feat(benchmark): 临时固定随机种子初始化值

.cmake/arm_optimization.cmake

@@ -1,75 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# ARM架构优化配置文件
-
-# 检查是否为ARM架构
-if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "aarch64" OR ${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm64" OR ${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm")
-
-    # 启用NEON指令集优化
-    if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "aarch64" OR ${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm64")
-        # ARM64架构
-        add_compile_options(-march=armv8-a+simd)
-        
-        # 如果支持 crypto 扩展，则启用
-        include(CheckCSourceCompiles)
-        check_c_source_compiles("
-            #include <arm_neon.h>
-            int main() {
-                uint8x16_t a = vdupq_n_u8(0);
-                uint8x16_t b = vaeseq_u8(a, vdupq_n_u8(0));
-                return 0;
-            }" HAVE_ARM64_CRYPTO)
-            
-        if (HAVE_ARM64_CRYPTO)
-            add_compile_options(-march=armv8-a+crypto)
-            add_compile_definitions(HAVE_ARM64_CRYPTO)
-        endif()
-        
-        # CPU特定优化
-        # 根据实际部署平台选择合适的CPU型号
-        add_compile_options(-mtune=cortex-a76)  # 默认使用cortex-a76
-        
-        # 更多ARM64优化选项
-        add_compile_options(
-            -moutline-atomics           # 内联原子操作
-            -mstrict-align              # 严格对齐优化
-        )
-        
-    else()
-        # ARM32架构
-        add_compile_options(-march=armv7-a -mfpu=neon)
-    endif()
-
-    # 通用ARM优化选项
-    add_compile_options(
-        -O3                         # 最高级别优化
-        -funroll-loops              # 循环展开
-        -fomit-frame-pointer        # 省略帧指针
-        -frename-registers          # 重命名寄存器
-        -fipa-pta                   # 点对点分析
-        -floop-optimize             # 循环优化
-        -fprefetch-loop-arrays      # 预取循环数组
-        -funroll-all-loops          # 展开所有循环
-        -fpeel-loops                # 循环剥离
-    )
-    
-    # 浮点运算优化
-    add_compile_options(
-        -ffast-math                 # 快速数学运算
-        -ffp-contract=fast          # 快速浮点收缩
-        -funsafe-math-optimizations # 不安全的数学优化
-        -ftree-vectorize            # 树向量化
-    )
-    
-    # 启用链接时优化(LTO)
-    include(CheckIPOSupported)
-    check_ipo_supported(RESULT result)
-    if(result)
-        set(CMAKE_INTERPROCEDURAL_OPTIMIZATION TRUE)
-        add_compile_options(-flto=auto)
-    endif()
-    
-    message(STATUS "ARM optimizations enabled for ${CMAKE_SYSTEM_PROCESSOR}")
-    
-    # 添加NEON支持的定义
-    add_compile_definitions(HAVE_NEON)
-endif()

.cmake/target.cmake

@@ -51,11 +51,6 @@ else()
     message("Warning: system architecture not detected, defaulting to 64 bit")
 endif()
 
-# 修改默认GF_RADIX为64
-if (NOT DEFINED GF_RADIX)
-    set(GF_RADIX 64)
-endif()
-
 if (NOT GF_RADIX STREQUAL "AUTO")
     if (NOT((GF_RADIX EQUAL 64) OR (GF_RADIX EQUAL 32)))
         message(FATAL_ERROR "Currently supported options for GF_RADIX: 32 or 64. Aborting")

CMakeLists.txt

@@ -13,21 +13,9 @@ option(ENABLE_STRICT "Build with strict compile options." ON)
 option(ENABLE_TESTS  "Enable compilation of tests." ON)
 option(ENABLE_CT_TESTING  "Enable compilation for constant time testing." OFF)
 option(ENABLE_SIGN "Build with sign functionality" ON)
-option(ENABLE_OPENMP "Enable OpenMP for parallel computation" OFF)
 set(GMP_LIBRARY "SYSTEM" CACHE STRING "Which version of GMP to use: SYSTEM, BUILD or MINI")
 set(GF_RADIX "AUTO" CACHE STRING "Set the radix for the gf module (currently supported values: 32 or 64), or AUTO.")
 
-if(ENABLE_OPENMP)
-    find_package(OpenMP REQUIRED)
-    if(OpenMP_C_FOUND)
-        add_compile_definitions(HAVE_OPENMP)
-        # 将OpenMP标志添加到全局编译选项，供后续目标使用
-        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}")
-        set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} ${OpenMP_EXE_LINKER_FLAGS}")
-        set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} ${OpenMP_SHARED_LINKER_FLAGS}")
-    endif()
-endif()
-
 if (NOT DEFINED SQISIGN_BUILD_TYPE)
   SET(SQISIGN_BUILD_TYPE "ref")
 endif()
@@ -47,8 +35,6 @@ SET(SVARIANT_S "lvl1;lvl3;lvl5")
 include(.cmake/flags.cmake)
 include(.cmake/sanitizers.cmake)
 include(.cmake/target.cmake)
-include(.cmake/arm_optimization.cmake)
-
 if(ENABLE_SIGN)
 	include(.cmake/gmpconfig.cmake)
 	add_compile_definitions(ENABLE_SIGN)

benchmark.sh

@@ -1,13 +0,0 @@
-# cmake -DSQISIGN_BUILD_TYPE=ref -DCMAKE_BUILD_TYPE=Release ..
-# 包含完整跑分和验证的测试脚本
-# BASE_DIR="./build_Neon_ml"
-BASE_DIR="./build_ref_release_test"
-
-
-echo "------------------------start benchmark------------------------------"
-
-$BASE_DIR/apps/benchmark_lvl1 --iterations=100 >> $BASE_DIR/test_result.txt
-$BASE_DIR/apps/benchmark_lvl3 --iterations=100 >> $BASE_DIR/test_result.txt
-$BASE_DIR/apps/benchmark_lvl5 --iterations=100 >> $BASE_DIR/test_result.txt
-
-echo "-------------------------end benchmark-----------------------------"

src/common/arm64crypto/randombytes_ctrdrbg.c

@@ -7,344 +7,270 @@
 
 static AES256_CTR_DRBG_struct DRBG_ctx;
 
-__attribute__((always_inline, hot))
+static inline uint32_t AES_sbox_x4(uint32_t in) {
-static inline uint32_t
+  uint8x16_t sbox_val = vreinterpretq_u8_u32(vdupq_n_u32(in));
-AES_sbox_x4(uint32_t in)
+  sbox_val = vaeseq_u8(sbox_val, vdupq_n_u8(0));
-{
-    uint8x16_t sbox_val = vreinterpretq_u8_u32(vdupq_n_u32(in));
-    sbox_val = vaeseq_u8(sbox_val, vdupq_n_u8(0));
 
-    return vgetq_lane_u32(vreinterpretq_u32_u8(sbox_val), 0);
+  return vgetq_lane_u32(vreinterpretq_u32_u8(sbox_val), 0);
 }
 
 #define ROTR32(x, n) ((x << (32 - n)) | (x >> n))
 
-typedef union
+typedef union {
-{
+  uint8_t u8[15][16];
-    uint8_t u8[15][16];
+  uint32_t u32[15][4];
-    uint32_t u32[15][4];
 } subkeys_t;
 
-__attribute__((hot))
+static void AES256_key_schedule(uint8_t subkeys[15][16], const uint8_t *key) {
-static void
+  subkeys_t *sk = (subkeys_t *)subkeys;
-AES256_key_schedule(uint8_t subkeys[15][16], const uint8_t *key)
+  uint8_t rcon = 1;
-{
+  uint32_t s;
-    uint8x16_t rcon = vdupq_n_u8(0x01);
+  int i, j;
-    // uint8x16_t rcon_step = vdupq_n_u8(0x1b);
 
-    // 一次性复制前两轮密钥
+  memcpy(&subkeys[0][0], key, 32 * sizeof(uint8_t));
-    memcpy(&subkeys[0][0], key, 32);
 
-    uint8x16_t prev_key = vld1q_u8(&subkeys[0][0]);
+  for (i = 2; i < 14; i += 2) {
-    uint8x16_t prev_prev_key = vld1q_u8(&subkeys[1][0]);
+    s = AES_sbox_x4(sk->u32[i - 1][3]);
+    sk->u32[i][0] = ROTR32(s, 8) ^ rcon ^ sk->u32[i - 2][0];
 
-    for (int i = 2; i < 15; i++) {
+    for (j = 1; j < 4; j++) {
-        // 提取最后一列并进行S-box变换
+      sk->u32[i][j] = sk->u32[i][j - 1] ^ sk->u32[i - 2][j];
-        uint8x16_t last_col = vextq_u8(prev_key, vdupq_n_u8(0), 12);
-        last_col = vaeseq_u8(last_col, vdupq_n_u8(0));
-
-        // RotWord
-        last_col = vextq_u8(last_col, last_col, 3);
-
-        // XOR with rcon
-        uint8x16_t new_key_first = veorq_u8(veorq_u8(last_col, rcon), prev_prev_key);
-
-        // 生成新密钥的剩余部分
-        uint8x16_t new_key = vextq_u8(prev_prev_key, new_key_first, 12);
-
-        // 保存新密钥
-        vst1q_u8(&subkeys[i][0], new_key);
-
-        // 更新rcon
-        uint8_t rcon_val = vgetq_lane_u8(rcon, 0);
-        rcon_val = (rcon_val << 1) ^ ((rcon_val >> 7) * 0x1b);
-        rcon = vdupq_n_u8(rcon_val);
-
-        // 更新前两个密钥
-        prev_prev_key = prev_key;
-        prev_key = new_key;
     }
+
+    s = AES_sbox_x4(sk->u32[i][3]);
+    sk->u32[i + 1][0] = s ^ sk->u32[i - 1][0];
+
+    for (j = 1; j < 4; j++) {
+      sk->u32[i + 1][j] = sk->u32[i + 1][j - 1] ^ sk->u32[i - 1][j];
+    }
+
+    rcon = (rcon << 1) ^ ((rcon >> 7) * 0x11b);
+  }
+
+  s = AES_sbox_x4(sk->u32[13][3]);
+  sk->u32[14][0] = ROTR32(s, 8) ^ rcon ^ sk->u32[12][0];
+
+  for (j = 1; j < 4; j++) {
+    sk->u32[14][j] = sk->u32[14][j - 1] ^ sk->u32[12][j];
+  }
 }
 
-#define AES256_ECB_XWAYS(ways, vsubkeys, ctr, out)                                                                     \
+#define AES256_ECB_XWAYS(ways, vsubkeys, ctr, out)                             \
-    do {                                                                                                               \
+  do {                                                                         \
-        uint8x16_t state[ways];                                                                                        \
+    uint8x16_t state[ways];                                                    \
-                                                                                                                       \
+                                                                               \
-        for (int j = 0; j < ways; j++) {                                                                               \
+    for (int j = 0; j < ways; j++) {                                           \
-            state[j] = vaeseq_u8(ctr[j], vsubkeys[0]);                                                                 \
+      state[j] = vaeseq_u8(ctr[j], vsubkeys[0]);                               \
-            state[j] = vaesmcq_u8(state[j]);                                                                           \
+      state[j] = vaesmcq_u8(state[j]);                                         \
-        }                                                                                                              \
+    }                                                                          \
-                                                                                                                       \
+                                                                               \
-        for (int i = 1; i < 13; i++) {                                                                                 \
+    for (int i = 1; i < 13; i++) {                                             \
-            for (int j = 0; j < ways; j++) {                                                                           \
+      for (int j = 0; j < ways; j++) {                                         \
-                state[j] = vaeseq_u8(state[j], vsubkeys[i]);                                                           \
+        state[j] = vaeseq_u8(state[j], vsubkeys[i]);                           \
-                state[j] = vaesmcq_u8(state[j]);                                                                       \
+        state[j] = vaesmcq_u8(state[j]);                                       \
-            }                                                                                                          \
+      }                                                                        \
-        }                                                                                                              \
+    }                                                                          \
-                                                                                                                       \
+                                                                               \
-        for (int j = 0; j < ways; j++) {                                                                               \
+    for (int j = 0; j < ways; j++) {                                           \
-            state[j] = vaeseq_u8(state[j], vsubkeys[13]);                                                              \
+      state[j] = vaeseq_u8(state[j], vsubkeys[13]);                            \
-            state[j] = veorq_u8(state[j], vsubkeys[14]);                                                               \
+      state[j] = veorq_u8(state[j], vsubkeys[14]);                             \
-            vst1q_u8(out + j * 16, state[j]);                                                                          \
+      vst1q_u8(out + j * 16, state[j]);                                        \
-        }                                                                                                              \
+    }                                                                          \
-    } while (0);
+  } while (0);
 
 //    subkeys - subkeys for AES-256
 //    ctr - a 128-bit plaintext value
 //    buffer - a 128-bit ciphertext value
-static void
+static void AES256_ECB(uint8x16_t vsubkeys[15], uint8x16_t ctr,
-AES256_ECB(uint8x16_t vsubkeys[15], uint8x16_t ctr, unsigned char *buffer)
+                       unsigned char *buffer) {
-{
+  AES256_ECB_XWAYS(1, vsubkeys, (&ctr), buffer);
-    AES256_ECB_XWAYS(1, vsubkeys, (&ctr), buffer);
 }
 
 // vsubkeys - subkeys for AES-256
 // ctr - an array of 3 x 128-bit plaintext value
 // buffer - an array of 3 x 128-bit ciphertext value
-static void
+static void AES256_ECB_x3(uint8x16_t vsubkeys[15], uint8x16_t ctr[3],
-AES256_ECB_x3(uint8x16_t vsubkeys[15], uint8x16_t ctr[3], unsigned char *buffer)
+                          unsigned char *buffer) {
-{
+  AES256_ECB_XWAYS(3, vsubkeys, ctr, buffer);
-    AES256_ECB_XWAYS(3, vsubkeys, ctr, buffer);
 }
 
-static void
+static void bswap128(__uint128_t *x) {
-bswap128(__uint128_t *x)
+  uint64_t *x64 = (uint64_t *)x;
-{
-    uint64_t *x64 = (uint64_t *)x;
 
-    uint64_t t = x64[0];
+  uint64_t t = x64[0];
-    x64[0] = x64[1];
+  x64[0] = x64[1];
-    x64[1] = t;
+  x64[1] = t;
 
-    x64[0] = __builtin_bswap64(x64[0]);
+  x64[0] = __builtin_bswap64(x64[0]);
-    x64[1] = __builtin_bswap64(x64[1]);
+  x64[1] = __builtin_bswap64(x64[1]);
 }
 
-static void
+static void add_to_V(unsigned char V[], int incr) {
-add_to_V(unsigned char V[], int incr)
+  __uint128_t *V128 = (__uint128_t *)V;
-{
+  bswap128(V128);
-    __uint128_t *V128 = (__uint128_t *)V;
+  (*V128) += incr;
-    bswap128(V128);
+  bswap128(V128);
-    (*V128) += incr;
-    bswap128(V128);
 }
 
-static void
+static void AES256_CTR_DRBG_Update(unsigned char *provided_data,
-AES256_CTR_DRBG_Update(unsigned char *provided_data, uint8x16_t vsubkeys[15], unsigned char *Key, unsigned char *V)
+                                   uint8x16_t vsubkeys[15], unsigned char *Key,
-{
+                                   unsigned char *V) {
-    unsigned char temp[48];
+  unsigned char temp[48];
-    __uint128_t V128, t;
+  __uint128_t V128, t;
-    uint64x2_t vV[3];
+  uint64x2_t vV[3];
 
-    memcpy(&V128, DRBG_ctx.V, sizeof(V128));
+  memcpy(&V128, DRBG_ctx.V, sizeof(V128));
 
-    bswap128(&V128);
+  bswap128(&V128);
 
-    for (int j = 0; j < 3; j++) {
+  for (int j = 0; j < 3; j++) {
-        V128++;
+    V128++;
-        t = V128;
+    t = V128;
-        bswap128(&t);
+    bswap128(&t);
-        vV[j] = vld1q_u64((uint64_t *)&t);
+    vV[j] = vld1q_u64((uint64_t *)&t);
-    }
+  }
 
-    AES256_ECB_x3(vsubkeys, (uint8x16_t *)vV, temp);
+  AES256_ECB_x3(vsubkeys, (uint8x16_t *)vV, temp);
 
-    //   if (provided_data != NULL)
+  if (provided_data != NULL)
-    //     for (int i = 0; i < 48; i++)
+    for (int i = 0; i < 48; i++)
-    //       temp[i] ^= provided_data[i];
+      temp[i] ^= provided_data[i];
-    if (provided_data != NULL) {
+  memcpy(Key, temp, 32);
-        // 使用 SIMD 进行批量 XOR 操作
+  memcpy(V, temp + 32, 16);
-        uint8x16_t *temp_vec = (uint8x16_t *)temp;
-        uint8x16_t *prov_vec = (uint8x16_t *)provided_data;
 
-        temp_vec[0] = veorq_u8(temp_vec[0], prov_vec[0]);
+  add_to_V(DRBG_ctx.V, 1);
-        temp_vec[1] = veorq_u8(temp_vec[1], prov_vec[1]);
-        temp_vec[2] = veorq_u8(temp_vec[2], prov_vec[2]);
-    }
-    memcpy(Key, temp, 32);
-    memcpy(V, temp + 32, 16);
-
-    add_to_V(DRBG_ctx.V, 1);
 }
 
-void
+void randombytes_init_arm64crypto(unsigned char *entropy_input,
-randombytes_init_arm64crypto(unsigned char *entropy_input, unsigned char *personalization_string, int security_strength)
+                                  unsigned char *personalization_string,
-{
+                                  int security_strength) {
-    (void)security_strength;
+  (void)security_strength;
 
-    unsigned char seed_material[48];
+  unsigned char seed_material[48];
-    uint8_t subkeys[15][16];
+  uint8_t subkeys[15][16];
-    uint8x16_t vsubkeys[15];
+  uint8x16_t vsubkeys[15];
 
-    memcpy(seed_material, entropy_input, 48);
+  memcpy(seed_material, entropy_input, 48);
-    // if (personalization_string)
+  if (personalization_string)
-    //     for (int i = 0; i < 48; i++)
+    for (int i = 0; i < 48; i++)
-    //         seed_material[i] ^= personalization_string[i];
+      seed_material[i] ^= personalization_string[i];
-    if (personalization_string) {
+  memset(DRBG_ctx.Key, 0x00, 32);
-        // 使用 SIMD 加速 XOR 操作
+  memset(DRBG_ctx.V, 0x00, 16);
-        uint8x16_t *seed_vec = (uint8x16_t *)seed_material;
-        uint8x16_t *pers_vec = (uint8x16_t *)personalization_string;
 
-        seed_vec[0] = veorq_u8(seed_vec[0], pers_vec[0]);
+  AES256_key_schedule(subkeys, DRBG_ctx.Key);
-        seed_vec[1] = veorq_u8(seed_vec[1], pers_vec[1]);
+  for (int i = 0; i < 15; i++) {
-        seed_vec[2] = veorq_u8(seed_vec[2], pers_vec[2]);
+    vsubkeys[i] = vld1q_u8(subkeys[i]);
-    }
+  }
 
-    memset(DRBG_ctx.Key, 0x00, 32);
+  AES256_CTR_DRBG_Update(seed_material, vsubkeys, DRBG_ctx.Key, DRBG_ctx.V);
-    memset(DRBG_ctx.V, 0x00, 16);
+  DRBG_ctx.reseed_counter = 1;
-
-    AES256_key_schedule(subkeys, DRBG_ctx.Key);
-    for (int i = 0; i < 15; i++) {
-        vsubkeys[i] = vld1q_u8(subkeys[i]);
-    }
-
-    AES256_CTR_DRBG_Update(seed_material, vsubkeys, DRBG_ctx.Key, DRBG_ctx.V);
-    DRBG_ctx.reseed_counter = 1;
 }
 
-#define WAYS 8
+#define WAYS 4
 
-int
+int randombytes_arm64crypto(unsigned char *x, unsigned long long xlen) {
-randombytes_arm64crypto(unsigned char *x, unsigned long long xlen)
+  uint8_t subkeys[15][16];
-{
+  unsigned char block[16];
-    uint8_t subkeys[15][16];
+  __uint128_t V[WAYS], Vle[WAYS];
-    unsigned char block[16];
+  uint8x16x4_t vV;
-    __uint128_t V[WAYS], Vle[WAYS];
+  uint8x16_t vsubkeys[15];
-    uint8x16x4_t vV;
-    uint8x16_t vsubkeys[15];
 
-    AES256_key_schedule(subkeys, DRBG_ctx.Key);
+  AES256_key_schedule(subkeys, DRBG_ctx.Key);
 
-    for (int j = 0; j < 15; j++) {
+  for (int j = 0; j < 15; j++) {
-        vsubkeys[j] = vld1q_u8(subkeys[j]);
+    vsubkeys[j] = vld1q_u8(subkeys[j]);
+  }
+
+  memcpy(&Vle[0], DRBG_ctx.V, sizeof(Vle[0]));
+  V[0] = Vle[0];
+  vV.val[0] = vld1q_u8((uint8_t *)&V[0]);
+  bswap128(&Vle[0]);
+  for (int j = 1; j < WAYS; j++) {
+    Vle[j] = Vle[j - 1] + 1;
+    V[j] = Vle[j];
+    bswap128(&V[j]);
+    vV.val[j] = vld1q_u8((uint8_t *)&V[j]);
+  }
+
+  int entered_fast_path = (xlen >= WAYS * 16) ? 1 : 0;
+
+  while (xlen >= WAYS * 16) {
+    for (int j = 0; j < WAYS; j++) {
+      Vle[j] += 4;
     }
 
-    memcpy(&Vle[0], DRBG_ctx.V, sizeof(Vle[0]));
+    for (int j = 0; j < WAYS; j++) {
-    V[0] = Vle[0];
+      vV.val[j] = vaeseq_u8(vV.val[j], vsubkeys[0]);
-    vV.val[0] = vld1q_u8((uint8_t *)&V[0]);
+      vV.val[j] = vaesmcq_u8(vV.val[j]);
-    bswap128(&Vle[0]);
-    for (int j = 1; j < WAYS; j++) {
-        Vle[j] = Vle[j - 1] + 1;
-        V[j] = Vle[j];
-        bswap128(&V[j]);
-        // 分批加载到向量寄存器中
-        if (j % 4 == 0 || j == WAYS - 1) {
-            vV = vld1q_u8_x4((uint8_t *)&V[j-3]);
-        }
     }
 
-    int entered_fast_path = (xlen >= WAYS * 16) ? 1 : 0;
+    for (int i = 1; i < 13; i++) {
-
+      for (int j = 0; j < WAYS; j++) {
-    while (xlen >= WAYS * 16) {
+        vV.val[j] = vaeseq_u8(vV.val[j], vsubkeys[i]);
-        // 添加预取指令
+        vV.val[j] = vaesmcq_u8(vV.val[j]);
-        __builtin_prefetch(&x[64], 1, 3);
+      }
-        for (int j = 0; j < WAYS; j++) {
-            Vle[j] += 8; // 增加步长以减少循环次数
-        }
-
-        // 展开循环以减少分支预测失败
-        for (int j = 0; j < WAYS; j += 4) {
-            uint8x16_t state0 = vaeseq_u8(vV.val[0], vsubkeys[0]);
-            uint8x16_t state1 = vaeseq_u8(vV.val[1], vsubkeys[0]);
-            uint8x16_t state2 = vaeseq_u8(vV.val[2], vsubkeys[0]);
-            uint8x16_t state3 = vaeseq_u8(vV.val[3], vsubkeys[0]);
-            
-            state0 = vaesmcq_u8(state0);
-            state1 = vaesmcq_u8(state1);
-            state2 = vaesmcq_u8(state2);
-            state3 = vaesmcq_u8(state3);
-
-            for (int i = 1; i < 13; i++) {
-                state0 = vaeseq_u8(state0, vsubkeys[i]);
-                state1 = vaeseq_u8(state1, vsubkeys[i]);
-                state2 = vaeseq_u8(state2, vsubkeys[i]);
-                state3 = vaeseq_u8(state3, vsubkeys[i]);
-                
-                state0 = vaesmcq_u8(state0);
-                state1 = vaesmcq_u8(state1);
-                state2 = vaesmcq_u8(state2);
-                state3 = vaesmcq_u8(state3);
-            }
-
-            state0 = vaeseq_u8(state0, vsubkeys[13]);
-            state1 = vaeseq_u8(state1, vsubkeys[13]);
-            state2 = vaeseq_u8(state2, vsubkeys[13]);
-            state3 = vaeseq_u8(state3, vsubkeys[13]);
-            
-            state0 = veorq_u8(state0, vsubkeys[14]);
-            state1 = veorq_u8(state1, vsubkeys[14]);
-            state2 = veorq_u8(state2, vsubkeys[14]);
-            state3 = veorq_u8(state3, vsubkeys[14]);
-            
-            vst1q_u8(x + 0 * 16, state0);
-            vst1q_u8(x + 1 * 16, state1);
-            vst1q_u8(x + 2 * 16, state2);
-            vst1q_u8(x + 3 * 16, state3);
-            
-            // 更新向量寄存器
-            if (j + 4 < WAYS) {
-                vV.val[0] = vld1q_u8((uint8_t *)&V[j+4]);
-                vV.val[1] = vld1q_u8((uint8_t *)&V[j+5]);
-                vV.val[2] = vld1q_u8((uint8_t *)&V[j+6]);
-                vV.val[3] = vld1q_u8((uint8_t *)&V[j+7]);
-            }
-        }
-
-        for (int j = 0; j < WAYS; j++) {
-            V[j] = Vle[j];
-            bswap128(&V[j]);
-        }
-
-        vV = vld1q_u8_x4((uint8_t *)V);
-
-        x += WAYS * 16;
-        xlen -= WAYS * 16;
     }
 
-    if (entered_fast_path && xlen == 0) {
+    for (int j = 0; j < WAYS; j++) {
-        asm volatile("" : "+r,m"(Vle[7]) : : "memory");
+      vV.val[j] = vaeseq_u8(vV.val[j], vsubkeys[13]);
-        V[0] = Vle[7] - 8;
+      vV.val[j] = veorq_u8(vV.val[j], vsubkeys[14]);
-        bswap128(&V[0]);
+      vst1q_u8(x + j * 16, vV.val[j]);
     }
 
-    while (xlen > 0) {
+    for (int j = 0; j < WAYS; j++) {
-        if (xlen > 16) {
+      V[j] = Vle[j];
-            AES256_ECB(vsubkeys, vld1q_u8((uint8_t *)&V[0]), x);
+      bswap128(&V[j]);
-            x += 16;
-            xlen -= 16;
-
-            Vle[0]++;
-            V[0] = Vle[0];
-            bswap128(&V[0]);
-        } else {
-            AES256_ECB(vsubkeys, vld1q_u8((uint8_t *)&V[0]), block);
-            memcpy(x, block, xlen);
-            xlen = 0;
-        }
     }
 
-    memcpy(DRBG_ctx.V, &V[0], sizeof(V[0]));
+    vV = vld1q_u8_x4((uint8_t *)V);
 
-    AES256_CTR_DRBG_Update(NULL, vsubkeys, DRBG_ctx.Key, DRBG_ctx.V);
+    x += WAYS * 16;
-    DRBG_ctx.reseed_counter++;
+    xlen -= WAYS * 16;
+  }
 
-    return RNG_SUCCESS;
+  if (entered_fast_path && xlen == 0) {
+    asm volatile("" : "+r,m"(Vle[3]) : : "memory");
+    V[0] = Vle[3] - 4;
+    bswap128(&V[0]);
+  }
+
+  while (xlen > 0) {
+    if (xlen > 16) {
+      AES256_ECB(vsubkeys, vld1q_u8((uint8_t *)&V[0]), x);
+      x += 16;
+      xlen -= 16;
+
+      Vle[0]++;
+      V[0] = Vle[0];
+      bswap128(&V[0]);
+    } else {
+      AES256_ECB(vsubkeys, vld1q_u8((uint8_t *)&V[0]), block);
+      memcpy(x, block, xlen);
+      xlen = 0;
+    }
+  }
+
+  memcpy(DRBG_ctx.V, &V[0], sizeof(V[0]));
+
+  AES256_CTR_DRBG_Update(NULL, vsubkeys, DRBG_ctx.Key, DRBG_ctx.V);
+  DRBG_ctx.reseed_counter++;
+
+  return RNG_SUCCESS;
 }
 
 #ifdef RANDOMBYTES_ARM64CRYPTO
-int
+int randombytes(unsigned char *random_array, unsigned long long nbytes) {
-randombytes(unsigned char *random_array, unsigned long long nbytes)
+  int ret = randombytes_arm64crypto(random_array, nbytes);
-{
-    int ret = randombytes_arm64crypto(random_array, nbytes);
 #ifdef ENABLE_CT_TESTING
-    VALGRIND_MAKE_MEM_UNDEFINED(random_array, ret);
+  VALGRIND_MAKE_MEM_UNDEFINED(random_array, ret);
 #endif
-    return ret;
+  return ret;
 }
 
-void
+void randombytes_init(unsigned char *entropy_input,
-randombytes_init(unsigned char *entropy_input, unsigned char *personalization_string, int security_strength)
+                      unsigned char *personalization_string,
-{
+                      int security_strength) {
-    randombytes_init_arm64crypto(entropy_input, personalization_string, security_strength);
+  randombytes_init_arm64crypto(entropy_input, personalization_string,
+                               security_strength);
 }
 #endif

src/quaternion/ref/generic/lll/lll_tests.c

@@ -655,48 +655,13 @@ quat_test_lll_lideal_lideal_mul_reduced()
     ibz_mat_4x4_mul(&(gram_test), &(gram_test), &(prod.lattice.basis));
     ibz_mat_4x4_transpose(&(gram_test), &(gram_test));
     ibz_mat_4x4_mul(&(gram_test), &(gram_test), &(prod.lattice.basis));
-    
+    for (int i = 0; i < 4; i++) {
-    // ARM优化: 循环展开以减少分支预测失败的可能性
+        ibz_vec_4_set(&vec, (i == 0), (i == 1), (i == 2), (i == 3));
-    // 原始循环:
+        quat_qf_eval(&norm, &gram, &vec);
-    // for (int i = 0; i < 4; i++) {
+        quat_qf_eval(&test_norm, &gram_test, &vec);
-    //     ibz_vec_4_set(&vec, (i == 0), (i == 1), (i == 2), (i == 3));
+        ibz_mul(&norm, &(prod.norm), &norm);
-    //     quat_qf_eval(&norm, &gram, &vec);
+        res = res || !(ibz_cmp(&norm, &test_norm) == 0);
-    //     quat_qf_eval(&test_norm, &gram_test, &vec);
+    }
-    //     ibz_mul(&norm, &(prod.norm), &norm);
-    //     res = res || !(ibz_cmp(&norm, &test_norm) == 0);
-    // }
-    
-    // 展开后的循环 - 减少循环开销，更适合ARM处理器流水线
-    ibz_vec_4_set(&vec, 1, 0, 0, 0);
-    quat_qf_eval(&norm, &gram, &vec);
-    quat_qf_eval(&test_norm, &gram_test, &vec);
-    ibz_mul(&norm, &(prod.norm), &norm);
-    res = res || !(ibz_cmp(&norm, &test_norm) == 0);
-    
-    ibz_vec_4_set(&vec, 0, 1, 0, 0);
-    quat_qf_eval(&norm, &gram, &vec);
-    quat_qf_eval(&test_norm, &gram_test, &vec);
-    ibz_mul(&norm, &(prod.norm), &norm);
-    res = res || !(ibz_cmp(&norm, &test_norm) == 0);
-    
-    ibz_vec_4_set(&vec, 0, 0, 1, 0);
-    quat_qf_eval(&norm, &gram, &vec);
-    quat_qf_eval(&test_norm, &gram_test, &vec);
-    ibz_mul(&norm, &(prod.norm), &norm);
-    res = res || !(ibz_cmp(&norm, &test_norm) == 0);
-    
-    ibz_vec_4_set(&vec, 0, 0, 0, 1);
-    quat_qf_eval(&norm, &gram, &vec);
-    quat_qf_eval(&test_norm, &gram_test, &vec);
-    ibz_mul(&norm, &(prod.norm), &norm);
-    res = res || !(ibz_cmp(&norm, &test_norm) == 0);
-    
-    // 使用NEON优化大整数运算（如果可用）
-#ifdef HAVE_NEON
-    // 在支持NEON的ARM64平台上并行处理多个规范评估
-    // 这里可以进一步优化，但需要重构底层的大整数运算库
-#endif
-    
     quat_lattice_hnf(&(prod.lattice));
 
     res = res || !quat_lideal_equals(&i1, &lideal1, &alg);

src/signature/ref/include/ml_predict.h

@@ -1,6 +0,0 @@
-#ifndef ML_PREDICT_H
-#define ML_PREDICT_H
-
-double ml_predict_success(long norm, int trace, int order);
-
-#endif

src/signature/ref/lvlx/keygen.c

@@ -3,9 +3,6 @@
 #include <quaternion_data.h>
 #include <id2iso.h>
 #include <torsion_constants.h>
-#ifdef _OPENMP
-#include <omp.h>
-#endif
 
 void
 secret_key_init(secret_key_t *sk)
@@ -30,59 +27,7 @@ protocols_keygen(public_key_t *pk, secret_key_t *sk)
 
     // iterating until a solution has been found
     while (!found) {
-        // 尝试并行生成多个理想，提高找到解决方案的概率
-#ifdef _OPENMP
-        int num_threads = omp_get_max_threads();
-        if (num_threads > 1) {
-            int local_found = 0;
-#pragma omp parallel shared(found, local_found) num_threads(num_threads)
-            {
-                if (!local_found) {
-                    secret_key_t local_sk;
-                    secret_key_init(&local_sk);
 
-                    int thread_found = quat_sampling_random_ideal_O0_given_norm(
-                        &local_sk.secret_ideal, &SEC_DEGREE, 1, &QUAT_represent_integer_params, NULL);
-
-                    // replacing the secret key ideal by a shorter equivalent one for efficiency
-                    thread_found = thread_found && quat_lideal_prime_norm_reduced_equivalent(
-                                         &local_sk.secret_ideal, &QUATALG_PINFTY, QUAT_primality_num_iter, QUAT_equiv_bound_coeff);
-
-                    // ideal to isogeny clapotis
-                    ec_basis_t local_B_0_two;
-                    thread_found = thread_found && dim2id2iso_arbitrary_isogeny_evaluation(&local_B_0_two, &local_sk.curve, &local_sk.secret_ideal);
-                    
-                    if (thread_found) {
-#pragma omp critical
-                        {
-                            if (!local_found) {
-                                local_found = 1;
-                                found = 1;
-                                // Copy local results to global variables
-                                quat_left_ideal_copy(&sk->secret_ideal, &local_sk.secret_ideal);
-                                ec_curve_copy(&sk->curve, &local_sk.curve);
-                                // Copy basis
-                                ec_point_copy(&B_0_two.P, &local_B_0_two.P);
-                                ec_point_copy(&B_0_two.Q, &local_B_0_two.Q);
-                                ec_point_copy(&B_0_two.PmQ, &local_B_0_two.PmQ);
-                            }
-                        }
-                    }
-                    secret_key_finalize(&local_sk);
-                }
-            }
-        } else {
-            found = quat_sampling_random_ideal_O0_given_norm(
-                &sk->secret_ideal, &SEC_DEGREE, 1, &QUAT_represent_integer_params, NULL);
-
-            // replacing the secret key ideal by a shorter equivalent one for efficiency
-            found = found && quat_lideal_prime_norm_reduced_equivalent(
-                                 &sk->secret_ideal, &QUATALG_PINFTY, QUAT_primality_num_iter, QUAT_equiv_bound_coeff);
-
-            // ideal to isogeny clapotis
-            found = found && dim2id2iso_arbitrary_isogeny_evaluation(&B_0_two, &sk->curve, &sk->secret_ideal);
-        }
-#else
         found = quat_sampling_random_ideal_O0_given_norm(
             &sk->secret_ideal, &SEC_DEGREE, 1, &QUAT_represent_integer_params, NULL);
 
@@ -91,8 +36,8 @@ protocols_keygen(public_key_t *pk, secret_key_t *sk)
                              &sk->secret_ideal, &QUATALG_PINFTY, QUAT_primality_num_iter, QUAT_equiv_bound_coeff);
 
         // ideal to isogeny clapotis
+
         found = found && dim2id2iso_arbitrary_isogeny_evaluation(&B_0_two, &sk->curve, &sk->secret_ideal);
-#endif
     }
 
     // Assert the isogeny was found and images have the correct order

src/signature/ref/lvlx/ml_predict.c

@@ -1,124 +0,0 @@
-#include "ml_predict.h"
-#include <stdio.h>
-#include <stdlib.h>
-#include <stdint.h>
-#include <time.h>
-#include <math.h>
-#include "../src/quaternion/ref/generic/include/intbig.h"
-#include "../src/quaternion/ref/generic/include/quaternion.h"
-
-// 下述是macOS为获取时间戳的操作
-#include <sys/stat.h>
-#include <sys/types.h>
-#include <unistd.h>
-
-// ==================== 辅助函数 ====================
-
-// 将 ibz_t 转为 long
-static long ibz_to_long_safe(const ibz_t *x) {
-    // 注意：仅用于调试，不要在生产环境中用 long 表示大数
-    return (long) mpz_get_si(*x);
-}
-
-// 计算理想的“迹”值（trace）
-// 我们使用生成元近似：trace ≈ 2 * a，其中 a 是 quat 元素实部
-int quat_ideal_trace(const quat_left_ideal_t *I) {
-    if (I == NULL) return 0;
-    // 理想中存储的 lattice.basis[0] 通常包含 generator 信息
-    // 取其第一个坐标近似为实部 a
-    ibz_t a;
-    ibz_init(&a);
-    ibz_copy(&a, &I->lattice.basis[0][0]);
-    long val = ibz_to_long_safe(&a);
-    ibz_finalize(&a);
-    return (int)(2 * val);
-}
-
-// 改进的机器学习预测函数，基于更复杂的启发式模型
-double ml_predict_success(long norm_val, int trace_val, int kernel_order) {
-    // 使用更细致的概率模型
-    double score = 0.0;
-    
-    // 对范数进行对数缩放并评估
-    double log_norm = (norm_val > 0) ? log10((double)norm_val + 1) : 0;
-    if (log_norm < 12) {
-        score += 0.5;  // 范数较小的理想更容易处理
-    } else if (log_norm < 15) {
-        score += 0.3;
-    } else if (log_norm < 18) {
-        score += 0.1;
-    }
-    
-    // 对迹值进行评估
-    int abs_trace = abs(trace_val);
-    if (abs_trace < 100000000) {
-        score += 0.3;  // 迹值较小时更优
-    } else if (abs_trace < 300000000) {
-        score += 0.15;
-    }
-    
-    // 对核阶数进行评估
-    if (kernel_order == 2) {
-        score += 0.2;  // 优先选择核阶数为2的理想
-    }
-    
-    // 基于组合特征的调整
-    if (log_norm < 12 && abs_trace < 100000000) {
-        score += 0.2;  // 范数和迹都很小的情况给予额外加分
-    }
-    
-    // 限制分数范围
-    if (score > 1.0) score = 1.0;
-    
-    // 转换为成功概率（非线性映射）
-    double probability = score * score; // 平方放大差异
-    
-    return probability;
-}
-
-// 日志函数：记录一次理想尝试
-// 记录到 CSV 文件，包含时间戳
-void ml_log_ideal_attempt(int attempt,
-                          const quat_left_ideal_t *lideal_com,
-                          int kernel_order,
-                          int success_flag)
-{
-    const char *dir_path = "./dataset";
-    struct stat st = {0};
-
-    if (stat(dir_path, &st) == -1) {
-        mkdir(dir_path, 0755);
-    }
-
-    // 使用固定文件名，避免频繁创建文件
-    char csv_path[512];
-    snprintf(csv_path, sizeof(csv_path), "%s/ideal_data.csv", dir_path);
-
-    // 使用追加模式，避免覆盖
-    FILE *logfile = fopen(csv_path, "a+");
-    if (!logfile) {
-        perror("无法打开 ideal_data CSV 文件");
-        return;
-    }
-
-    // 如果是第一次写入，添加表头
-    if (ftell(logfile) == 0) {
-        fprintf(logfile, "Timestamp,Attempt,Norm,Trace,KernelOrder,Prob,Success\n");
-    }
-
-    // 提取特征和记录
-    long norm_val = ibz_to_long_safe(&lideal_com->norm);
-    int trace_val = quat_ideal_trace(lideal_com);
-    double prob = ml_predict_success(norm_val, trace_val, kernel_order);
-
-    time_t t = time(NULL);
-    struct tm tm_info;
-    localtime_r(&t, &tm_info);
-    char time_str[32];
-    strftime(time_str, sizeof(time_str), "%Y-%m-%d-%H-%M-%S", &tm_info);
-
-    fprintf(logfile, "%s,%d,%ld,%d,%d,%.3f,%d\n",
-            time_str, attempt, norm_val, trace_val, kernel_order, prob, success_flag);
-
-    fclose(logfile);
-}

src/signature/ref/lvlx/sign.c

@@ -1,148 +1,27 @@
 #include <signature.h>
-#include <ml_predict.h> // 包含机器学习模型的头文件
-#include "ml_predict.c" // 包含实现文件（仅用于测试，生产环境应编译链接）
-#include <assert.h>
 #include <tools.h>
 #include <quaternion_data.h>
 #include <id2iso.h>
 #include <torsion_constants.h>
 #include <encoded_sizes.h>
-#include <intbig.h>
-#ifdef _OPENMP
-#include <omp.h>
-#endif
-
-// 添加函数声明
-static bool execute_commit_serial(ec_curve_t *E_com, ec_basis_t *basis_even_com,
-                                 quat_left_ideal_t *lideal_com, int *attempt_counter, int kernel_order);
 
 // compute the commitment with ideal to isogeny clapotis
 // and apply it to the basis of E0 (together with the multiplication by some scalar u)
-
-// 原代码如下，为了进行相应的测试，暂时注释掉
-// static bool
-// commit(ec_curve_t *E_com, ec_basis_t *basis_even_com, quat_left_ideal_t *lideal_com)
-// {
-
-//     bool found = false;
-
-//     found = quat_sampling_random_ideal_O0_given_norm(lideal_com, &COM_DEGREE, 1, &QUAT_represent_integer_params, NULL);
-//     // replacing it with a shorter prime norm equivalent ideal
-//     found = found && quat_lideal_prime_norm_reduced_equivalent(
-//                          lideal_com, &QUATALG_PINFTY, QUAT_primality_num_iter, QUAT_equiv_bound_coeff);
-//     // ideal to isogeny clapotis
-//     found = found && dim2id2iso_arbitrary_isogeny_evaluation(basis_even_com, E_com, lideal_com);
-//     return found;
-// }
-
-
-// 这是我的第一个版本的代码，需要进一步测试和调试，于是为了收集更多正确和错误的数据集，我在下面改了第二版
 static bool
 commit(ec_curve_t *E_com, ec_basis_t *basis_even_com, quat_left_ideal_t *lideal_com)
 {
-    static int attempt_counter = 0;
-    bool found = false;
-    int kernel_order = 2;
 
-    // 使用并行计算来加速随机理想采样
-#ifdef _OPENMP
-    int num_threads = omp_get_max_threads();
-    if (num_threads > 1) {
-#pragma omp parallel for reduction(|:found) num_threads(num_threads)
-        for (int i = 0; i < num_threads && !found; i++) {
-            // 为每个线程创建独立的临时变量
-            quat_left_ideal_t local_ideal;
-            quat_left_ideal_init(&local_ideal);
-            
-            // 添加线程特定的随机性
-            random_state_t local_rand;
-            random_init(&local_rand);
-            random_add_entropy(&local_rand, (unsigned char*)&i, sizeof(i));
-            
-            // 尝试采样理想
-            bool local_found = quat_sampling_random_ideal_O0_given_norm(
-                &local_ideal, &COM_DEGREE, 1, &QUAT_represent_integer_params, &local_rand);
-            
-            if (local_found) {
-                // 用机器学习预测理想是否值得尝试
-                long norm_val = mpz_get_si(local_ideal.norm);
-                int trace_val = quat_ideal_trace(&local_ideal);
-                double prob = ml_predict_success(norm_val, trace_val, kernel_order);
-                
-                if (prob >= 0.3) { // 概率足够高才继续
-                    // 尝试优化等价理想
-                    local_found = local_found && 
-                        quat_lideal_prime_norm_reduced_equivalent(&local_ideal, &QUATALG_PINFTY,
-                            QUAT_primality_num_iter, QUAT_equiv_bound_coeff);
-                    
-                    if (local_found) {
-#pragma omp critical
-                        {
-                            if (!found) {
-                                // 复制成功的结果到输出参数
-                                quat_left_ideal_copy(lideal_com, &local_ideal);
-                                found = true;
-                            }
-                        }
-                    }
-                }
-            }
-            
-            quat_left_ideal_finalize(&local_ideal);
-            random_finalize(&local_rand);
-        }
-        
-        // 如果并行搜索成功，则执行同源映射评估
-        if (found) {
-            found = dim2id2iso_arbitrary_isogeny_evaluation(basis_even_com, E_com, lideal_com);
-        }
-    } else {
-        // 单线程情况 - 原有逻辑
-        found = execute_commit_serial(E_com, basis_even_com, lideal_com, &attempt_counter, kernel_order);
-    }
-#else
-    // 没有OpenMP的情况 - 原有逻辑
-    found = execute_commit_serial(E_com, basis_even_com, lideal_com, &attempt_counter, kernel_order);
-#endif
-
-    return found;
-}
-
-// 辅助函数：串行提交逻辑
-static bool execute_commit_serial(ec_curve_t *E_com, ec_basis_t *basis_even_com, 
-                                 quat_left_ideal_t *lideal_com, int *attempt_counter, int kernel_order)
-{
     bool found = false;
 
-    // Step 1. 随机生成理想
+    found = quat_sampling_random_ideal_O0_given_norm(lideal_com, &COM_DEGREE, 1, &QUAT_represent_integer_params, NULL);
-    found = quat_sampling_random_ideal_O0_given_norm(lideal_com, &COM_DEGREE, 1,
+    // replacing it with a shorter prime norm equivalent ideal
-                                                     &QUAT_represent_integer_params, NULL);
+    found = found && quat_lideal_prime_norm_reduced_equivalent(
-    // Step 2. 用机器学习预测理想是否值得尝试
+                         lideal_com, &QUATALG_PINFTY, QUAT_primality_num_iter, QUAT_equiv_bound_coeff);
-    long norm_val = mpz_get_si(lideal_com->norm);
+    // ideal to isogeny clapotis
-    int trace_val = quat_ideal_trace(lideal_com);
-    double prob = ml_predict_success(norm_val, trace_val, kernel_order);
-
-    if (prob < 0.3) {
-        // 概率太低，跳过以节省计算
-        ml_log_ideal_attempt((*attempt_counter)++, lideal_com, kernel_order, 0);
-        return false;
-    }
-
-    // Step 3. 尝试优化等价理想
-    found = found && quat_lideal_prime_norm_reduced_equivalent(lideal_com, &QUATALG_PINFTY,
-                                                               QUAT_primality_num_iter,
-                                                               QUAT_equiv_bound_coeff);
-
-    // Step 4. 理想到同源映射
     found = found && dim2id2iso_arbitrary_isogeny_evaluation(basis_even_com, E_com, lideal_com);
-
-    // Step 5. 记录结果
-    ml_log_ideal_attempt((*attempt_counter)++, lideal_com, kernel_order, found ? 1 : 0);
-
     return found;
 }
 
-
 static void
 compute_challenge_ideal_signature(quat_left_ideal_t *lideal_chall_two, const signature_t *sig, const secret_key_t *sk)
 {