tweaks to the monerov1 cryptonight algorithm

2 files changed, 57 insertions, 21 deletions

diff --git a/src/crypto/slow-hash.c b/src/crypto/slow-hash.c
index f81aadfab..c733d7b39 100644
--- a/src/crypto/slow-hash.c
+++ b/src/crypto/slow-hash.c
@@ -61,16 +61,33 @@ extern int aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *exp
 #define VARIANT1_2(p) \
   do if (variant > 0) \
   { \
-    ((uint32_t*)p)[2] ^= nonce; \
+    xor64(p, tweak1_2); \
   } while(0)
 
-#define VARIANT1_INIT() \
-  if (variant > 0 && length < 43) \
+#define VARIANT1_CHECK() \
+  do if (length < 43) \
   { \
     fprintf(stderr, "Cryptonight variants need at least 43 bytes of data"); \
     _exit(1); \
+  } while(0)
+
+#define NONCE_POINTER (((const uint8_t*)data)+35)
+
+#define VARIANT1_PORTABLE_INIT() \
+  uint8_t tweak1_2[8]; \
+  do if (variant > 0) \
+  { \
+    VARIANT1_CHECK(); \
+    memcpy(&tweak1_2, &state.hs.b[192], sizeof(tweak1_2)); \
+    xor64(tweak1_2, NONCE_POINTER); \
+  } while(0)
+
+#define VARIANT1_INIT64() \
+  if (variant > 0) \
+  { \
+    VARIANT1_CHECK(); \
   } \
-  const uint32_t nonce = variant > 0 ? *(const uint32_t*)(((const uint8_t*)data)+39) : 0
+  const uint64_t tweak1_2 = variant > 0 ? (state.hs.w[24] ^ (*((const uint64_t*)NONCE_POINTER))) : 0
 
 #if !defined NO_AES && (defined(__x86_64__) || (defined(_MSC_VER) && defined(_WIN64)))
 // Optimised code below, uses x86-specific intrinsics, SSE2, AES-NI
@@ -159,8 +176,8 @@ extern int aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *exp
   p = U64(&hp_state[j]); \
   p[0] = a[0];  p[1] = a[1]; \
   a[0] ^= b[0]; a[1] ^= b[1]; \
+  VARIANT1_2(p + 1); \
   _b = _c; \
-  VARIANT1_2(&hp_state[j]); \
 
 #if defined(_MSC_VER)
 #define THREADV __declspec(thread)
@@ -210,6 +227,11 @@ STATIC INLINE void xor_blocks(uint8_t *a, const uint8_t *b)
     U64(a)[1] ^= U64(b)[1];
 }
 
+STATIC INLINE void xor64(uint64_t *a, const uint64_t b)
+{
+    *a ^= b;
+}
+
 /**
  * @brief uses cpuid to determine if the CPU supports the AES instructions
  * @return true if the CPU supports AES, false otherwise
@@ -568,8 +590,6 @@ void cn_slow_hash_pre(const void *data, size_t length, char *hash, int variant,
         hash_extra_blake, hash_extra_groestl, hash_extra_jh, hash_extra_skein
     };
 
-    VARIANT1_INIT();
-
     // this isn't supposed to happen, but guard against it for now.
     if(hp_state == NULL)
         slow_hash_allocate_state();
@@ -582,6 +602,8 @@ void cn_slow_hash_pre(const void *data, size_t length, char *hash, int variant,
     }
     memcpy(text, state.init, INIT_SIZE_BYTE);
 
+    VARIANT1_INIT64();
+
     /* CryptoNight Step 2:  Iteratively encrypt the results from Keccak to fill
      * the 2MB large random access buffer.
      */
@@ -705,6 +727,11 @@ void slow_hash_free_state(void)
 
 #define U64(x) ((uint64_t *) (x))
 
+STATIC INLINE void xor64(uint64 *a, const uint64 b)
+{
+    *a ^= b;
+}
+
 #pragma pack(push, 1)
 union cn_slow_hash_state
 {
@@ -750,7 +777,7 @@ union cn_slow_hash_state
   p = U64(&hp_state[j]); \
   p[0] = a[0];  p[1] = a[1]; \
   a[0] ^= b[0]; a[1] ^= b[1]; \
-  VARIANT1_2(p); \
+  VARIANT1_2(p + 1); \
   _b = _c; \
 
 
@@ -903,13 +930,13 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant)
         hash_extra_blake, hash_extra_groestl, hash_extra_jh, hash_extra_skein
     };
 
-    VARIANT1_INIT();
-
     /* CryptoNight Step 1:  Use Keccak1600 to initialize the 'state' (and 'text') buffers from the data. */
 
     hash_process(&state.hs, data, length);
     memcpy(text, state.init, INIT_SIZE_BYTE);
 
+    VARIANT1_INIT64();
+
     /* CryptoNight Step 2:  Iteratively encrypt the results from Keccak to fill
      * the 2MB large random access buffer.
      */
@@ -1097,8 +1124,6 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant)
         hash_extra_blake, hash_extra_groestl, hash_extra_jh, hash_extra_skein
     };
 
-    VARIANT1_INIT();
-
 #ifndef FORCE_USE_HEAP
     uint8_t long_state[MEMORY];
 #else
@@ -1109,6 +1134,8 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant)
     hash_process(&state.hs, data, length);
     memcpy(text, state.init, INIT_SIZE_BYTE);
 
+    VARIANT1_INIT64();
+
     aes_ctx = (oaes_ctx *) oaes_alloc();
     oaes_key_import_data(aes_ctx, state.hs.b, AES_KEY_SIZE);
 
@@ -1148,7 +1175,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant)
       swap_blocks(b, p);
       xor_blocks(b, p);
       swap_blocks(a, b);
-      VARIANT1_2(p);
+      VARIANT1_2(U64(p) + 1);
     }
 
     memcpy(text, state.init, INIT_SIZE_BYTE);
@@ -1243,6 +1270,15 @@ static void xor_blocks(uint8_t* a, const uint8_t* b) {
   }
 }
 
+static void xor64(uint8_t* left, const uint8_t* right)
+{
+  size_t i;
+  for (i = 0; i < 8; ++i)
+  {
+    left[i] ^= right[i];
+  }
+}
+
 #pragma pack(push, 1)
 union cn_slow_hash_state {
   union hash_state hs;
@@ -1265,13 +1301,13 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant) {
   uint8_t aes_key[AES_KEY_SIZE];
   oaes_ctx *aes_ctx;
 
-  VARIANT1_INIT();
-
   hash_process(&state.hs, data, length);
   memcpy(text, state.init, INIT_SIZE_BYTE);
   memcpy(aes_key, state.hs.b, AES_KEY_SIZE);
   aes_ctx = (oaes_ctx *) oaes_alloc();
 
+  VARIANT1_PORTABLE_INIT();
+
   oaes_key_import_data(aes_ctx, aes_key, AES_KEY_SIZE);
   for (i = 0; i < MEMORY / INIT_SIZE_BYTE; i++) {
     for (j = 0; j < INIT_SIZE_BLK; j++) {
@@ -1307,10 +1343,10 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant) {
     sum_half_blocks(b, d);
     swap_blocks(b, c);
     xor_blocks(b, c);
+    VARIANT1_2(c + 8);
     copy_block(&long_state[j * AES_BLOCK_SIZE], c);
     assert(j == e2i(a, MEMORY / AES_BLOCK_SIZE));
     swap_blocks(a, b);
-    VARIANT1_2(&long_state[j * AES_BLOCK_SIZE]);
   }
 
   memcpy(text, state.init, INIT_SIZE_BYTE);
diff --git a/tests/hash/tests-slow-1.txt b/tests/hash/tests-slow-1.txt
index 0c571f86f..ed8eccc22 100644
--- a/tests/hash/tests-slow-1.txt
+++ b/tests/hash/tests-slow-1.txt
@@ -1,5 +1,5 @@
-b46aab7facce3eb4b4679d1d526f2d500d43736988b7881853c4c0c4af04ac0e 00000000000000000000000000000000000000000000000000000000000000000000000000000000000000
-f0f0ab8a50a809ccd82c76996e494b20a1e01e90ed4814a27db2558511559091 00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
-13310922971e1112b382357aeaa8b0bf0cd941dc6f5e398979f3fa77d13749bd 8519e039172b0d70e5ca7b3383d6b3167315a422747b73f019cf9528f0fde341fd0f2a63030ba6450525cf6de31837669af6f1df8131faf50aaab8d3a7405589
-0d9f641d14a782748ed5548f72a20c83c4a2cfe606015ad9eda5b36a00947d72 37a636d7dafdf259b7287eddca2f58099e98619d2f99bdb8969d7b14498102cc065201c8be90bd777323f449848b215d2977c92c4c1c2da36ab46b2e389689ed97c18fec08cd3b03235c5e4c62a37ad88c7b67932495a71090e85dd4020a9300
-d74a0c9b603aa9f9af0bfec0c36d3d383c14930a0ee2f08dab44536fccda5bee 38274c97c45a172cfc97679870422e3a1ab0784960c60514d816271415c306ee3a3ed1a77e31f6a885c3cb
+b5a7f63abb94d07d1a6445c36c07c7e8327fe61b1647e391b4c7edae5de57a3d 00000000000000000000000000000000000000000000000000000000000000000000000000000000000000
+80563c40ed46575a9e44820d93ee095e2851aa22483fd67837118c6cd951ba61 00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
+5bb40c5880cef2f739bdb6aaaf16161eaae55530e7b10d7ea996b751a299e949 8519e039172b0d70e5ca7b3383d6b3167315a422747b73f019cf9528f0fde341fd0f2a63030ba6450525cf6de31837669af6f1df8131faf50aaab8d3a7405589
+613e638505ba1fd05f428d5c9f8e08f8165614342dac419adc6a47dce257eb3e 37a636d7dafdf259b7287eddca2f58099e98619d2f99bdb8969d7b14498102cc065201c8be90bd777323f449848b215d2977c92c4c1c2da36ab46b2e389689ed97c18fec08cd3b03235c5e4c62a37ad88c7b67932495a71090e85dd4020a9300
+ed082e49dbd5bbe34a3726a0d1dad981146062b39d36d62c71eb1ed8ab49459b 38274c97c45a172cfc97679870422e3a1ab0784960c60514d816271415c306ee3a3ed1a77e31f6a885c3cb