Expand the first 12 iterations of phi, though more simplification is needed.

1 files changed, 58 insertions, 2 deletions

diff --git a/src/hash/gost_3411/gost_3411.cpp b/src/hash/gost_3411/gost_3411.cpp
index 25118752f..0ac875532 100644
--- a/src/hash/gost_3411/gost_3411.cpp
+++ b/src/hash/gost_3411/gost_3411.cpp
@@ -137,8 +137,64 @@ void GOST_34_11::compress_n(const byte input[], u32bit blocks)
          V[3] = AA_V_2;
          }
 
-      for(u32bit j = 0; j != 12; ++j)
-         psi(S);
+      byte X0  = S[ 0] ^ S[ 2] ^ S[ 4] ^ S[ 6] ^ S[24] ^ S[30];
+      byte X1  = S[ 1] ^ S[ 3] ^ S[ 5] ^ S[ 7] ^ S[25] ^ S[31];
+      byte X2  = S[ 0] ^ S[ 8] ^ S[24] ^ S[26] ^ S[30];
+      byte X3  = S[ 1] ^ S[ 9] ^ S[25] ^ S[27] ^ S[31];
+      byte X4  = S[ 4] ^ S[ 6] ^ S[10] ^ S[28] ^ S[ 0] ^ S[24] ^ S[26] ^ S[30];
+      byte X5  = S[ 5] ^ S[ 7] ^ S[11] ^ S[29] ^ S[27] ^ S[ 1] ^ S[25] ^ S[31];
+      byte X6  = S[ 6] ^ S[ 8] ^ S[10] ^ S[12] ^ S[30] ^ X4;
+      byte X7  = S[ 7] ^ S[ 9] ^ S[11] ^ S[13] ^ S[31] ^ X5;
+      byte X8  = S[ 8] ^ S[10] ^ S[12] ^ S[14] ^ X0 ^ X6;
+      byte X9  = S[ 9] ^ S[11] ^ S[13] ^ S[15] ^ X1 ^ X7;
+      byte X10 = S[10] ^ S[12] ^ S[14] ^ S[16] ^ X2 ^ X8;
+      byte X11 = S[11] ^ S[13] ^ S[15] ^ S[17] ^ X3 ^ X9;
+      byte X12 = S[12] ^ S[14] ^ S[16] ^ S[18] ^ X4 ^ X10;
+      byte X13 = S[13] ^ S[15] ^ S[17] ^ S[19] ^ X5 ^ X11;
+      byte X14 = S[14] ^ S[16] ^ S[18] ^ S[20] ^ X6 ^ X12;
+      byte X15 = S[15] ^ S[17] ^ S[19] ^ S[21] ^ X7 ^ X13;
+      byte X16 = S[16] ^ S[18] ^ S[20] ^ S[22] ^ X8 ^ X14;
+      byte X17 = S[17] ^ S[19] ^ S[21] ^ S[23] ^ X9 ^ X15;
+      byte X18 = S[18] ^ S[20] ^ S[22] ^ S[24] ^ X10 ^ X16;
+      byte X19 = S[19] ^ S[21] ^ S[23] ^ S[25] ^ X11 ^ X17;
+      byte X20 = S[20] ^ S[22] ^ S[24] ^ S[26] ^ X12 ^ X18;
+      byte X21 = S[21] ^ S[23] ^ S[25] ^ S[27] ^ X13 ^ X19;
+      byte X22 = S[22] ^ S[24] ^ S[26] ^ S[28] ^ X14 ^ X20;
+      byte X23 = S[23] ^ S[25] ^ S[27] ^ S[29] ^ X15 ^ X21;
+
+      S[0] = S[24];
+      S[1] = S[25];
+      S[2] = S[26];
+      S[3] = S[27];
+      S[4] = S[28];
+      S[5] = S[29];
+      S[6] = S[30];
+      S[7] = S[31];
+      S[8] = X0;
+      S[9] = X1;
+      S[10] = X2;
+      S[11] = X3;
+      S[12] = X4;
+      S[13] = X5;
+      S[14] = X6;
+      S[15] = X7;
+      S[16] = X8;
+      S[17] = X9;
+      S[18] = X10;
+      S[19] = X11;
+      S[20] = X12;
+      S[21] = X13;
+      S[22] = X14;
+      S[23] = X15;
+      S[24] = X16;
+      S[25] = X17;
+      S[26] = X18;
+      S[27] = X19;
+      S[28] = X20;
+      S[29] = X21;
+      S[30] = X22;
+      S[31] = X23;
+
       xor_buf(S, input + 32*i, 32);
       psi(S);
       xor_buf(S, hash, 32);