kris/pqc
1
1
フォーク 0
コード イシュー プルリクエスト リリース Wiki アクティビティ

remove preprocessor conditionals from frodokem640shake

ソースを参照
このコミットが含まれているのは:
Matthias J. Kannwischer 2019-05-21 15:51:56 +02:00
コミット 57ad79ae3a
1個のファイルの変更、0行の追加、76行の削除
統計情報を表示 Patchファイルをダウンロード Diffファイルをダウンロード すべてのファイルを展開 すべてのファイルを折り畳む

76
crypto_kem/frodokem640shake/opt/matrix_shake.c
ファイルの表示

@ -13,8 +13,6 @@
#include "common.h"
#include "params.h"
#define USE_SHAKE128_FOR_A 1
int PQCLEAN_FRODOKEM640SHAKE_OPT_mul_add_as_plus_e(uint16_t *out, const uint16_t *s, const uint16_t *e, const uint8_t *seed_A) {
// Generate-and-multiply: generate matrix A (N x N) row-wise, multiply by s on the right.
// Inputs: s, e (N x N_BAR)
@ -26,28 +24,6 @@ int PQCLEAN_FRODOKEM640SHAKE_OPT_mul_add_as_plus_e(uint16_t *out, const uint16_t
*((uint32_t *)&out[i]) = *((uint32_t *)&e[i]);
}
#if defined(USE_AES128_FOR_A)
int16_t a_row_temp[4 * PARAMS_N] = {0}; // Take four lines of A at once
uint8_t aes_key_schedule[16 * 11];
AES128_load_schedule(seed_A, aes_key_schedule);
for (j = 0; j < PARAMS_N; j += PARAMS_STRIPE_STEP) {
a_row_temp[j + 1 + 0 * PARAMS_N] = PQCLEAN_FRODOKEM640SHAKE_OPT_UINT16_TO_LE(j); // Loading values in the little-endian order
a_row_temp[j + 1 + 1 * PARAMS_N] = PQCLEAN_FRODOKEM640SHAKE_OPT_UINT16_TO_LE(j);
a_row_temp[j + 1 + 2 * PARAMS_N] = PQCLEAN_FRODOKEM640SHAKE_OPT_UINT16_TO_LE(j);
a_row_temp[j + 1 + 3 * PARAMS_N] = PQCLEAN_FRODOKEM640SHAKE_OPT_UINT16_TO_LE(j);
}
for (i = 0; i < PARAMS_N; i += 4) {
for (j = 0; j < PARAMS_N; j += PARAMS_STRIPE_STEP) { // Go through A, four rows at a time
a_row_temp[j + 0 * PARAMS_N] = PQCLEAN_FRODOKEM640SHAKE_OPT_UINT16_TO_LE(i + 0); // Loading values in the little-endian order
a_row_temp[j + 1 * PARAMS_N] = PQCLEAN_FRODOKEM640SHAKE_OPT_UINT16_TO_LE(i + 1);
a_row_temp[j + 2 * PARAMS_N] = PQCLEAN_FRODOKEM640SHAKE_OPT_UINT16_TO_LE(i + 2);
a_row_temp[j + 3 * PARAMS_N] = PQCLEAN_FRODOKEM640SHAKE_OPT_UINT16_TO_LE(i + 3);
}
AES128_ECB_enc_sch((uint8_t *)a_row_temp, 4 * PARAMS_N * sizeof(int16_t), aes_key_schedule, (uint8_t *)a_row);
#elif defined (USE_SHAKE128_FOR_A)
uint8_t seed_A_separated[2 + BYTES_SEED_A];
uint16_t *seed_A_origin = (uint16_t *)&seed_A_separated;
memcpy(&seed_A_separated[2], seed_A, BYTES_SEED_A);
@ -60,7 +36,6 @@ int PQCLEAN_FRODOKEM640SHAKE_OPT_mul_add_as_plus_e(uint16_t *out, const uint16_t
shake128((unsigned char *)(a_row + 2 * PARAMS_N), (unsigned long long)(2 * PARAMS_N), seed_A_separated, 2 + BYTES_SEED_A);
seed_A_origin[0] = PQCLEAN_FRODOKEM640SHAKE_OPT_UINT16_TO_LE(i + 3);
shake128((unsigned char *)(a_row + 3 * PARAMS_N), (unsigned long long)(2 * PARAMS_N), seed_A_separated, 2 + BYTES_SEED_A);
#endif
for (k = 0; k < 4 * PARAMS_N; k++) {
a_row[k] = PQCLEAN_FRODOKEM640SHAKE_OPT_LE_TO_UINT16(a_row[k]);
}
@ -79,10 +54,6 @@ int PQCLEAN_FRODOKEM640SHAKE_OPT_mul_add_as_plus_e(uint16_t *out, const uint16_t
out[(i + 3)*PARAMS_NBAR + k] += sum[3];
}
}
#if defined(USE_AES128_FOR_A)
AES128_free_schedule(aes_key_schedule);
#endif
return 1;
}
@ -99,51 +70,6 @@ int PQCLEAN_FRODOKEM640SHAKE_OPT_mul_add_sa_plus_e(uint16_t *out, const uint16_t
*((uint32_t *)&out[i]) = *((uint32_t *)&e[i]);
}
#if defined(USE_AES128_FOR_A)
int k;
uint16_t a_cols[PARAMS_N * PARAMS_STRIPE_STEP] = {0};
uint16_t a_cols_t[PARAMS_N * PARAMS_STRIPE_STEP];
uint16_t a_cols_temp[PARAMS_N * PARAMS_STRIPE_STEP] = {0};
uint8_t aes_key_schedule[16 * 11];
AES128_load_schedule(seed_A, aes_key_schedule);
for (i = 0, j = 0; i < PARAMS_N; i++, j += PARAMS_STRIPE_STEP) {
a_cols_temp[j] = PQCLEAN_FRODOKEM640SHAKE_OPT_UINT16_TO_LE(i); // Loading values in the little-endian order
}
for (kk = 0; kk < PARAMS_N; kk += PARAMS_STRIPE_STEP) { // Go through A's columns, 8 (== PARAMS_STRIPE_STEP) columns at a time.
for (i = 0; i < (PARAMS_N * PARAMS_STRIPE_STEP); i += PARAMS_STRIPE_STEP) {
a_cols_temp[i + 1] = PQCLEAN_FRODOKEM640SHAKE_OPT_UINT16_TO_LE(kk); // Loading values in the little-endian order
}
AES128_ECB_enc_sch((uint8_t *)a_cols_temp, PARAMS_N * PARAMS_STRIPE_STEP * sizeof(int16_t), aes_key_schedule, (uint8_t *)a_cols);
for (i = 0; i < PARAMS_N; i++) { // Transpose a_cols to have access to it in the column-major order.
for (k = 0; k < PARAMS_STRIPE_STEP; k++) {
a_cols_t[k * PARAMS_N + i] = PQCLEAN_FRODOKEM640SHAKE_OPT_LE_TO_UINT16(a_cols[i * PARAMS_STRIPE_STEP + k]);
}
}
for (i = 0; i < PARAMS_NBAR; i++) {
for (k = 0; k < PARAMS_STRIPE_STEP; k += PARAMS_PARALLEL) {
uint16_t sum[PARAMS_PARALLEL] = {0};
for (j = 0; j < PARAMS_N; j++) { // Matrix-vector multiplication
uint16_t sp = s[i * PARAMS_N + j];
sum[0] += sp * a_cols_t[(k + 0) * PARAMS_N + j];
sum[1] += sp * a_cols_t[(k + 1) * PARAMS_N + j];
sum[2] += sp * a_cols_t[(k + 2) * PARAMS_N + j];
sum[3] += sp * a_cols_t[(k + 3) * PARAMS_N + j];
}
out[i * PARAMS_N + kk + k + 0] += sum[0];
out[i * PARAMS_N + kk + k + 2] += sum[2];
out[i * PARAMS_N + kk + k + 1] += sum[1];
out[i * PARAMS_N + kk + k + 3] += sum[3];
}
}
}
AES128_free_schedule(aes_key_schedule);
#elif defined (USE_SHAKE128_FOR_A) // SHAKE128
int t = 0;
uint16_t a_cols[4 * PARAMS_N];
@ -177,7 +103,5 @@ int PQCLEAN_FRODOKEM640SHAKE_OPT_mul_add_sa_plus_e(uint16_t *out, const uint16_t
}
}
}
#endif
return 1;
}