Revision 4:0da19393bd57, committed 2013-09-12

Comitter:

feb11

Date:

Thu Sep 12 10:18:57 2013 +0000

Parent:

3:85c6ee25cf3e

Child:

5:06cd9c8afa0b

Commit message:

improved performance of SHA-2 (32 & 64bits)

Changed in this revision

--- a/MD2.cpp	Wed Sep 11 17:22:40 2013 +0000
+++ b/MD2.cpp	Thu Sep 12 10:18:57 2013 +0000
@@ -1,3 +1,8 @@
+/**
+    Implementation of MD2 as described here:
+    http://tools.ietf.org/html/rfc1319
+*/
+
 #include "MD2.h"
 #include <string.h>
 

--- a/MD5.cpp	Wed Sep 11 17:22:40 2013 +0000
+++ b/MD5.cpp	Thu Sep 12 10:18:57 2013 +0000
@@ -1,3 +1,8 @@
+/**
+    Implementation of MD5 as described here:
+    http://tools.ietf.org/html/rfc1321
+*/
+
 #include "MD5.h"
 #include <string.h>
 

--- a/SHA1.cpp	Wed Sep 11 17:22:40 2013 +0000
+++ b/SHA1.cpp	Thu Sep 12 10:18:57 2013 +0000
@@ -1,3 +1,8 @@
+/**
+    Implementation of SHA-1 as described here:
+    http://tools.ietf.org/html/rfc1319
+*/
+
 #include "SHA1.h"
 #include <string.h>
 #include <stdio.h>

--- a/SHA2_32.cpp	Wed Sep 11 17:22:40 2013 +0000
+++ b/SHA2_32.cpp	Thu Sep 12 10:18:57 2013 +0000
@@ -3,25 +3,9 @@
 #include <stdio.h>
 #include <stdlib.h>
 
-static const uint32_t K[] =
-{
-    0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5,
-    0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5,
-    0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3,
-    0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174,
-    0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc,
-    0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da,
-    0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7,
-    0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967,
-    0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13,
-    0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85,
-    0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3,
-    0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070,
-    0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5,
-    0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3,
-    0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208,
-    0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2
-};
+
+static const uint8_t MASK = 0x0F;
+#define W(t) (w[(t)] = SSIG1(w[((t)+14)&MASK]) + w[((t)+9)&MASK] + SSIG0(w[((t)+1)&MASK]) + w[t])
 
 #define ROTL(W,N) (((W) << (N)) | ((W) >> (32-(N))))
 #define ROTR(W,N) (((W) >> (N)) | ((W) << (32-(N))))
@@ -31,7 +15,11 @@
 #define BSIG1(X) (ROTR(X,6) ^ ROTR(X,11) ^ ROTR(X,25))
 #define SSIG0(X) (ROTR((X),7) ^ ROTR((X),18) ^ ((X) >> 3))
 #define SSIG1(X) (ROTR((X),17) ^ ROTR((X),19) ^ ((X) >> 10))
-#define R(A,B,C,D,E,F,G,H,T)  T1 = H + BSIG1(E) + CH(E,F,G) + K[T] + w[T]; \
+#define R(A,B,C,D,E,F,G,H,T,K)  T1 = H + BSIG1(E) + CH(E,F,G) + K + (w[T] = __rev(buffer2[T])); \
+                              T2 = BSIG0(A) + MAJ(A,B,C); \
+                              D += T1; \
+                              H = T1 + T2;
+#define R2(A,B,C,D,E,F,G,H,T,K)  T1 = H + BSIG1(E) + CH(E,F,G) + K + W(T&MASK); \
                               T2 = BSIG0(A) + MAJ(A,B,C); \
                               D += T1; \
                               H = T1 + T2;
@@ -119,7 +107,7 @@
     {
         memset(&buffer[bufferLength], 0, 64-bufferLength);
         computeBlock(&h0, &h1, &h2, &h3, &h4, &h5, &h6, &h7, buffer);
-        memset(buffer, 0, bufferLength);
+        memset(buffer, 0, 48);
     }
     
     uint64_t lengthBit = totalBufferLength << 3;
@@ -191,102 +179,83 @@
                         uint32_t *h72,
                         uint8_t *buffer)
 {
-    uint32_t w[64];
+    uint32_t w[16];
     uint32_t *buffer2 = (uint32_t*)buffer;
-    w[0] = __rev(buffer2[0]);
-    w[1] = __rev(buffer2[1]);
-    w[2] = __rev(buffer2[2]);
-    w[3] = __rev(buffer2[3]);
-    w[4] = __rev(buffer2[4]);
-    w[5] = __rev(buffer2[5]);
-    w[6] = __rev(buffer2[6]);
-    w[7] = __rev(buffer2[7]);
-    w[8] = __rev(buffer2[8]);
-    w[9] = __rev(buffer2[9]);
-    w[10] = __rev(buffer2[10]);
-    w[11] = __rev(buffer2[11]);
-    w[12] = __rev(buffer2[12]);
-    w[13] = __rev(buffer2[13]);
-    w[14] = __rev(buffer2[14]);
-    w[15] = __rev(buffer2[15]);
-
-    for(int t = 16; t < 64; ++t)
-        w[t] = SSIG1(w[t-2]) + w[t-7] + SSIG0(w[t-15]) + w[t-16];
-    
     uint32_t a = *h02, b = *h12, c = *h22, d = *h32, e = *h42, f = *h52, g = *h62, h = *h72;
     uint32_t T1, T2;
-    
-    R(a,b,c,d,e,f,g,h,0)
-    R(h,a,b,c,d,e,f,g,1)
-    R(g,h,a,b,c,d,e,f,2)
-    R(f,g,h,a,b,c,d,e,3)
-    R(e,f,g,h,a,b,c,d,4)
-    R(d,e,f,g,h,a,b,c,5)
-    R(c,d,e,f,g,h,a,b,6)
-    R(b,c,d,e,f,g,h,a,7)
+
+
+    R(a,b,c,d,e,f,g,h,0,0x428a2f98)
+    R(h,a,b,c,d,e,f,g,1,0x71374491)
+    R(g,h,a,b,c,d,e,f,2,0xb5c0fbcf)
+    R(f,g,h,a,b,c,d,e,3,0xe9b5dba5)
+    R(e,f,g,h,a,b,c,d,4,0x3956c25b)
+    R(d,e,f,g,h,a,b,c,5,0x59f111f1)
+    R(c,d,e,f,g,h,a,b,6,0x923f82a4)
+    R(b,c,d,e,f,g,h,a,7,0xab1c5ed5)
 
-    R(a,b,c,d,e,f,g,h,8)
-    R(h,a,b,c,d,e,f,g,9)
-    R(g,h,a,b,c,d,e,f,10)
-    R(f,g,h,a,b,c,d,e,11)
-    R(e,f,g,h,a,b,c,d,12)
-    R(d,e,f,g,h,a,b,c,13)
-    R(c,d,e,f,g,h,a,b,14)
-    R(b,c,d,e,f,g,h,a,15)
-    
-    R(a,b,c,d,e,f,g,h,16)
-    R(h,a,b,c,d,e,f,g,17)
-    R(g,h,a,b,c,d,e,f,18)
-    R(f,g,h,a,b,c,d,e,19)
-    R(e,f,g,h,a,b,c,d,20)
-    R(d,e,f,g,h,a,b,c,21)
-    R(c,d,e,f,g,h,a,b,22)
-    R(b,c,d,e,f,g,h,a,23)
+    R(a,b,c,d,e,f,g,h,8,0xd807aa98)
+    R(h,a,b,c,d,e,f,g,9,0x12835b01)
+    R(g,h,a,b,c,d,e,f,10,0x243185be)
+    R(f,g,h,a,b,c,d,e,11,0x550c7dc3)
+    R(e,f,g,h,a,b,c,d,12,0x72be5d74)
+    R(d,e,f,g,h,a,b,c,13,0x80deb1fe)
+    R(c,d,e,f,g,h,a,b,14,0x9bdc06a7)
+    R(b,c,d,e,f,g,h,a,15,0xc19bf174)
+
+    R2(a,b,c,d,e,f,g,h,16,0xe49b69c1)
+    R2(h,a,b,c,d,e,f,g,17,0xefbe4786)
+    R2(g,h,a,b,c,d,e,f,18,0x0fc19dc6)
+    R2(f,g,h,a,b,c,d,e,19,0x240ca1cc)
+    R2(e,f,g,h,a,b,c,d,20,0x2de92c6f)
+    R2(d,e,f,g,h,a,b,c,21,0x4a7484aa)
+    R2(c,d,e,f,g,h,a,b,22,0x5cb0a9dc)
+    R2(b,c,d,e,f,g,h,a,23,0x76f988da)
     
-    R(a,b,c,d,e,f,g,h,24)
-    R(h,a,b,c,d,e,f,g,25)
-    R(g,h,a,b,c,d,e,f,26)
-    R(f,g,h,a,b,c,d,e,27)
-    R(e,f,g,h,a,b,c,d,28)
-    R(d,e,f,g,h,a,b,c,29)
-    R(c,d,e,f,g,h,a,b,30)
-    R(b,c,d,e,f,g,h,a,31) 
-    
-    R(a,b,c,d,e,f,g,h,32)
-    R(h,a,b,c,d,e,f,g,33)
-    R(g,h,a,b,c,d,e,f,34)
-    R(f,g,h,a,b,c,d,e,35)
-    R(e,f,g,h,a,b,c,d,36)
-    R(d,e,f,g,h,a,b,c,37)
-    R(c,d,e,f,g,h,a,b,38)
-    R(b,c,d,e,f,g,h,a,39)
+    R2(a,b,c,d,e,f,g,h,24,0x983e5152)
+    R2(h,a,b,c,d,e,f,g,25,0xa831c66d)
+    R2(g,h,a,b,c,d,e,f,26,0xb00327c8)
+    R2(f,g,h,a,b,c,d,e,27,0xbf597fc7)
+    R2(e,f,g,h,a,b,c,d,28,0xc6e00bf3)
+    R2(d,e,f,g,h,a,b,c,29,0xd5a79147)
+    R2(c,d,e,f,g,h,a,b,30,0x06ca6351)
+    R2(b,c,d,e,f,g,h,a,31,0x14292967) 
+
+    R2(a,b,c,d,e,f,g,h,32,0x27b70a85)
+    R2(h,a,b,c,d,e,f,g,33,0x2e1b2138)
+    R2(g,h,a,b,c,d,e,f,34,0x4d2c6dfc)
+    R2(f,g,h,a,b,c,d,e,35,0x53380d13)
+    R2(e,f,g,h,a,b,c,d,36,0x650a7354)
+    R2(d,e,f,g,h,a,b,c,37,0x766a0abb)
+    R2(c,d,e,f,g,h,a,b,38,0x81c2c92e)
+    R2(b,c,d,e,f,g,h,a,39,0x92722c85)
     
-    R(a,b,c,d,e,f,g,h,40)
-    R(h,a,b,c,d,e,f,g,41)
-    R(g,h,a,b,c,d,e,f,42)
-    R(f,g,h,a,b,c,d,e,43)
-    R(e,f,g,h,a,b,c,d,44)
-    R(d,e,f,g,h,a,b,c,45)
-    R(c,d,e,f,g,h,a,b,46)
-    R(b,c,d,e,f,g,h,a,47)
-
-    R(a,b,c,d,e,f,g,h,48)
-    R(h,a,b,c,d,e,f,g,49)
-    R(g,h,a,b,c,d,e,f,50)
-    R(f,g,h,a,b,c,d,e,51)
-    R(e,f,g,h,a,b,c,d,52)
-    R(d,e,f,g,h,a,b,c,53)
-    R(c,d,e,f,g,h,a,b,54)
-    R(b,c,d,e,f,g,h,a,55)
+    R2(a,b,c,d,e,f,g,h,40,0xa2bfe8a1)
+    R2(h,a,b,c,d,e,f,g,41,0xa81a664b)
+    R2(g,h,a,b,c,d,e,f,42,0xc24b8b70)
+    R2(f,g,h,a,b,c,d,e,43,0xc76c51a3)
+    R2(e,f,g,h,a,b,c,d,44,0xd192e819)
+    R2(d,e,f,g,h,a,b,c,45,0xd6990624)
+    R2(c,d,e,f,g,h,a,b,46,0xf40e3585)
+    R2(b,c,d,e,f,g,h,a,47,0x106aa070)
     
-    R(a,b,c,d,e,f,g,h,56)
-    R(h,a,b,c,d,e,f,g,57)
-    R(g,h,a,b,c,d,e,f,58)
-    R(f,g,h,a,b,c,d,e,59)
-    R(e,f,g,h,a,b,c,d,60)
-    R(d,e,f,g,h,a,b,c,61)
-    R(c,d,e,f,g,h,a,b,62)
-    R(b,c,d,e,f,g,h,a,63)
+    R2(a,b,c,d,e,f,g,h,48,0x19a4c116)
+    R2(h,a,b,c,d,e,f,g,49,0x1e376c08)
+    R2(g,h,a,b,c,d,e,f,50,0x2748774c)
+    R2(f,g,h,a,b,c,d,e,51,0x34b0bcb5)
+    R2(e,f,g,h,a,b,c,d,52,0x391c0cb3)
+    R2(d,e,f,g,h,a,b,c,53,0x4ed8aa4a)
+    R2(c,d,e,f,g,h,a,b,54,0x5b9cca4f)
+    R2(b,c,d,e,f,g,h,a,55,0x682e6ff3)
+    
+    R2(a,b,c,d,e,f,g,h,56,0x748f82ee)
+    R2(h,a,b,c,d,e,f,g,57,0x78a5636f)
+    R2(g,h,a,b,c,d,e,f,58,0x84c87814)
+    R2(f,g,h,a,b,c,d,e,59,0x8cc70208)
+    R2(e,f,g,h,a,b,c,d,60,0x90befffa)
+    R2(d,e,f,g,h,a,b,c,61,0xa4506ceb)
+    R2(c,d,e,f,g,h,a,b,62,0xbef9a3f7)
+    R2(b,c,d,e,f,g,h,a,63,0xc67178f2)
     
     
     *h02 += a;
@@ -326,7 +295,7 @@
     {
         memset(&buffer[length], 0, 64-length);
         computeBlock(&h0, &h1, &h2, &h3, &h4, &h5, &h6, &h7, buffer);
-        memset(buffer, 0, length);
+        memset(buffer, 0, 48);
     }
     
     uint32_t lengthBitLow = lengthBit;

--- a/SHA2_64.cpp	Wed Sep 11 17:22:40 2013 +0000
+++ b/SHA2_64.cpp	Thu Sep 12 10:18:57 2013 +0000
@@ -2,30 +2,6 @@
 #include <string.h>
 
 
-static const uint64_t K[] =
-{
-   0x428a2f98d728ae22, 0x7137449123ef65cd, 0xb5c0fbcfec4d3b2f, 0xe9b5dba58189dbbc,
-   0x3956c25bf348b538, 0x59f111f1b605d019, 0x923f82a4af194f9b, 0xab1c5ed5da6d8118,
-   0xd807aa98a3030242, 0x12835b0145706fbe, 0x243185be4ee4b28c, 0x550c7dc3d5ffb4e2,
-   0x72be5d74f27b896f, 0x80deb1fe3b1696b1, 0x9bdc06a725c71235, 0xc19bf174cf692694,
-   0xe49b69c19ef14ad2, 0xefbe4786384f25e3, 0x0fc19dc68b8cd5b5, 0x240ca1cc77ac9c65,
-   0x2de92c6f592b0275, 0x4a7484aa6ea6e483, 0x5cb0a9dcbd41fbd4, 0x76f988da831153b5,
-   0x983e5152ee66dfab, 0xa831c66d2db43210, 0xb00327c898fb213f, 0xbf597fc7beef0ee4,
-   0xc6e00bf33da88fc2, 0xd5a79147930aa725, 0x06ca6351e003826f, 0x142929670a0e6e70,
-   0x27b70a8546d22ffc, 0x2e1b21385c26c926, 0x4d2c6dfc5ac42aed, 0x53380d139d95b3df,
-   0x650a73548baf63de, 0x766a0abb3c77b2a8, 0x81c2c92e47edaee6, 0x92722c851482353b,
-   0xa2bfe8a14cf10364, 0xa81a664bbc423001, 0xc24b8b70d0f89791, 0xc76c51a30654be30,
-   0xd192e819d6ef5218, 0xd69906245565a910, 0xf40e35855771202a, 0x106aa07032bbd1b8,
-   0x19a4c116b8d2d0c8, 0x1e376c085141ab53, 0x2748774cdf8eeb99, 0x34b0bcb5e19b48a8,
-   0x391c0cb3c5c95a63, 0x4ed8aa4ae3418acb, 0x5b9cca4f7763e373, 0x682e6ff3d6b2b8a3,
-   0x748f82ee5defb2fc, 0x78a5636f43172f60, 0x84c87814a1f0ab72, 0x8cc702081a6439ec,
-   0x90befffa23631e28, 0xa4506cebde82bde9, 0xbef9a3f7b2c67915, 0xc67178f2e372532b,
-   0xca273eceea26619c, 0xd186b8c721c0c207, 0xeada7dd6cde0eb1e, 0xf57d4f7fee6ed178,
-   0x06f067aa72176fba, 0x0a637dc5a2c898a6, 0x113f9804bef90dae, 0x1b710b35131c471b,
-   0x28db77f523047d84, 0x32caab7b40c72493, 0x3c9ebe0a15c9bebc, 0x431d67c49c100d4c,
-   0x4cc5d4becb3e42b6, 0x597f299cfc657e2a, 0x5fcb6fab3ad6faec, 0x6c44198c4a475817
-};
-
 static const uint64_t H[] =
 {
     // SHA-384
@@ -39,14 +15,8 @@
 
 static uint64_t revWord(uint64_t w)
 {
-    return (w >> 56) 
-        | ((w & 0x00FF000000000000) >> 40) 
-        | ((w & 0x0000FF0000000000) >> 24) 
-        | ((w & 0x000000FF00000000) >> 8) 
-        | ((w & 0x00000000FF000000) << 8) 
-        | ((w & 0x0000000000FF0000) << 24) 
-        | ((w & 0x000000000000FF00) << 40) 
-        | ((w & 0x00000000000000FF) << 56);
+    return __rev((w & 0xFFFFFFFF00000000) >> 32) 
+         | ((uint64_t)(__rev(w & 0x00000000FFFFFFFF)) << 32);
 }
 
 #define ROTL(W,N) (((W) << (N)) | ((W) >> (64-(N))))
@@ -58,7 +28,11 @@
 #define SSIG0(X) (ROTR((X),1) ^ ROTR((X),8) ^ ((X) >> 7))
 #define SSIG1(X) (ROTR((X),19) ^ ROTR((X),61) ^ ((X) >> 6))
 
-    
+#define R(A,B,C,D,E,F,G,H,K,T)  T1 = H + BSIG1(E) + CH(E,F,G) + K + w[T]; \
+                              T2 = BSIG0(A) + MAJ(A,B,C); \
+                              D += T1; \
+                              H = T1 + T2;    
+                          
 
 SHA2_64::SHA2_64(SHA2_64_TYPE t):
 type(t),
@@ -198,28 +172,124 @@
                      uint8_t *buffer)
 {
     uint64_t w[80];
-    for(int t = 0; t < 16; ++t) 
-    {
-        memcpy(&w[t], &buffer[t*8], 8);
-        w[t] = revWord(w[t]);
-    }
+    uint64_t *buffer2 = (uint64_t*)buffer;
+
+    w[0] = revWord(buffer2[0]);
+    w[1] = revWord(buffer2[1]);
+    w[2] = revWord(buffer2[2]);
+    w[3] = revWord(buffer2[3]);
+    w[4] = revWord(buffer2[4]);
+    w[5] = revWord(buffer2[5]);
+    w[6] = revWord(buffer2[6]);
+    w[7] = revWord(buffer2[7]); 
+    w[8] = revWord(buffer2[8]);
+    w[9] = revWord(buffer2[9]);
+    w[10] = revWord(buffer2[10]);
+    w[11] = revWord(buffer2[11]);
+    w[12] = revWord(buffer2[12]);
+    w[13] = revWord(buffer2[13]);
+    w[14] = revWord(buffer2[14]);
+    w[15] = revWord(buffer2[15]);     
+    
     for(int t = 16; t < 80; ++t)
         w[t] = SSIG1(w[t-2]) + w[t-7] + SSIG0(w[t-15]) + w[t-16];
     
     uint64_t a = *h02, b = *h12, c = *h22, d = *h32, e = *h42, f = *h52, g = *h62, h = *h72;
-    for(int t = 0; t < 80; ++t)
-    {
-        uint64_t T1 = h + BSIG1(e) + CH(e,f,g) + K[t] + w[t];
-        uint64_t T2 = BSIG0(a) + MAJ(a,b,c);
-        h = g;
-        g = f;
-        f = e;
-        e = d + T1;
-        d = c;
-        c = b;
-        b = a;
-        a = T1 + T2;
-    }
+    uint64_t T1, T2;
+    
+
+    R(a,b,c,d,e,f,g,h,0x428a2f98d728ae22,0)
+    R(h,a,b,c,d,e,f,g,0x7137449123ef65cd,1)
+    R(g,h,a,b,c,d,e,f,0xb5c0fbcfec4d3b2f,2)
+    R(f,g,h,a,b,c,d,e,0xe9b5dba58189dbbc,3)
+    R(e,f,g,h,a,b,c,d,0x3956c25bf348b538,4)
+    R(d,e,f,g,h,a,b,c,0x59f111f1b605d019,5)
+    R(c,d,e,f,g,h,a,b,0x923f82a4af194f9b,6)
+    R(b,c,d,e,f,g,h,a,0xab1c5ed5da6d8118,7)
+
+    R(a,b,c,d,e,f,g,h,0xd807aa98a3030242,8)
+    R(h,a,b,c,d,e,f,g,0x12835b0145706fbe,9)
+    R(g,h,a,b,c,d,e,f,0x243185be4ee4b28c,10)
+    R(f,g,h,a,b,c,d,e,0x550c7dc3d5ffb4e2,11)
+    R(e,f,g,h,a,b,c,d,0x72be5d74f27b896f,12)
+    R(d,e,f,g,h,a,b,c,0x80deb1fe3b1696b1,13)
+    R(c,d,e,f,g,h,a,b,0x9bdc06a725c71235,14)
+    R(b,c,d,e,f,g,h,a,0xc19bf174cf692694,15)
+    
+    
+    R(a,b,c,d,e,f,g,h,0xe49b69c19ef14ad2,16)
+    R(h,a,b,c,d,e,f,g,0xefbe4786384f25e3,17)
+    R(g,h,a,b,c,d,e,f,0x0fc19dc68b8cd5b5,18)
+    R(f,g,h,a,b,c,d,e,0x240ca1cc77ac9c65,19)
+    R(e,f,g,h,a,b,c,d,0x2de92c6f592b0275,20)
+    R(d,e,f,g,h,a,b,c,0x4a7484aa6ea6e483,21)
+    R(c,d,e,f,g,h,a,b,0x5cb0a9dcbd41fbd4,22)
+    R(b,c,d,e,f,g,h,a,0x76f988da831153b5,23)
+    
+    R(a,b,c,d,e,f,g,h,0x983e5152ee66dfab,24)
+    R(h,a,b,c,d,e,f,g,0xa831c66d2db43210,25)
+    R(g,h,a,b,c,d,e,f,0xb00327c898fb213f,26)
+    R(f,g,h,a,b,c,d,e,0xbf597fc7beef0ee4,27)
+    R(e,f,g,h,a,b,c,d,0xc6e00bf33da88fc2,28)
+    R(d,e,f,g,h,a,b,c,0xd5a79147930aa725,29)
+    R(c,d,e,f,g,h,a,b,0x06ca6351e003826f,30)
+    R(b,c,d,e,f,g,h,a,0x142929670a0e6e70,31) 
+    
+    
+    R(a,b,c,d,e,f,g,h,0x27b70a8546d22ffc,32)
+    R(h,a,b,c,d,e,f,g,0x2e1b21385c26c926,33)
+    R(g,h,a,b,c,d,e,f,0x4d2c6dfc5ac42aed,34)
+    R(f,g,h,a,b,c,d,e,0x53380d139d95b3df,35)
+    R(e,f,g,h,a,b,c,d,0x650a73548baf63de,36)
+    R(d,e,f,g,h,a,b,c,0x766a0abb3c77b2a8,37)
+    R(c,d,e,f,g,h,a,b,0x81c2c92e47edaee6,38)
+    R(b,c,d,e,f,g,h,a,0x92722c851482353b,39)
+    
+    R(a,b,c,d,e,f,g,h,0xa2bfe8a14cf10364,40)
+    R(h,a,b,c,d,e,f,g,0xa81a664bbc423001,41)
+    R(g,h,a,b,c,d,e,f,0xc24b8b70d0f89791,42)
+    R(f,g,h,a,b,c,d,e,0xc76c51a30654be30,43)
+    R(e,f,g,h,a,b,c,d,0xd192e819d6ef5218,44)
+    R(d,e,f,g,h,a,b,c,0xd69906245565a910,45)
+    R(c,d,e,f,g,h,a,b,0xf40e35855771202a,46)
+    R(b,c,d,e,f,g,h,a,0x106aa07032bbd1b8,47)
+
+    R(a,b,c,d,e,f,g,h,0x19a4c116b8d2d0c8,48)
+    R(h,a,b,c,d,e,f,g,0x1e376c085141ab53,49)
+    R(g,h,a,b,c,d,e,f,0x2748774cdf8eeb99,50)
+    R(f,g,h,a,b,c,d,e,0x34b0bcb5e19b48a8,51)
+    R(e,f,g,h,a,b,c,d,0x391c0cb3c5c95a63,52)
+    R(d,e,f,g,h,a,b,c,0x4ed8aa4ae3418acb,53)
+    R(c,d,e,f,g,h,a,b,0x5b9cca4f7763e373,54)
+    R(b,c,d,e,f,g,h,a,0x682e6ff3d6b2b8a3,55)
+    
+    R(a,b,c,d,e,f,g,h,0x748f82ee5defb2fc,56)
+    R(h,a,b,c,d,e,f,g,0x78a5636f43172f60,57)
+    R(g,h,a,b,c,d,e,f,0x84c87814a1f0ab72,58)
+    R(f,g,h,a,b,c,d,e,0x8cc702081a6439ec,59)
+    R(e,f,g,h,a,b,c,d,0x90befffa23631e28,60)
+    R(d,e,f,g,h,a,b,c,0xa4506cebde82bde9,61)
+    R(c,d,e,f,g,h,a,b,0xbef9a3f7b2c67915,62)
+    R(b,c,d,e,f,g,h,a,0xc67178f2e372532b,63)
+
+    R(a,b,c,d,e,f,g,h,0xca273eceea26619c,64)
+    R(h,a,b,c,d,e,f,g,0xd186b8c721c0c207,65)
+    R(g,h,a,b,c,d,e,f,0xeada7dd6cde0eb1e,66)
+    R(f,g,h,a,b,c,d,e,0xf57d4f7fee6ed178,67)
+    R(e,f,g,h,a,b,c,d,0x06f067aa72176fba,68)
+    R(d,e,f,g,h,a,b,c,0x0a637dc5a2c898a6,69)
+    R(c,d,e,f,g,h,a,b,0x113f9804bef90dae,70)
+    R(b,c,d,e,f,g,h,a,0x1b710b35131c471b,71)
+
+    R(a,b,c,d,e,f,g,h,0x28db77f523047d84,72)
+    R(h,a,b,c,d,e,f,g,0x32caab7b40c72493,73)
+    R(g,h,a,b,c,d,e,f,0x3c9ebe0a15c9bebc,74)
+    R(f,g,h,a,b,c,d,e,0x431d67c49c100d4c,75)
+    R(e,f,g,h,a,b,c,d,0x4cc5d4becb3e42b6,76)
+    R(d,e,f,g,h,a,b,c,0x597f299cfc657e2a,77)
+    R(c,d,e,f,g,h,a,b,0x5fcb6fab3ad6faec,78)
+    R(b,c,d,e,f,g,h,a,0x6c44198c4a475817,79)
+            
     *h02 += a;
     *h12 += b;
     *h22 += c;