Unroll the _encipher() loop. Saves ~80ms on my machine (at cost 12).

dcodeIO · Jan 20, 2017 · 0b7c11a · 0b7c11a
1 parent 9350b93
commit 0b7c11a
Showing 1 changed file with 94 additions and 0 deletions.
diff --git a/src/bcrypt/impl.js b/src/bcrypt/impl.js
@@ -275,6 +275,8 @@ function _encipher(lr, off, P, S) { // This is our bottleneck: 1714/1905 ticks /
         r = lr[off + 1];
 
     l ^= P[0];
+
+    /*
     for (var i=0, k=BLOWFISH_NUM_ROUNDS-2; i<=k;)
         // Feistel substitution on left word
         n  = S[l >>> 24],
@@ -288,6 +290,98 @@ function _encipher(lr, off, P, S) { // This is our bottleneck: 1714/1905 ticks /
         n ^= S[0x200 | ((r >> 8) & 0xff)],
         n += S[0x300 | (r & 0xff)],
         l ^= n ^ P[++i];
+    */
+
+    //The following is an unrolled version of the above loop.
+    //Iteration 0
+    n  = S[l >>> 24];
+    n += S[0x100 | ((l >> 16) & 0xff)];
+    n ^= S[0x200 | ((l >> 8) & 0xff)];
+    n += S[0x300 | (l & 0xff)];
+    r ^= n ^ P[1];
+    n  = S[r >>> 24];
+    n += S[0x100 | ((r >> 16) & 0xff)];
+    n ^= S[0x200 | ((r >> 8) & 0xff)];
+    n += S[0x300 | (r & 0xff)];
+    l ^= n ^ P[2];
+    //Iteration 1
+    n  = S[l >>> 24];
+    n += S[0x100 | ((l >> 16) & 0xff)];
+    n ^= S[0x200 | ((l >> 8) & 0xff)];
+    n += S[0x300 | (l & 0xff)];
+    r ^= n ^ P[3];
+    n  = S[r >>> 24];
+    n += S[0x100 | ((r >> 16) & 0xff)];
+    n ^= S[0x200 | ((r >> 8) & 0xff)];
+    n += S[0x300 | (r & 0xff)];
+    l ^= n ^ P[4];
+    //Iteration 2
+    n  = S[l >>> 24];
+    n += S[0x100 | ((l >> 16) & 0xff)];
+    n ^= S[0x200 | ((l >> 8) & 0xff)];
+    n += S[0x300 | (l & 0xff)];
+    r ^= n ^ P[5];
+    n  = S[r >>> 24];
+    n += S[0x100 | ((r >> 16) & 0xff)];
+    n ^= S[0x200 | ((r >> 8) & 0xff)];
+    n += S[0x300 | (r & 0xff)];
+    l ^= n ^ P[6];
+    //Iteration 3
+    n  = S[l >>> 24];
+    n += S[0x100 | ((l >> 16) & 0xff)];
+    n ^= S[0x200 | ((l >> 8) & 0xff)];
+    n += S[0x300 | (l & 0xff)];
+    r ^= n ^ P[7];
+    n  = S[r >>> 24];
+    n += S[0x100 | ((r >> 16) & 0xff)];
+    n ^= S[0x200 | ((r >> 8) & 0xff)];
+    n += S[0x300 | (r & 0xff)];
+    l ^= n ^ P[8];
+    //Iteration 4
+    n  = S[l >>> 24];
+    n += S[0x100 | ((l >> 16) & 0xff)];
+    n ^= S[0x200 | ((l >> 8) & 0xff)];
+    n += S[0x300 | (l & 0xff)];
+    r ^= n ^ P[9];
+    n  = S[r >>> 24];
+    n += S[0x100 | ((r >> 16) & 0xff)];
+    n ^= S[0x200 | ((r >> 8) & 0xff)];
+    n += S[0x300 | (r & 0xff)];
+    l ^= n ^ P[10];
+    //Iteration 5
+    n  = S[l >>> 24];
+    n += S[0x100 | ((l >> 16) & 0xff)];
+    n ^= S[0x200 | ((l >> 8) & 0xff)];
+    n += S[0x300 | (l & 0xff)];
+    r ^= n ^ P[11];
+    n  = S[r >>> 24];
+    n += S[0x100 | ((r >> 16) & 0xff)];
+    n ^= S[0x200 | ((r >> 8) & 0xff)];
+    n += S[0x300 | (r & 0xff)];
+    l ^= n ^ P[12];
+    //Iteration 6
+    n  = S[l >>> 24];
+    n += S[0x100 | ((l >> 16) & 0xff)];
+    n ^= S[0x200 | ((l >> 8) & 0xff)];
+    n += S[0x300 | (l & 0xff)];
+    r ^= n ^ P[13];
+    n  = S[r >>> 24];
+    n += S[0x100 | ((r >> 16) & 0xff)];
+    n ^= S[0x200 | ((r >> 8) & 0xff)];
+    n += S[0x300 | (r & 0xff)];
+    l ^= n ^ P[14];
+    //Iteration 7
+    n  = S[l >>> 24];
+    n += S[0x100 | ((l >> 16) & 0xff)];
+    n ^= S[0x200 | ((l >> 8) & 0xff)];
+    n += S[0x300 | (l & 0xff)];
+    r ^= n ^ P[15];
+    n  = S[r >>> 24];
+    n += S[0x100 | ((r >> 16) & 0xff)];
+    n ^= S[0x200 | ((r >> 8) & 0xff)];
+    n += S[0x300 | (r & 0xff)];
+    l ^= n ^ P[16];
+
     lr[off] = r ^ P[BLOWFISH_NUM_ROUNDS + 1];
     lr[off + 1] = l;
     return lr;