kris/pqc
1
1
Fork 0
Code Issues Pull Requests Releases Wiki Activity

dilithium/fips202x4: align _mm_storel_pd result using double on stack

Browse Source
This commit is contained in:
John M. Schanck 2021-02-09 11:07:00 -05:00
parent 52750b86b1
commit 52851284ab
3 changed files with 27 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
crypto_sign/dilithium2/avx2/fips202x4.c
View File

@ -91,17 +91,22 @@ static void keccakx4_squeezeblocks(uint8_t *out0,
unsigned int r, unsigned int r,
__m256i s[25]) { __m256i s[25]) {
unsigned int i; unsigned int i;
double temp0, temp1;
__m128d t; __m128d t;
while (nblocks > 0) { while (nblocks > 0) {
PQCLEAN_DILITHIUM2_AVX2_f1600x4(s, KeccakF_RoundConstants); PQCLEAN_DILITHIUM2_AVX2_f1600x4(s, KeccakF_RoundConstants);
for (i = 0; i < r / 8; ++i) { for (i = 0; i < r / 8; ++i) {
t = _mm_castsi128_pd(_mm256_castsi256_si128(s[i])); t = _mm_castsi128_pd(_mm256_castsi256_si128(s[i]));
_mm_storel_pd((double *)&out0[8 * i], t); _mm_storel_pd(&temp0, t);
_mm_storeh_pd((double *)&out1[8 * i], t); _mm_storeh_pd(&temp1, t);
memmove(&out0[8 * i], &temp0, sizeof(double));
memmove(&out1[8 * i], &temp1, sizeof(double));
t = _mm_castsi128_pd(_mm256_extracti128_si256(s[i], 1)); t = _mm_castsi128_pd(_mm256_extracti128_si256(s[i], 1));
_mm_storel_pd((double *)&out2[8 * i], t); _mm_storel_pd(&temp0, t);
_mm_storeh_pd((double *)&out3[8 * i], t); _mm_storeh_pd(&temp1, t);
memmove(&out2[8 * i], &temp0, sizeof(double));
memmove(&out3[8 * i], &temp1, sizeof(double));
} }
out0 += r; out0 += r;

13
crypto_sign/dilithium3/avx2/fips202x4.c
View File

@ -91,17 +91,22 @@ static void keccakx4_squeezeblocks(uint8_t *out0,
unsigned int r, unsigned int r,
__m256i s[25]) { __m256i s[25]) {
unsigned int i; unsigned int i;
double temp0, temp1;
__m128d t; __m128d t;
while (nblocks > 0) { while (nblocks > 0) {
PQCLEAN_DILITHIUM3_AVX2_f1600x4(s, KeccakF_RoundConstants); PQCLEAN_DILITHIUM3_AVX2_f1600x4(s, KeccakF_RoundConstants);
for (i = 0; i < r / 8; ++i) { for (i = 0; i < r / 8; ++i) {
t = _mm_castsi128_pd(_mm256_castsi256_si128(s[i])); t = _mm_castsi128_pd(_mm256_castsi256_si128(s[i]));
_mm_storel_pd((double *)&out0[8 * i], t); _mm_storel_pd(&temp0, t);
_mm_storeh_pd((double *)&out1[8 * i], t); _mm_storeh_pd(&temp1, t);
memmove(&out0[8 * i], &temp0, sizeof(double));
memmove(&out1[8 * i], &temp1, sizeof(double));
t = _mm_castsi128_pd(_mm256_extracti128_si256(s[i], 1)); t = _mm_castsi128_pd(_mm256_extracti128_si256(s[i], 1));
_mm_storel_pd((double *)&out2[8 * i], t); _mm_storel_pd(&temp0, t);
_mm_storeh_pd((double *)&out3[8 * i], t); _mm_storeh_pd(&temp1, t);
memmove(&out2[8 * i], &temp0, sizeof(double));
memmove(&out3[8 * i], &temp1, sizeof(double));
} }
out0 += r; out0 += r;

13
crypto_sign/dilithium5/avx2/fips202x4.c
View File

@ -91,17 +91,22 @@ static void keccakx4_squeezeblocks(uint8_t *out0,
unsigned int r, unsigned int r,
__m256i s[25]) { __m256i s[25]) {
unsigned int i; unsigned int i;
double temp0, temp1;
__m128d t; __m128d t;
while (nblocks > 0) { while (nblocks > 0) {
PQCLEAN_DILITHIUM5_AVX2_f1600x4(s, KeccakF_RoundConstants); PQCLEAN_DILITHIUM5_AVX2_f1600x4(s, KeccakF_RoundConstants);
for (i = 0; i < r / 8; ++i) { for (i = 0; i < r / 8; ++i) {
t = _mm_castsi128_pd(_mm256_castsi256_si128(s[i])); t = _mm_castsi128_pd(_mm256_castsi256_si128(s[i]));
_mm_storel_pd((double *)&out0[8 * i], t); _mm_storel_pd(&temp0, t);
_mm_storeh_pd((double *)&out1[8 * i], t); _mm_storeh_pd(&temp1, t);
memmove(&out0[8 * i], &temp0, sizeof(double));
memmove(&out1[8 * i], &temp1, sizeof(double));
t = _mm_castsi128_pd(_mm256_extracti128_si256(s[i], 1)); t = _mm_castsi128_pd(_mm256_extracti128_si256(s[i], 1));
_mm_storel_pd((double *)&out2[8 * i], t); _mm_storel_pd(&temp0, t);
_mm_storeh_pd((double *)&out3[8 * i], t); _mm_storeh_pd(&temp1, t);
memmove(&out2[8 * i], &temp0, sizeof(double));
memmove(&out3[8 * i], &temp1, sizeof(double));
} }
out0 += r; out0 += r;