedback), 6);
tmp1 = _mm512_alignr_epi64(*d1, *d0, 6);
tmp2 = _mm512_alignr_epi64(*d2, *d1, 6);
*feedback = set_feedback(*d2, partial_blocks);
tmp0 = _mm512_xor_si512(tmp0, _mm512_broadcast_i32x4(roundKeys[0]));
tmp1 = _mm512_xor_si512(tmp1, _mm512_broadcast_i32x4(roundKeys[0]));
tmp2 = _mm512_xor_si512(tmp2, _mm512_broadcast_i32x4(roundKeys[0]));
int rounds;
for (rounds = 1; rounds < max_rounds; rounds++) {
tmp0 = _mm512_aesenc_epi128(tmp0, _mm512_broadcast_i32x4(roundKeys[rounds]));
tmp1 = _mm512_aesenc_epi128(tmp1, _mm512_broadcast_i32x4(roundKeys[rounds]));
tmp2 = _mm512_aesenc_epi128(tmp2, _mm512_broadcast_i32x4(roundKeys[rounds]));
}
tmp0 = _mm512_aesenclast_epi128(tmp0, _mm512_broadcast_i32x4(roundKeys[rounds]));
tmp1 = _mm512_aesenclast_epi128(tmp1, _mm512_broadcast_i32x4(roundKeys[rounds]));
tmp2 = _mm512_aesenclast_epi128(tmp2, _mm512_broadcast_i32x4(roundKeys[rounds]));
*d0 = _mm512_xor_si512(*d0, tmp0);
*d1 = _mm512_xor_si512(*d1, tmp1);
*d2 = _mm512_xor_si512(*d2, tmp2);
} else if (blocks >= 8) {
const uint32_t partial_blocks = blocks - 8;
tmp0 = _mm512_alignr_epi64(*d0, _mm512_broadcast_i32x4(*feedback), 6);
tmp1 = _mm512_alignr_epi64(*d1, *d0, 6);
*feedback = set_feedback(*d1, partial_blocks);
tmp0 = _mm512_xor_si512(tmp0, _mm512_broadcast_i32x4(roundKeys[0]));
tmp1 = _mm512_xor_si512(tmp1, _mm512_broadcast_i32x4(roundKeys[0]));
int rounds;
for (rounds = 1; rounds < max_rounds; rounds++) {
tmp0 = _mm512_aesenc_epi128(tmp0, _mm512_broadcast_i32x4(roundKeys[rounds]));
tmp1 = _mm512_aesenc_epi128(tmp1, _mm512_broadcast_i32x4(roundKeys[rounds]));
}
tmp0 = _mm512_aesenclast_epi128(tmp0, _mm512_broadcast_i32x4(roundKeys[rounds]));
tmp1 = _mm512_aesenclast_epi128(tmp1, _mm512_broadcast_i32x4(roundKeys[rounds]));
*d0 = _mm512_xor_si512(*d0, tmp0);
*d1 = _mm512_xor_si512(*d1, tmp1);
} else {
const uint32_t partial_blocks = blocks - 4;
tmp0 = _mm512_alignr_epi64(*d0, _mm512_broadcast_i32x4(*feedback), 6);
*feedback = set_feedback(*d0, partial_blocks);
tmp0 = _mm512_xor_si512(tmp0, _mm512_broadcast_i32x4(roundKeys[0]));
int rounds;
for (rounds = 1; rounds < max_rounds; rounds++) {
tmp0 = _mm512_aesenc_epi128(tmp0, _mm512_broadcast_i32x4(roundKeys[rounds]));
}
tmp0 = _mm512_aesenclast_epi128(tmp0, _mm512_broadcast_i32x4(roundKeys[rounds]));
*d0 = _mm512_xor_si512(*d0, tmp0);
}
}
size_t cfb_decrypt(cfb_ctx *cfb, uint8_t *src, size_t len, unsigned char *dest) {
unsigned char *destStart = dest;
//
// Round out buffer.
//
while (cfb->buf_index > 0 && len > 0) {
*dest = cfb_decrypt_byte(cfb, *src);
len--;
dest++;
src++;
}
while (len >= 16 * CFB_BLOCK_SIZE) {
__m512i d0 = _mm512_loadu_si512((__m512i *) &src[0 * 64]);
__m512i d1 = _mm512_loadu_si512((__m512i *) &src[1 * 64]);
__m512i d2 = _mm512_loadu_si512((__m512i *) &src[2 * 64]);
__m512i d3 = _mm512_loadu_si512((__m512i *) &src[3 * 64]);
aes512w_cfb_decrypt(&d0, &d1, &d2, &d3, &cfb->feedback, cfb->roundKeys, 16, cfb->num_rounds);
_mm512_storeu_si512((__m512i *) &dest[0 * 64], d0);
_mm512_storeu_si512((__m512i *) &dest[1 * 64], d1);
_mm512_storeu_si512((__m512i *) &dest[2 * 64], d2);
_mm512_storeu_si512((__m512i *) &dest[3 * 64], d3);
len -= 16 * CFB_BLOCK_SIZE;
src += 16 * CFB_BLOCK_SIZE;
dest += 16 * CFB_BLOCK_SIZE;
}
while (len >= CFB_BLOCK_SIZE) {
if (len >= 12 * CFB_BLOCK_SIZE) {
__m512i d0 = _mm512_loadu_si512((__m512i *) &src[0 * 64]);
__m512i d1 = _mm512_loadu_si512((__m512i *) &src[1 * 64]);
__m512i d2 = _mm512_loadu_si512((__m512i *) &src[2 * 64]);
aes512w_cfb_decrypt(&d0, &d1, &d2, &d2, &cfb->feedback, cfb->roundKeys, 12, cfb->num_rounds);
_mm512_storeu_si512((__m512i *) &dest[0 * 64], d0);
_mm512_storeu_si512((__m512i *) &dest[1 * 64], d1);
_mm512_storeu_si512((__m512i *) &dest[2 * 64], d2);
len -= 12 * CFB_BLOCK_SIZE;
src += 12 * CFB_BLOCK_SIZE;
dest += 12 * CFB_BLOCK_SIZE;
} else if (len >= 8 * CFB_BLOCK_SIZE) {
__m512i d0 = _mm512_loadu_si512((__m512i *) &src[0 * 64]);
__m512i d1 = _mm512_loadu_si512((__m512i *) &src[1 * 64]);
aes512w_cfb_decrypt(&d0, &d1, &d1, &d1, &cfb->feedback, cfb->roundKeys, 8, cfb->num_rounds);
_mm512_storeu_si512((__m512i *) &dest[0 * 64], d0);
_mm512_storeu_si512((__m512i *) &dest[1 * 64], d1);
len -= 8 * CFB_BLOCK_SIZE;
src += 8 * CFB_BLOCK_SIZE;
dest += 8 * CFB_BLOCK_SIZE;
} else if (len >= 4 * CFB_BLOCK_SIZE) {
__m512i d0 = _mm512_loadu_si512((__m512i *) &src[0 * 64]);
aes512w_cfb_decrypt(&d0, &d0, &d0, &d0, &cfb->feedback, cfb->roundKeys, 4, cfb->num_rounds);
_mm512_storeu_si512((__m512i *) &dest[0 * 64], d0);
len -= 4 * CFB_BLOCK_SIZE;
src += 4 * CFB_BLOCK_SIZE;
dest += 4 * CFB_BLOCK_SIZE;
} else if (len >= 2 * CFB_BLOCK_SIZE) {
__m256i d0 = _mm256_loadu_si256((__m256i *) &src[0 * 32]);
aes256_trailing(&d0, &cfb->feedback, cfb->roundKeys, 2,
cfb->num_rounds);
_mm256_storeu_si256((__m256i *) &dest[0 * 32], d0);
len -= 2 * CFB_BLOCK_SIZE;
src += 2 * CFB_BLOCK_SIZE;
dest += 2 * CFB_BLOCK_SIZE;
} else { // single block
__m128i d0 = _mm_loadu_si128((__m128i *) &src[0 * 16]);
aes128_trailing(&d0, &cfb->feedback, cfb->roundKeys, 1,
cfb->num_rounds);
_mm_storeu_si128(((__m128i *) &dest[0 * 16]), d0);
len -= CFB_BLOCK_SIZE;
src += CFB_BLOCK_SIZE;
dest += CFB_BLOCK_SIZE;
}
}
//
// load any trailing bytes into the buffer, the expectation is that
// whatever is passed in has to be decrypted, ideally callers will
// try and stick to the AES block size for as long as possible.
//
while (len > 0) {
*dest = cfb_decrypt_byte(cfb, *src);
len--;
dest++;
src++;
}
return (size_t) (dest - destStart);
}
unsigned char cfb_decrypt_byte(cfb_ctx *cfbCtx, unsigned char b) {
if (cfbCtx->buf_index == 0) {
// We need to generate a new encrypted feedback block to xor into the data.
cfbCtx->mask = _mm_xor_si128(cfbCtx->feedback, cfbCtx->roundKeys[0]);
int j;
for (j = 1; j < cfbCtx->num_rounds; j++) {
cfbCtx->mask = _mm_aesenc_si128(cfbCtx->mask, cfbCtx->roundKeys[j]);
}
cfbCtx->mask = _mm_aesenclast_si128(cfbCtx->mask, cfbCtx->roundKeys[j]);
}
//
// incrementally mask becomes the last block of cipher text
//
unsigned char pt = ((unsigned char *) &cfbCtx->mask)[cfbCtx->buf_index] ^ b;
((unsigned char *) &cfbCtx->mask)[cfbCtx->buf_index++] = b; // Mask fills with last cipher text directly.
if (cfbCtx->buf_index == CFB_BLOCK_SIZE) {
cfbCtx->buf_index = 0;
cfbCtx->feedback = cfbCtx->mask;
}
return pt;
}