diff options
Diffstat (limited to '')
-rw-r--r-- | README.md | 4 | ||||
-rw-r--r-- | python2/sha3.py | 221 | ||||
-rw-r--r-- | python3/sha3.py | 221 |
3 files changed, 226 insertions, 220 deletions
@@ -21,8 +21,8 @@ PERFORMANCE COMPARISON: C ISO C90 :: 0,082s ~ 1 Java/C JNI :: 0,175s ~ 2,13 Java 1.2+ :: 0,228s ~ 2,78 - Python 3 :: 25,057s ~ 306 - Python 2 :: 37,488s ~ 457 + Python 3 :: 24,373s ~ 297 + Python 2 :: 34,595s ~ 422 Based on test against a 2,3 MB file. diff --git a/python2/sha3.py b/python2/sha3.py index a79fd79..7b62563 100644 --- a/python2/sha3.py +++ b/python2/sha3.py @@ -280,36 +280,35 @@ class SHA3: self.keccakFRound(A, self.RC[i] & self.wmod) - def toLane(self, message, rr, ww, off): + def toLane(self, message, n, ww, off): ''' Convert a chunk of byte:s to a word @param message:bytes The message - @param rr:int Bitrate in bytes + @param n:int `min(len(message), rr)` + rr:int Bitrate in bytes @param ww:int Word size in bytes @param off:int The offset in the message @return :int Lane ''' rc = 0 i = off + ww - 1 - n = min(len(message), rr) while i >= off: rc = (rc << 8) | (message[i] if (i < n) else 0) i -= 1 return rc - def toLane64(self, message, rr, off): + def toLane64(self, message, n, off): ''' Convert a chunk of byte:s to a 64-bit word @param message:bytes The message - @param rr:int Bitrate in bytes + @param n:int `min(len(message), rr)` + rr:int Bitrate in bytes @param off:int The offset in the message @return :int Lane ''' - n = min(len(message), rr) - return ((message[off + 7] << 56) if (off + 7 < n) else 0) | \ ((message[off + 6] << 48) if (off + 6 < n) else 0) | \ ((message[off + 5] << 40) if (off + 5 < n) else 0) | \ @@ -392,62 +391,64 @@ class SHA3: # Absorbing phase if ww == 8: for i in range(0, nnn, rr): - self.S[ 0] ^= self.toLane64(message, rr, 0) - self.S[ 5] ^= self.toLane64(message, rr, 8) - self.S[10] ^= self.toLane64(message, rr, 16) - self.S[15] ^= self.toLane64(message, rr, 24) - self.S[20] ^= self.toLane64(message, rr, 32) - self.S[ 1] ^= self.toLane64(message, rr, 40) - self.S[ 6] ^= self.toLane64(message, rr, 48) - self.S[11] ^= self.toLane64(message, rr, 56) - self.S[16] ^= self.toLane64(message, rr, 64) - self.S[21] ^= self.toLane64(message, rr, 72) - self.S[ 2] ^= self.toLane64(message, rr, 80) - self.S[ 7] ^= self.toLane64(message, rr, 88) - self.S[12] ^= self.toLane64(message, rr, 96) - self.S[17] ^= self.toLane64(message, rr, 104) - self.S[22] ^= self.toLane64(message, rr, 112) - self.S[ 3] ^= self.toLane64(message, rr, 120) - self.S[ 8] ^= self.toLane64(message, rr, 128) - self.S[13] ^= self.toLane64(message, rr, 136) - self.S[18] ^= self.toLane64(message, rr, 144) - self.S[23] ^= self.toLane64(message, rr, 152) - self.S[ 4] ^= self.toLane64(message, rr, 160) - self.S[ 9] ^= self.toLane64(message, rr, 168) - self.S[14] ^= self.toLane64(message, rr, 176) - self.S[19] ^= self.toLane64(message, rr, 184) - self.S[24] ^= self.toLane64(message, rr, 192) + n = min(len(message), rr) + self.S[ 0] ^= self.toLane64(message, n, 0) + self.S[ 5] ^= self.toLane64(message, n, 8) + self.S[10] ^= self.toLane64(message, n, 16) + self.S[15] ^= self.toLane64(message, n, 24) + self.S[20] ^= self.toLane64(message, n, 32) + self.S[ 1] ^= self.toLane64(message, n, 40) + self.S[ 6] ^= self.toLane64(message, n, 48) + self.S[11] ^= self.toLane64(message, n, 56) + self.S[16] ^= self.toLane64(message, n, 64) + self.S[21] ^= self.toLane64(message, n, 72) + self.S[ 2] ^= self.toLane64(message, n, 80) + self.S[ 7] ^= self.toLane64(message, n, 88) + self.S[12] ^= self.toLane64(message, n, 96) + self.S[17] ^= self.toLane64(message, n, 104) + self.S[22] ^= self.toLane64(message, n, 112) + self.S[ 3] ^= self.toLane64(message, n, 120) + self.S[ 8] ^= self.toLane64(message, n, 128) + self.S[13] ^= self.toLane64(message, n, 136) + self.S[18] ^= self.toLane64(message, n, 144) + self.S[23] ^= self.toLane64(message, n, 152) + self.S[ 4] ^= self.toLane64(message, n, 160) + self.S[ 9] ^= self.toLane64(message, n, 168) + self.S[14] ^= self.toLane64(message, n, 176) + self.S[19] ^= self.toLane64(message, n, 184) + self.S[24] ^= self.toLane64(message, n, 192) self.keccakF(self.S) message = message[rr:] else: for i in range(0, nnn, rr): - self.S[ 0] ^= self.toLane(message, rr, ww, 0) - self.S[ 5] ^= self.toLane(message, rr, ww, ww) - self.S[10] ^= self.toLane(message, rr, ww, 2 * ww) - self.S[15] ^= self.toLane(message, rr, ww, 3 * ww) - self.S[20] ^= self.toLane(message, rr, ww, 4 * ww) - self.S[ 1] ^= self.toLane(message, rr, ww, 5 * ww) - self.S[ 6] ^= self.toLane(message, rr, ww, 6 * ww) - self.S[11] ^= self.toLane(message, rr, ww, 7 * ww) - self.S[16] ^= self.toLane(message, rr, ww, 8 * ww) - self.S[21] ^= self.toLane(message, rr, ww, 9 * ww) - self.S[ 2] ^= self.toLane(message, rr, ww, 10 * ww) - self.S[ 7] ^= self.toLane(message, rr, ww, 11 * ww) - self.S[12] ^= self.toLane(message, rr, ww, 12 * ww) - self.S[17] ^= self.toLane(message, rr, ww, 13 * ww) - self.S[22] ^= self.toLane(message, rr, ww, 14 * ww) - self.S[ 3] ^= self.toLane(message, rr, ww, 15 * ww) - self.S[ 8] ^= self.toLane(message, rr, ww, 16 * ww) - self.S[13] ^= self.toLane(message, rr, ww, 17 * ww) - self.S[18] ^= self.toLane(message, rr, ww, 18 * ww) - self.S[23] ^= self.toLane(message, rr, ww, 19 * ww) - self.S[ 4] ^= self.toLane(message, rr, ww, 20 * ww) - self.S[ 9] ^= self.toLane(message, rr, ww, 21 * ww) - self.S[14] ^= self.toLane(message, rr, ww, 22 * ww) - self.S[19] ^= self.toLane(message, rr, ww, 23 * ww) - self.S[24] ^= self.toLane(message, rr, ww, 24 * ww) - message = message[rr:] + n = min(len(message), rr) + self.S[ 0] ^= self.toLane(message, n, ww, 0) + self.S[ 5] ^= self.toLane(message, n, ww, ww) + self.S[10] ^= self.toLane(message, n, ww, 2 * ww) + self.S[15] ^= self.toLane(message, n, ww, 3 * ww) + self.S[20] ^= self.toLane(message, n, ww, 4 * ww) + self.S[ 1] ^= self.toLane(message, n, ww, 5 * ww) + self.S[ 6] ^= self.toLane(message, n, ww, 6 * ww) + self.S[11] ^= self.toLane(message, n, ww, 7 * ww) + self.S[16] ^= self.toLane(message, n, ww, 8 * ww) + self.S[21] ^= self.toLane(message, n, ww, 9 * ww) + self.S[ 2] ^= self.toLane(message, n, ww, 10 * ww) + self.S[ 7] ^= self.toLane(message, n, ww, 11 * ww) + self.S[12] ^= self.toLane(message, n, ww, 12 * ww) + self.S[17] ^= self.toLane(message, n, ww, 13 * ww) + self.S[22] ^= self.toLane(message, n, ww, 14 * ww) + self.S[ 3] ^= self.toLane(message, n, ww, 15 * ww) + self.S[ 8] ^= self.toLane(message, n, ww, 16 * ww) + self.S[13] ^= self.toLane(message, n, ww, 17 * ww) + self.S[18] ^= self.toLane(message, n, ww, 18 * ww) + self.S[23] ^= self.toLane(message, n, ww, 19 * ww) + self.S[ 4] ^= self.toLane(message, n, ww, 20 * ww) + self.S[ 9] ^= self.toLane(message, n, ww, 21 * ww) + self.S[14] ^= self.toLane(message, n, ww, 22 * ww) + self.S[19] ^= self.toLane(message, n, ww, 23 * ww) + self.S[24] ^= self.toLane(message, n, ww, 24 * ww) self.keccakF(self.S) + message = message[rr:] def digest(self, msg = None, msglen = None, withReturn = None): @@ -478,62 +479,64 @@ class SHA3: # Absorbing phase if ww == 8: for i in range(0, nnn, rr): - self.S[ 0] ^= self.toLane64(message, rr, 0) - self.S[ 5] ^= self.toLane64(message, rr, 8) - self.S[10] ^= self.toLane64(message, rr, 16) - self.S[15] ^= self.toLane64(message, rr, 24) - self.S[20] ^= self.toLane64(message, rr, 32) - self.S[ 1] ^= self.toLane64(message, rr, 40) - self.S[ 6] ^= self.toLane64(message, rr, 48) - self.S[11] ^= self.toLane64(message, rr, 56) - self.S[16] ^= self.toLane64(message, rr, 64) - self.S[21] ^= self.toLane64(message, rr, 72) - self.S[ 2] ^= self.toLane64(message, rr, 80) - self.S[ 7] ^= self.toLane64(message, rr, 88) - self.S[12] ^= self.toLane64(message, rr, 96) - self.S[17] ^= self.toLane64(message, rr, 104) - self.S[22] ^= self.toLane64(message, rr, 112) - self.S[ 3] ^= self.toLane64(message, rr, 120) - self.S[ 8] ^= self.toLane64(message, rr, 128) - self.S[13] ^= self.toLane64(message, rr, 136) - self.S[18] ^= self.toLane64(message, rr, 144) - self.S[23] ^= self.toLane64(message, rr, 152) - self.S[ 4] ^= self.toLane64(message, rr, 160) - self.S[ 9] ^= self.toLane64(message, rr, 168) - self.S[14] ^= self.toLane64(message, rr, 176) - self.S[19] ^= self.toLane64(message, rr, 184) - self.S[24] ^= self.toLane64(message, rr, 192) + n = min(len(message), rr) + self.S[ 0] ^= self.toLane64(message, n, 0) + self.S[ 5] ^= self.toLane64(message, n, 8) + self.S[10] ^= self.toLane64(message, n, 16) + self.S[15] ^= self.toLane64(message, n, 24) + self.S[20] ^= self.toLane64(message, n, 32) + self.S[ 1] ^= self.toLane64(message, n, 40) + self.S[ 6] ^= self.toLane64(message, n, 48) + self.S[11] ^= self.toLane64(message, n, 56) + self.S[16] ^= self.toLane64(message, n, 64) + self.S[21] ^= self.toLane64(message, n, 72) + self.S[ 2] ^= self.toLane64(message, n, 80) + self.S[ 7] ^= self.toLane64(message, n, 88) + self.S[12] ^= self.toLane64(message, n, 96) + self.S[17] ^= self.toLane64(message, n, 104) + self.S[22] ^= self.toLane64(message, n, 112) + self.S[ 3] ^= self.toLane64(message, n, 120) + self.S[ 8] ^= self.toLane64(message, n, 128) + self.S[13] ^= self.toLane64(message, n, 136) + self.S[18] ^= self.toLane64(message, n, 144) + self.S[23] ^= self.toLane64(message, n, 152) + self.S[ 4] ^= self.toLane64(message, n, 160) + self.S[ 9] ^= self.toLane64(message, n, 168) + self.S[14] ^= self.toLane64(message, n, 176) + self.S[19] ^= self.toLane64(message, n, 184) + self.S[24] ^= self.toLane64(message, n, 192) self.keccakF(self.S) message = message[rr:] else: for i in range(0, nnn, rr): - self.S[ 0] ^= self.toLane(message, rr, ww, 0) - self.S[ 5] ^= self.toLane(message, rr, ww, ww) - self.S[10] ^= self.toLane(message, rr, ww, 2 * ww) - self.S[15] ^= self.toLane(message, rr, ww, 3 * ww) - self.S[20] ^= self.toLane(message, rr, ww, 4 * ww) - self.S[ 1] ^= self.toLane(message, rr, ww, 5 * ww) - self.S[ 6] ^= self.toLane(message, rr, ww, 6 * ww) - self.S[11] ^= self.toLane(message, rr, ww, 7 * ww) - self.S[16] ^= self.toLane(message, rr, ww, 8 * ww) - self.S[21] ^= self.toLane(message, rr, ww, 9 * ww) - self.S[ 2] ^= self.toLane(message, rr, ww, 10 * ww) - self.S[ 7] ^= self.toLane(message, rr, ww, 11 * ww) - self.S[12] ^= self.toLane(message, rr, ww, 12 * ww) - self.S[17] ^= self.toLane(message, rr, ww, 13 * ww) - self.S[22] ^= self.toLane(message, rr, ww, 14 * ww) - self.S[ 3] ^= self.toLane(message, rr, ww, 15 * ww) - self.S[ 8] ^= self.toLane(message, rr, ww, 16 * ww) - self.S[13] ^= self.toLane(message, rr, ww, 17 * ww) - self.S[18] ^= self.toLane(message, rr, ww, 18 * ww) - self.S[23] ^= self.toLane(message, rr, ww, 19 * ww) - self.S[ 4] ^= self.toLane(message, rr, ww, 20 * ww) - self.S[ 9] ^= self.toLane(message, rr, ww, 21 * ww) - self.S[14] ^= self.toLane(message, rr, ww, 22 * ww) - self.S[19] ^= self.toLane(message, rr, ww, 23 * ww) - self.S[24] ^= self.toLane(message, rr, ww, 24 * ww) - message = message[rr:] + n = min(len(message), rr) + self.S[ 0] ^= self.toLane(message, n, ww, 0) + self.S[ 5] ^= self.toLane(message, n, ww, ww) + self.S[10] ^= self.toLane(message, n, ww, 2 * ww) + self.S[15] ^= self.toLane(message, n, ww, 3 * ww) + self.S[20] ^= self.toLane(message, n, ww, 4 * ww) + self.S[ 1] ^= self.toLane(message, n, ww, 5 * ww) + self.S[ 6] ^= self.toLane(message, n, ww, 6 * ww) + self.S[11] ^= self.toLane(message, n, ww, 7 * ww) + self.S[16] ^= self.toLane(message, n, ww, 8 * ww) + self.S[21] ^= self.toLane(message, n, ww, 9 * ww) + self.S[ 2] ^= self.toLane(message, n, ww, 10 * ww) + self.S[ 7] ^= self.toLane(message, n, ww, 11 * ww) + self.S[12] ^= self.toLane(message, n, ww, 12 * ww) + self.S[17] ^= self.toLane(message, n, ww, 13 * ww) + self.S[22] ^= self.toLane(message, n, ww, 14 * ww) + self.S[ 3] ^= self.toLane(message, n, ww, 15 * ww) + self.S[ 8] ^= self.toLane(message, n, ww, 16 * ww) + self.S[13] ^= self.toLane(message, n, ww, 17 * ww) + self.S[18] ^= self.toLane(message, n, ww, 18 * ww) + self.S[23] ^= self.toLane(message, n, ww, 19 * ww) + self.S[ 4] ^= self.toLane(message, n, ww, 20 * ww) + self.S[ 9] ^= self.toLane(message, n, ww, 21 * ww) + self.S[14] ^= self.toLane(message, n, ww, 22 * ww) + self.S[19] ^= self.toLane(message, n, ww, 23 * ww) + self.S[24] ^= self.toLane(message, n, ww, 24 * ww) self.keccakF(self.S) + message = message[rr:] # Squeezing phase if withReturn: diff --git a/python3/sha3.py b/python3/sha3.py index a2e301d..ca34d5b 100644 --- a/python3/sha3.py +++ b/python3/sha3.py @@ -280,36 +280,35 @@ class SHA3: self.keccakFRound(A, self.RC[i] & self.wmod) - def toLane(self, message, rr, ww, off): + def toLane(self, message, n, ww, off): ''' Convert a chunk of byte:s to a word @param message:bytes The message - @param rr:int Bitrate in bytes + @param n:int `min(len(message), rr)` + rr:int Bitrate in bytes @param ww:int Word size in bytes @param off:int The offset in the message @return :int Lane ''' rc = 0 i = off + ww - 1 - n = min(len(message), rr) while i >= off: rc = (rc << 8) | (message[i] if (i < n) else 0) i -= 1 return rc - def toLane64(self, message, rr, off): + def toLane64(self, message, n, off): ''' Convert a chunk of byte:s to a 64-bit word @param message:bytes The message - @param rr:int Bitrate in bytes + @param n:int `min(len(message), rr)` + rr:int Bitrate in bytes @param off:int The offset in the message @return :int Lane ''' - n = min(len(message), rr) - return ((message[off + 7] << 56) if (off + 7 < n) else 0) | ((message[off + 6] << 48) if (off + 6 < n) else 0) | ((message[off + 5] << 40) if (off + 5 < n) else 0) | ((message[off + 4] << 32) if (off + 4 < n) else 0) | ((message[off + 3] << 24) if (off + 3 < n) else 0) | ((message[off + 2] << 16) if (off + 2 < n) else 0) | ((message[off + 1] << 8) if (off + 1 < n) else 0) | ((message[off]) if (off < n) else 0) @@ -385,62 +384,64 @@ class SHA3: # Absorbing phase if ww == 8: for i in range(0, nnn, rr): - self.S[ 0] ^= self.toLane64(message, rr, 0) - self.S[ 5] ^= self.toLane64(message, rr, 8) - self.S[10] ^= self.toLane64(message, rr, 16) - self.S[15] ^= self.toLane64(message, rr, 24) - self.S[20] ^= self.toLane64(message, rr, 32) - self.S[ 1] ^= self.toLane64(message, rr, 40) - self.S[ 6] ^= self.toLane64(message, rr, 48) - self.S[11] ^= self.toLane64(message, rr, 56) - self.S[16] ^= self.toLane64(message, rr, 64) - self.S[21] ^= self.toLane64(message, rr, 72) - self.S[ 2] ^= self.toLane64(message, rr, 80) - self.S[ 7] ^= self.toLane64(message, rr, 88) - self.S[12] ^= self.toLane64(message, rr, 96) - self.S[17] ^= self.toLane64(message, rr, 104) - self.S[22] ^= self.toLane64(message, rr, 112) - self.S[ 3] ^= self.toLane64(message, rr, 120) - self.S[ 8] ^= self.toLane64(message, rr, 128) - self.S[13] ^= self.toLane64(message, rr, 136) - self.S[18] ^= self.toLane64(message, rr, 144) - self.S[23] ^= self.toLane64(message, rr, 152) - self.S[ 4] ^= self.toLane64(message, rr, 160) - self.S[ 9] ^= self.toLane64(message, rr, 168) - self.S[14] ^= self.toLane64(message, rr, 176) - self.S[19] ^= self.toLane64(message, rr, 184) - self.S[24] ^= self.toLane64(message, rr, 192) + n = min(len(message), rr) + self.S[ 0] ^= self.toLane64(message, n, 0) + self.S[ 5] ^= self.toLane64(message, n, 8) + self.S[10] ^= self.toLane64(message, n, 16) + self.S[15] ^= self.toLane64(message, n, 24) + self.S[20] ^= self.toLane64(message, n, 32) + self.S[ 1] ^= self.toLane64(message, n, 40) + self.S[ 6] ^= self.toLane64(message, n, 48) + self.S[11] ^= self.toLane64(message, n, 56) + self.S[16] ^= self.toLane64(message, n, 64) + self.S[21] ^= self.toLane64(message, n, 72) + self.S[ 2] ^= self.toLane64(message, n, 80) + self.S[ 7] ^= self.toLane64(message, n, 88) + self.S[12] ^= self.toLane64(message, n, 96) + self.S[17] ^= self.toLane64(message, n, 104) + self.S[22] ^= self.toLane64(message, n, 112) + self.S[ 3] ^= self.toLane64(message, n, 120) + self.S[ 8] ^= self.toLane64(message, n, 128) + self.S[13] ^= self.toLane64(message, n, 136) + self.S[18] ^= self.toLane64(message, n, 144) + self.S[23] ^= self.toLane64(message, n, 152) + self.S[ 4] ^= self.toLane64(message, n, 160) + self.S[ 9] ^= self.toLane64(message, n, 168) + self.S[14] ^= self.toLane64(message, n, 176) + self.S[19] ^= self.toLane64(message, n, 184) + self.S[24] ^= self.toLane64(message, n, 192) self.keccakF(self.S) message = message[rr:] else: for i in range(0, nnn, rr): - self.S[ 0] ^= self.toLane(message, rr, ww, 0) - self.S[ 5] ^= self.toLane(message, rr, ww, ww) - self.S[10] ^= self.toLane(message, rr, ww, 2 * ww) - self.S[15] ^= self.toLane(message, rr, ww, 3 * ww) - self.S[20] ^= self.toLane(message, rr, ww, 4 * ww) - self.S[ 1] ^= self.toLane(message, rr, ww, 5 * ww) - self.S[ 6] ^= self.toLane(message, rr, ww, 6 * ww) - self.S[11] ^= self.toLane(message, rr, ww, 7 * ww) - self.S[16] ^= self.toLane(message, rr, ww, 8 * ww) - self.S[21] ^= self.toLane(message, rr, ww, 9 * ww) - self.S[ 2] ^= self.toLane(message, rr, ww, 10 * ww) - self.S[ 7] ^= self.toLane(message, rr, ww, 11 * ww) - self.S[12] ^= self.toLane(message, rr, ww, 12 * ww) - self.S[17] ^= self.toLane(message, rr, ww, 13 * ww) - self.S[22] ^= self.toLane(message, rr, ww, 14 * ww) - self.S[ 3] ^= self.toLane(message, rr, ww, 15 * ww) - self.S[ 8] ^= self.toLane(message, rr, ww, 16 * ww) - self.S[13] ^= self.toLane(message, rr, ww, 17 * ww) - self.S[18] ^= self.toLane(message, rr, ww, 18 * ww) - self.S[23] ^= self.toLane(message, rr, ww, 19 * ww) - self.S[ 4] ^= self.toLane(message, rr, ww, 20 * ww) - self.S[ 9] ^= self.toLane(message, rr, ww, 21 * ww) - self.S[14] ^= self.toLane(message, rr, ww, 22 * ww) - self.S[19] ^= self.toLane(message, rr, ww, 23 * ww) - self.S[24] ^= self.toLane(message, rr, ww, 24 * ww) - message = message[rr:] + n = min(len(message), rr) + self.S[ 0] ^= self.toLane(message, n, ww, 0) + self.S[ 5] ^= self.toLane(message, n, ww, ww) + self.S[10] ^= self.toLane(message, n, ww, 2 * ww) + self.S[15] ^= self.toLane(message, n, ww, 3 * ww) + self.S[20] ^= self.toLane(message, n, ww, 4 * ww) + self.S[ 1] ^= self.toLane(message, n, ww, 5 * ww) + self.S[ 6] ^= self.toLane(message, n, ww, 6 * ww) + self.S[11] ^= self.toLane(message, n, ww, 7 * ww) + self.S[16] ^= self.toLane(message, n, ww, 8 * ww) + self.S[21] ^= self.toLane(message, n, ww, 9 * ww) + self.S[ 2] ^= self.toLane(message, n, ww, 10 * ww) + self.S[ 7] ^= self.toLane(message, n, ww, 11 * ww) + self.S[12] ^= self.toLane(message, n, ww, 12 * ww) + self.S[17] ^= self.toLane(message, n, ww, 13 * ww) + self.S[22] ^= self.toLane(message, n, ww, 14 * ww) + self.S[ 3] ^= self.toLane(message, n, ww, 15 * ww) + self.S[ 8] ^= self.toLane(message, n, ww, 16 * ww) + self.S[13] ^= self.toLane(message, n, ww, 17 * ww) + self.S[18] ^= self.toLane(message, n, ww, 18 * ww) + self.S[23] ^= self.toLane(message, n, ww, 19 * ww) + self.S[ 4] ^= self.toLane(message, n, ww, 20 * ww) + self.S[ 9] ^= self.toLane(message, n, ww, 21 * ww) + self.S[14] ^= self.toLane(message, n, ww, 22 * ww) + self.S[19] ^= self.toLane(message, n, ww, 23 * ww) + self.S[24] ^= self.toLane(message, n, ww, 24 * ww) self.keccakF(self.S) + message = message[rr:] def digest(self, msg = None, msglen = None, withReturn = None): @@ -471,62 +472,64 @@ class SHA3: # Absorbing phase if ww == 8: for i in range(0, nnn, rr): - self.S[ 0] ^= self.toLane64(message, rr, 0) - self.S[ 5] ^= self.toLane64(message, rr, 8) - self.S[10] ^= self.toLane64(message, rr, 16) - self.S[15] ^= self.toLane64(message, rr, 24) - self.S[20] ^= self.toLane64(message, rr, 32) - self.S[ 1] ^= self.toLane64(message, rr, 40) - self.S[ 6] ^= self.toLane64(message, rr, 48) - self.S[11] ^= self.toLane64(message, rr, 56) - self.S[16] ^= self.toLane64(message, rr, 64) - self.S[21] ^= self.toLane64(message, rr, 72) - self.S[ 2] ^= self.toLane64(message, rr, 80) - self.S[ 7] ^= self.toLane64(message, rr, 88) - self.S[12] ^= self.toLane64(message, rr, 96) - self.S[17] ^= self.toLane64(message, rr, 104) - self.S[22] ^= self.toLane64(message, rr, 112) - self.S[ 3] ^= self.toLane64(message, rr, 120) - self.S[ 8] ^= self.toLane64(message, rr, 128) - self.S[13] ^= self.toLane64(message, rr, 136) - self.S[18] ^= self.toLane64(message, rr, 144) - self.S[23] ^= self.toLane64(message, rr, 152) - self.S[ 4] ^= self.toLane64(message, rr, 160) - self.S[ 9] ^= self.toLane64(message, rr, 168) - self.S[14] ^= self.toLane64(message, rr, 176) - self.S[19] ^= self.toLane64(message, rr, 184) - self.S[24] ^= self.toLane64(message, rr, 192) + n = min(len(message), rr) + self.S[ 0] ^= self.toLane64(message, n, 0) + self.S[ 5] ^= self.toLane64(message, n, 8) + self.S[10] ^= self.toLane64(message, n, 16) + self.S[15] ^= self.toLane64(message, n, 24) + self.S[20] ^= self.toLane64(message, n, 32) + self.S[ 1] ^= self.toLane64(message, n, 40) + self.S[ 6] ^= self.toLane64(message, n, 48) + self.S[11] ^= self.toLane64(message, n, 56) + self.S[16] ^= self.toLane64(message, n, 64) + self.S[21] ^= self.toLane64(message, n, 72) + self.S[ 2] ^= self.toLane64(message, n, 80) + self.S[ 7] ^= self.toLane64(message, n, 88) + self.S[12] ^= self.toLane64(message, n, 96) + self.S[17] ^= self.toLane64(message, n, 104) + self.S[22] ^= self.toLane64(message, n, 112) + self.S[ 3] ^= self.toLane64(message, n, 120) + self.S[ 8] ^= self.toLane64(message, n, 128) + self.S[13] ^= self.toLane64(message, n, 136) + self.S[18] ^= self.toLane64(message, n, 144) + self.S[23] ^= self.toLane64(message, n, 152) + self.S[ 4] ^= self.toLane64(message, n, 160) + self.S[ 9] ^= self.toLane64(message, n, 168) + self.S[14] ^= self.toLane64(message, n, 176) + self.S[19] ^= self.toLane64(message, n, 184) + self.S[24] ^= self.toLane64(message, n, 192) self.keccakF(self.S) message = message[rr:] else: for i in range(0, nnn, rr): - self.S[ 0] ^= self.toLane(message, rr, ww, 0) - self.S[ 5] ^= self.toLane(message, rr, ww, ww) - self.S[10] ^= self.toLane(message, rr, ww, 2 * ww) - self.S[15] ^= self.toLane(message, rr, ww, 3 * ww) - self.S[20] ^= self.toLane(message, rr, ww, 4 * ww) - self.S[ 1] ^= self.toLane(message, rr, ww, 5 * ww) - self.S[ 6] ^= self.toLane(message, rr, ww, 6 * ww) - self.S[11] ^= self.toLane(message, rr, ww, 7 * ww) - self.S[16] ^= self.toLane(message, rr, ww, 8 * ww) - self.S[21] ^= self.toLane(message, rr, ww, 9 * ww) - self.S[ 2] ^= self.toLane(message, rr, ww, 10 * ww) - self.S[ 7] ^= self.toLane(message, rr, ww, 11 * ww) - self.S[12] ^= self.toLane(message, rr, ww, 12 * ww) - self.S[17] ^= self.toLane(message, rr, ww, 13 * ww) - self.S[22] ^= self.toLane(message, rr, ww, 14 * ww) - self.S[ 3] ^= self.toLane(message, rr, ww, 15 * ww) - self.S[ 8] ^= self.toLane(message, rr, ww, 16 * ww) - self.S[13] ^= self.toLane(message, rr, ww, 17 * ww) - self.S[18] ^= self.toLane(message, rr, ww, 18 * ww) - self.S[23] ^= self.toLane(message, rr, ww, 19 * ww) - self.S[ 4] ^= self.toLane(message, rr, ww, 20 * ww) - self.S[ 9] ^= self.toLane(message, rr, ww, 21 * ww) - self.S[14] ^= self.toLane(message, rr, ww, 22 * ww) - self.S[19] ^= self.toLane(message, rr, ww, 23 * ww) - self.S[24] ^= self.toLane(message, rr, ww, 24 * ww) - message = message[rr:] + n = min(len(message), rr) + self.S[ 0] ^= self.toLane(message, n, ww, 0) + self.S[ 5] ^= self.toLane(message, n, ww, ww) + self.S[10] ^= self.toLane(message, n, ww, 2 * ww) + self.S[15] ^= self.toLane(message, n, ww, 3 * ww) + self.S[20] ^= self.toLane(message, n, ww, 4 * ww) + self.S[ 1] ^= self.toLane(message, n, ww, 5 * ww) + self.S[ 6] ^= self.toLane(message, n, ww, 6 * ww) + self.S[11] ^= self.toLane(message, n, ww, 7 * ww) + self.S[16] ^= self.toLane(message, n, ww, 8 * ww) + self.S[21] ^= self.toLane(message, n, ww, 9 * ww) + self.S[ 2] ^= self.toLane(message, n, ww, 10 * ww) + self.S[ 7] ^= self.toLane(message, n, ww, 11 * ww) + self.S[12] ^= self.toLane(message, n, ww, 12 * ww) + self.S[17] ^= self.toLane(message, n, ww, 13 * ww) + self.S[22] ^= self.toLane(message, n, ww, 14 * ww) + self.S[ 3] ^= self.toLane(message, n, ww, 15 * ww) + self.S[ 8] ^= self.toLane(message, n, ww, 16 * ww) + self.S[13] ^= self.toLane(message, n, ww, 17 * ww) + self.S[18] ^= self.toLane(message, n, ww, 18 * ww) + self.S[23] ^= self.toLane(message, n, ww, 19 * ww) + self.S[ 4] ^= self.toLane(message, n, ww, 20 * ww) + self.S[ 9] ^= self.toLane(message, n, ww, 21 * ww) + self.S[14] ^= self.toLane(message, n, ww, 22 * ww) + self.S[19] ^= self.toLane(message, n, ww, 23 * ww) + self.S[24] ^= self.toLane(message, n, ww, 24 * ww) self.keccakF(self.S) + message = message[rr:] # Squeezing phase if withReturn: |